YouTube2Text

Why the Next 10 Years May Add 50 to Your Lifespan | Dr. Derya Unutmaz — Transcript

by FoundMyFitness · 26,886 words · 3,719 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This is probably the most critical time
  2. 0:02in human history. So try not to die
  3. 0:04[music] for the next 10 years.
  4. 0:06>> Can you explain and unpack why you think
  5. 0:09that?
  6. 0:10>> The reason is that the technology
  7. 0:12because of AI is expanding
  8. 0:14exponentially. Cancer is going to be
  9. 0:16100% curable probably less than a
  10. 0:18decade. We'll get to a point where we'll
  11. 0:20have hundreds of new drugs coming out
  12. 0:22every month maybe. And then we'll get to
  13. 0:25a point uh probably 15 maximum 20 years
  14. 0:29where we will be able to completely
  15. 0:31reverse the aging process. So if you're
  16. 0:3380 years old, 90 years old, you will get
  17. 0:35back to [music] uh age 30, 40, whatever.
  18. 0:38>> Why do you have such an optimistic view?
  19. 0:40>> AI is an incredible enabler. It gives
  20. 0:43you superpowers. [music] The key risk is
  21. 0:45is humans. Humans misusing AI. There's
  22. 0:49only [music] one existential threat to
  23. 0:51humanity and that's humanity.
  24. 0:55Hey everyone, today's episode explores
  25. 0:57an extraordinarily exciting convergence,
  26. 1:00the accelerating pace of artificial
  27. 1:02intelligence and a growing optimism
  28. 1:04about the future of science and
  29. 1:05medicine. In this episode, I discuss
  30. 1:08with Dr. Duria Enautmas how AI could
  31. 1:10dramatically improve our ability to
  32. 1:12detect, prevent, and treat human disease
  33. 1:14and ultimately extend human life
  34. 1:16expectancy. Before we begin, I just want
  35. 1:19to mention one quick thing. Only about
  36. 1:2130% of the people who watch this podcast
  37. 1:24are subscribed to the YouTube channel.
  38. 1:26Taking a moment to subscribe and enable
  39. 1:28notifications is one of the simplest
  40. 1:30ways to support the show and help us
  41. 1:32bring these conversations to a wider
  42. 1:34audience. We greatly appreciate it.
  43. 1:37Thank you so much and I really hope you
  44. 1:39enjoy this episode with Dr. Duria
  45. 1:40Enutmas.
  46. 1:42I'm so excited to be sitting here with
  47. 1:44Dr. Dura Unutmas who is one of the
  48. 1:48handful of scientists that has had
  49. 1:51access to collaborate with open AI one
  50. 1:54of the you know world's leader in in
  51. 1:58artificial intelligence. He's also an
  52. 2:00aging researcher. He's an immunologist
  53. 2:03really just a matchmade in heaven to sit
  54. 2:06down and talk about the role of AI in in
  55. 2:10aging research and in medicine. So I'm
  56. 2:12super excited to have you here today.
  57. 2:14I'm very excited to be here. Thank you.
  58. 2:16>> As we both know, aging is a very, very
  59. 2:20complex
  60. 2:21process. Many factors involved. It's
  61. 2:24heterogeneous. It's so complex. And it
  62. 2:28just seems like so almost impossible to
  63. 2:31solve. And yet, I've heard you say
  64. 2:34something that's very interesting. I've
  65. 2:37heard you say if you could try not to
  66. 2:39die within the next 10 to 15 years, you
  67. 2:43might want to try to do that because you
  68. 2:45could live an extra 50 years.
  69. 2:48>> Yeah.
  70. 2:48>> Can you explain and unpack why you think
  71. 2:52that? What makes you believe that?
  72. 2:54>> Thank you. So, first of all, I'm very
  73. 2:56excited to be here. I'm a big follower
  74. 2:58of your podcast. I think it's maybe the
  75. 3:01best uh aging or longevity podcast. So,
  76. 3:04uh this is this is a great pleasure. Um
  77. 3:07yeah so I've I've said that um quite a
  78. 3:10few times uh in the last year or two
  79. 3:13actually um and it may not even take 10
  80. 3:1515 years might be uh even closer. Uh the
  81. 3:19reason is that uh the technology
  82. 3:23especially because of AI uh is expanding
  83. 3:26exponentially. So our minds think in a
  84. 3:29linear term. So we think that the next
  85. 3:3110 years is going to be as much advanced
  86. 3:33as the last 10 years or the last 15
  87. 3:36years. But that's not what's going to
  88. 3:37happen. In the next 10 years, you can
  89. 3:39think of it as more advanced than the
  90. 3:41last century. So imagine that you were
  91. 3:43living in early 1900s. Uh and somebody
  92. 3:47told you that you know we're going to uh
  93. 3:48have vaccines and you will never get
  94. 3:51small pox or uh you won't die of
  95. 3:54tuberculosis. You know people would
  96. 3:57laugh at you. So that's not that's not
  97. 3:59possible. Um so so that's this the speed
  98. 4:02that we're talking about. But there's
  99. 4:04something uh even more important because
  100. 4:06of this acceleration. [clears throat]
  101. 4:08The u the advances of treating diseases
  102. 4:11is also going to accelerate
  103. 4:13dramatically. So uh we will get to a
  104. 4:16point what's called the longevity escape
  105. 4:18velocity. This was coined by Aubry de
  106. 4:20Gray who's as you know is a great uh
  107. 4:23aging um researcher. Uh so the point is
  108. 4:26that we will come to a point in the next
  109. 4:29I would say probably eight to 10 years
  110. 4:32where every year you live is going to
  111. 4:35add more than a year to your life. So
  112. 4:38let's just say um you know 10 years ago
  113. 4:4210 years later you get a a cancer uh
  114. 4:45that's normally is not curable and you
  115. 4:48only have one or two years to live. uh
  116. 4:51um but that during that one year there
  117. 4:54is going to be a new treatment that will
  118. 4:56cure that cancer. So automatically it's
  119. 4:59going to add several years or maybe 10
  120. 5:0115 years to your life or uh we're
  121. 5:04already starting to see that with the
  122. 5:05GLP1 uh drugs uh receptor agonist which
  123. 5:10uh which are adding about 5 to 10 years
  124. 5:12to lifespan of people who are obese or
  125. 5:16who have chronic uh conditions uh will
  126. 5:18have sort of the muscle generators uh uh
  127. 5:21which I think will have tremendous
  128. 5:23impact on the aging population because
  129. 5:25as you know that's a huge problem. So
  130. 5:27all of these things will add up and and
  131. 5:29and the technology and AI is going to
  132. 5:32keep accelerating. So 10 years later uh
  133. 5:36what will happen in a year will be like
  134. 5:39what happens in 20 years of advance and
  135. 5:42then we'll get to a point
  136. 5:4515 maximum 20 years where we will be
  137. 5:48able to completely reverse the aging pro
  138. 5:51process. So if you're 80 years old, 90
  139. 5:54years old, you will get back to uh age
  140. 5:5630, 40, whatever. So that's going to add
  141. 5:58up uh 50 years or 100 years to to your
  142. 6:01lifespan. Um and then you can keep doing
  143. 6:03that and extend it almost uh
  144. 6:06indefinitely. So I think this is
  145. 6:08probably the most critical time in human
  146. 6:10history. So try not to die for the next
  147. 6:1210 years.
  148. 6:13>> And we're going to talk about all these
  149. 6:15things. I want to talk about curing
  150. 6:16disease. I want to talk about reversing
  151. 6:18aging, age reversal. Um, all of that is
  152. 6:22on on on my agenda to talk about with
  153. 6:24you today. But you mentioned something.
  154. 6:26You mentioned that right now the, you
  155. 6:30know, artificial intelligence as a
  156. 6:32general term, you know, is is
  157. 6:34accelerating at was an exponential rate.
  158. 6:37I've heard you talk about this Moore's
  159. 6:38law and how the, you know, the software
  160. 6:43itself is accelerating right at this
  161. 6:46exponential rate. Um maybe you could
  162. 6:49explain a little bit about like what
  163. 6:52what does that mean and then how how do
  164. 6:55you think that'll translate into biology
  165. 6:57because you know humans we're not
  166. 7:00software and there are things that at
  167. 7:03least in my opinion you know you have to
  168. 7:05still test safety right I mean so like
  169. 7:08if you're you know accelerating the
  170. 7:11computational speed and therefore you
  171. 7:13can test a lot of things that are what
  172. 7:15are what's called incilico for people
  173. 7:17listening we're talking about testing
  174. 7:19things like just modeling them and maybe
  175. 7:21you can explain this a little bit
  176. 7:22better. Um, but then at a certain point
  177. 7:26you still have to test about, you know,
  178. 7:29safety and you you definitely that that
  179. 7:31there's there are things that I think
  180. 7:32need to still be done in human trials.
  181. 7:34So I'd love to hear how you think that's
  182. 7:36going to happen.
  183. 7:37>> I think that's the the most critical
  184. 7:38question because people always bring
  185. 7:40that up. Okay, you know, if you generate
  186. 7:42drugs uh within hours, you still have to
  187. 7:45test them on humans for five years,
  188. 7:47maybe some sometimes longer. How how are
  189. 7:49you going to deal with that? But let me
  190. 7:50let me first uh start with how AI is
  191. 7:54accelerating biology now. So we we can
  192. 7:57think of it in in terms of phases and
  193. 7:59and because now and 5 years later is
  194. 8:02going to be very very different. Um so
  195. 8:04right now especially in the last year or
  196. 8:07two since uh you know LLM came out um
  197. 8:12you know their intelligence have been
  198. 8:14accelerating. Initially it was uh fairly
  199. 8:17um smaller u productivity gains. For
  200. 8:21example you know when GPT4 was was out I
  201. 8:25I would ask it to sort of scan the
  202. 8:27literature and tell me what's the the
  203. 8:29latest on this topic or that topic. Um
  204. 8:32and that saved me you know hours
  205. 8:34sometimes days. Uh but then as the
  206. 8:37models advanced especially the after01
  207. 8:39model the reasoning models uh started to
  208. 8:41come out and and now we have the GPT5 uh
  209. 8:44pro model 5.5 pro model. Uh what
  210. 8:47happened was that now they were able to
  211. 8:49think and plan. So uh you could start to
  212. 8:53ask very sophisticated questions. For
  213. 8:56example, here is a huge biological data
  214. 8:59set, a million data points or 10 million
  215. 9:02data points. Go over this. Not only just
  216. 9:05analyze it and group them, but uh what
  217. 9:09what is the insight from that data? Uh
  218. 9:12human mind is not able to do that. And
  219. 9:14in fact, we had such data sets which
  220. 9:16took us months to analyze like you know
  221. 9:19a PhD student work on it using deep
  222. 9:21learning. we still couldn't really truly
  223. 9:25understand what that data meant. We we
  224. 9:27know these genes are up, this
  225. 9:29metabolites are changing, this is
  226. 9:30happening. How do you bring all that
  227. 9:32together? Um, and so now AI models are
  228. 9:36able to do that. So you I I've tested
  229. 9:38for example uh latest GPT5 uh pro model.
  230. 9:42You can upload uh millions of data sets
  231. 9:45that we accumulate over years maybe uh
  232. 9:48and then in matter of minutes you get
  233. 9:51not only the complete analysis recently
  234. 9:54I had a 40page report from GPT5 uh pro
  235. 9:59um which was an analysis of this what's
  236. 10:01called the RNA sequencing lots of
  237. 10:03millions of data points but it also
  238. 10:05provided incredible insight like what is
  239. 10:08the what does this data mean what should
  240. 10:10be the next questions to ask so that
  241. 10:12automatically contracts months sometimes
  242. 10:15years of analytic work into matter of
  243. 10:18minutes or hours. Um so so there that
  244. 10:22that is already accelerating of course
  245. 10:23in the drug uh design uh parts uh I
  246. 10:27think every pharmaceutical company is
  247. 10:30going to eventually use AI generated AI
  248. 10:34generation for developing new drugs.
  249. 10:37things that took years of screening of
  250. 10:39small molecules now take you know hours
  251. 10:42or days. So, so tremendous acceleration
  252. 10:45there and then uh I think again more
  253. 10:48recently because the models have
  254. 10:50advanced so much that you can also ask
  255. 10:52things like okay so uh this is great um
  256. 10:56this is the hypothesis in fact AI can
  257. 10:59even generate hypods for you but what
  258. 11:01sort of experiment I should do to
  259. 11:03address that uh people have to realize
  260. 11:06that uh what we do in in biology is
  261. 11:10experiments but we don't really know
  262. 11:12what's the best experiment to do. I mean
  263. 11:14that's kind of my job but I have some
  264. 11:17intuition we should do this to address
  265. 11:19that question but is that the ideal
  266. 11:21experiment is does that have all the
  267. 11:23controls everything so AI models are now
  268. 11:26able to tell you sort of simulating if
  269. 11:29out of this 100 potential experiments
  270. 11:31you can do this two are the best ones
  271. 11:34because this is going to give you the
  272. 11:36best uh output and I' I've been testing
  273. 11:39that so so that is another acceleration
  274. 11:41now you don't have to try 100 things for
  275. 11:45a year, you can just try two things for
  276. 11:47few weeks and and get get the output. So
  277. 11:50that's what's possible now already
  278. 11:53tremendously accelerating the R&D part.
  279. 11:55But then uh the second part which I
  280. 11:58think is more important part is how do
  281. 12:00we uh apply that to clinical trials and
  282. 12:03regulations. Um, so it still takes years
  283. 12:07to try everything on humans and I think
  284. 12:09the solution to that will be what I call
  285. 12:12the digital twin and this this term is
  286. 12:14around for for several years. So the
  287. 12:17idea is that if we have lots of lots of
  288. 12:20biological data and when I say lots it's
  289. 12:22it's a lot pupy bytes of of data. If if
  290. 12:27AI comes to a point where we're going to
  291. 12:29need much more compute than we have
  292. 12:31today today is to able to compute all
  293. 12:34that and really kind of simulate a whole
  294. 12:36biological organism, a whole human
  295. 12:39being, but not just your um phenotype
  296. 12:42but but also your metabolism, your
  297. 12:44immune system, your gut microbiome, uh
  298. 12:47your genetics and and all kinds of data
  299. 12:50sets are put together. And so it it it
  300. 12:53knows your biology in a temporal way in
  301. 12:56in in a in a totally functional way.
  302. 12:58Then you can ask the question okay so if
  303. 13:01I give this drug to this person what
  304. 13:04kind of effect it will have if they have
  305. 13:07this disruption is it going to have a
  306. 13:09side effect or is it going to be
  307. 13:10effective. So literally we can cut down
  308. 13:13clinical trial time from years to to a
  309. 13:16matter of months or or even weeks. So
  310. 13:18you can actually do the trials in a very
  311. 13:21small subset of patients because you can
  312. 13:24choose the patients. You can say okay AI
  313. 13:27told me that these these these people
  314. 13:29this drug is going to be effective 100%
  315. 13:31to them. And so so you just test it on
  316. 13:33those people and in fact that will go
  317. 13:36into the personalization. There's going
  318. 13:38to be thousands of drugs for for
  319. 13:39different people. So that that will
  320. 13:41cause tremendous acceleration. We're not
  321. 13:44there yet, but I'm I'm betting on that
  322. 13:46that within the five five to 10 years we
  323. 13:49will get there. So, uh the iteration
  324. 13:52process the on humans is going to be all
  325. 13:55digital as well and then maybe the the
  326. 13:58manufacturing will be a little bit um uh
  327. 14:01still will take time but but we can even
  328. 14:04improve that part too. So uh we at some
  329. 14:07point we will come uh to a point where
  330. 14:11treatment on demand. So you go to an AI
  331. 14:14model analyzes your genome, your biology
  332. 14:19uh orders the this small molecule or the
  333. 14:21drug or treatment just for you to the
  334. 14:24manufacturing facility and next week you
  335. 14:27get your drug and and you get treated.
  336. 14:29That's the world I'm imagining. So I
  337. 14:32want to get back to this concept of
  338. 14:33digital twin um again when we talk about
  339. 14:35personalized medicine but if I
  340. 14:37understand correctly so you know if we
  341. 14:39are if we have this digital twin which
  342. 14:41is all the genetic data metabolomic
  343. 14:43proteomic biomarker just everything
  344. 14:45right all this data um and more that
  345. 14:48we're not talking about um
  346. 14:51and and now we have AI which can then
  347. 14:54you know do all these scenarios and
  348. 14:56figure out like how this drug is going
  349. 14:58to affect or how this treatment is going
  350. 14:59to affect this And you're saying that
  351. 15:02the clinical trial that may have taken,
  352. 15:04you know, a few years can be condensed
  353. 15:07down and perhaps we can look at after
  354. 15:10doing the incilico experiments, you can
  355. 15:11look at some biomarkers and know like is
  356. 15:13this going to affect their fertility
  357. 15:14like you don't want to give some some
  358. 15:17someone a treatment that's going to make
  359. 15:18them infertile or you know so you think
  360. 15:21that's going to be uh AI is going to be
  361. 15:23able to identify how to know if it's
  362. 15:27going to affect like fertility or
  363. 15:29cognition or life expectancy or you know
  364. 15:32just just from
  365. 15:34>> the whole composition of the person and
  366. 15:36doing I don't know all these tests.
  367. 15:38>> Yeah.
  368. 15:40So uh I mean the path there uh requires
  369. 15:44uh several steps of validation uh and
  370. 15:47that I think we will get to a point
  371. 15:50where when we have super intelligence
  372. 15:52that we'll be able to trust super
  373. 15:54intelligence
  374. 15:55you know almost 100% that we don't need
  375. 15:58to validate it even with biomarkers or
  376. 16:01or whatnot but to get to that point it's
  377. 16:04sort of like the self-driving cars right
  378. 16:06so um to get to a self-driving being
  379. 16:09leveled. I mean, it has to be 99.999%
  380. 16:13safety. Um, you you have to sort of
  381. 16:16validate it. Um, uh, you know what
  382. 16:19happens if somebody's crossing the
  383. 16:21street, right? So, so that scenario has
  384. 16:24to happen and then you you record it and
  385. 16:26sometimes uh you won't do the right
  386. 16:29thing. Maybe, you know, it won't stop.
  387. 16:31That's why we still have to like look at
  388. 16:33this, you know, be ready to to take
  389. 16:35control. But if it does stop and it
  390. 16:38stops uh and and saves lives again and
  391. 16:40again and again and right now you know
  392. 16:42self-driving cars are probably about 10
  393. 16:45times safer. They will be maybe hundred
  394. 16:46times safer. So you get to a point that
  395. 16:49you trust the AI rather than the the
  396. 16:52driver, right? So you say okay so I I
  397. 16:55trust I want the AI to decide for me uh
  398. 16:59to to drive. So I think we'll get to
  399. 17:01that point for biology. it will take a
  400. 17:02little bit longer uh because of the
  401. 17:05extreme complexity. Um and then we'll
  402. 17:07have to have uh very um clever
  403. 17:11benchmarking and validation
  404. 17:14uh ways there. the biomarker is going to
  405. 17:17be really important because again, you
  406. 17:20know, if you're developing an aging drug
  407. 17:22uh that you claim will let people to
  408. 17:25live to 150, well, you can't wait uh you
  409. 17:28know, even even if somebody 100 years
  410. 17:30old takes it, you still have to wait 150
  411. 17:33years,
  412. 17:3450 more years to to validate that. So
  413. 17:36that that's not going to work out. So we
  414. 17:38have to be able to predict that. But but
  415. 17:41actually probably aging is is the
  416. 17:43easiest in some ways uh to predict
  417. 17:47because uh we have so many biomarkers or
  418. 17:51functional outputs we can measure. We
  419. 17:54know how they are in an old person and
  420. 17:56in a young person. So if your vi
  421. 17:59suddenly gets uh you know like a
  422. 18:0220-year-old wow that's amazing. If your
  423. 18:04muscles are as good as a 30 year old, uh
  424. 18:08if your skin looks uh like a 20- year
  425. 18:10old, that's what my mom is waiting for.
  426. 18:13Uh you know, that that's that's proof.
  427. 18:15And you'll you'll immediately see that.
  428. 18:17I mean, immediately weeks or or or or or
  429. 18:20whatnot. So, I think um
  430. 18:23again, it will take time. That's the
  431. 18:25part that's going to take time, the sort
  432. 18:27of trusting AI to um to tell you yes, if
  433. 18:33you take this drug, you will you will be
  434. 18:36treated or you will reverse aging. Um uh
  435. 18:39we we we still have about a decade.
  436. 18:40That's why I'm saying like you know
  437. 18:43other otherwise it would it would take
  438. 18:45it would happen even earlier. You
  439. 18:47mentioned super intelligence, artificial
  440. 18:49super intelligence, ASI. Maybe you could
  441. 18:52talk a little bit about just for people
  442. 18:54to have an understanding right now the
  443. 18:56difference between artificial
  444. 18:57intelligence, artificial generalized
  445. 18:59intelligence, AGI, then the super
  446. 19:01intelligence because you said
  447. 19:03>> once we get to the super intelligence,
  448. 19:05we're going to trust it, right? So I I
  449. 19:07mean I don't know do we know what those
  450. 19:08differences are or can you explain a
  451. 19:10little bit?
  452. 19:10>> Yeah, of course. you know this changes
  453. 19:13on a daily basis what the definition are
  454. 19:15depending on who's uh whose definition
  455. 19:18uh but the you know I've been thinking
  456. 19:20about AGI ASI for decades I mean it's
  457. 19:23not something that I started to think
  458. 19:25about it recently um and so the way u I
  459. 19:29originally uh defined AGI it's it's
  460. 19:32artificial general intelligence so what
  461. 19:35that means is that first of all it's
  462. 19:37artificial right so it's not human
  463. 19:40intelligence artificial intelligence and
  464. 19:43then it's general. What that means is
  465. 19:45that um if AI learns uh one set of uh
  466. 19:49rules or one set of knowledge that it
  467. 19:52can generalize that to something else
  468. 19:54and that's how our brains are
  469. 19:56intelligent um because uh you you can be
  470. 20:00an amazing chess player. In fact, you
  471. 20:02know AI beat the chess champion Kasper
  472. 20:06in 1997 I think like decades ago but
  473. 20:09that was not general intelligence. It
  474. 20:12was super uh good or Alpha Go beat you
  475. 20:15know the the the world champion in Go
  476. 20:18which is much more difficult uh uh game
  477. 20:21uh to be general uh Alph Go you know
  478. 20:26learning how to play Go or chess should
  479. 20:29be able to I don't know solve a problem
  480. 20:32in aging right so it should be able to
  481. 20:35transfer that information I think the uh
  482. 20:38the amazing thing about LLMs what we
  483. 20:40call large language models is that they
  484. 20:42acquire this ability uh which honestly I
  485. 20:46didn't think uh this would happen so so
  486. 20:48easily. I was expecting AGI to to happen
  487. 20:51uh maybe a decade ago. So in my opinion
  488. 20:53we have already achieved uh what I call
  489. 20:56level one AGI artificial unit because if
  490. 21:00I ask GPT5 uh pro model you know
  491. 21:04something that hasn't it hasn't trained
  492. 21:06on like an experiment that I have done
  493. 21:09or if I say okay think of the experiment
  494. 21:12as a video game design another
  495. 21:14experiment for me like like you are
  496. 21:16playing a video game so that's
  497. 21:18transferring completely different area
  498. 21:21to a biological system and is able to do
  499. 21:23that in an amazing way. But we have we
  500. 21:26we still need to go through several
  501. 21:29levels. I I think the next level is
  502. 21:31going to be memory. So they don't have
  503. 21:33persistent memory right now. They have
  504. 21:36some memory. They know about you. Uh
  505. 21:39they know about what they've learned in
  506. 21:40the internet, but they need to be able
  507. 21:43to uh manage the context, you know,
  508. 21:45because there's a continuum. Life is a
  509. 21:47continuum. Um and then the the other one
  510. 21:51is going to be the self-learning right
  511. 21:53so maybe that's level three I it doesn't
  512. 21:56matter uh and that's coming soon you
  513. 21:58know AI companies are saying like we
  514. 22:00think that the real time learning uh is
  515. 22:02is is coming maybe by by next year um
  516. 22:05and then uh the third level uh uh what I
  517. 22:08call the physical intelligence so people
  518. 22:10again confuse this greatly because the
  519. 22:14true human level intelligence is
  520. 22:16physical intelligence it's not cognitive
  521. 22:18intelligence. So for uh millions of
  522. 22:21years, we evolved to survive in a
  523. 22:23physical world. We we didn't have
  524. 22:25language up to I don't know 10,000 years
  525. 22:27ago like we didn't know how to write. Um
  526. 22:31this cognitive part uh has developed in
  527. 22:34the last you know maybe 10 20,000 years.
  528. 22:37Uh before that in fact animals have very
  529. 22:40good physical intelligence. We're we're
  530. 22:42imprinted and born with that
  531. 22:43intelligence. So uh an admiral or a
  532. 22:46child knows already have a world model.
  533. 22:49They know that you know if I drop this
  534. 22:51it's going to fall and and doesn't have
  535. 22:53to test it a million times. And that's
  536. 22:56of course what we need for robots for
  537. 22:58embodiment. And and you can see that you
  538. 23:01know that's taken a long time. You know
  539. 23:03it's more difficult to train a robot to
  540. 23:07behave like a child than have GPT5 solve
  541. 23:10the most difficult math problem. So and
  542. 23:14we'll get there. I think people are
  543. 23:15working on these moral models and
  544. 23:17physical intelligence whether we need
  545. 23:18another algorithm or not. So that will
  546. 23:21be the the final level of the AGI level.
  547. 23:24Um once we have all those levels then
  548. 23:27and once the uh AI is able to self-learn
  549. 23:31um then that's the super intelligence
  550. 23:34because at that point it can train
  551. 23:36itself you know maybe thousands maybe
  552. 23:39millions fold faster than we are able to
  553. 23:42do. Um and and there is a there's a
  554. 23:44limit to human intelligence right so
  555. 23:46even the smartest person in the world
  556. 23:49can only do so much uh and super
  557. 23:51intelligence what I would define is that
  558. 23:54you will have the intelligence of
  559. 23:57combined totality of humanity at some
  560. 24:00point like if you if I bring uh a
  561. 24:03million top scientists in the world of
  562. 24:06course they can solve you know like a
  563. 24:08Manhattan project they brought all these
  564. 24:10brilliant minds it wasn't one person's
  565. 24:12uh they were able to solve very hard
  566. 24:14problems. Super intelligence will get to
  567. 24:16that level. We'll be able to do what
  568. 24:18thousands of scientists can do uh in a
  569. 24:21year will be able to do uh in a day. So
  570. 24:24um I I would probably trust that.
  571. 24:26>> Wow, [laughter]
  572. 24:28that's pretty exciting. I mean and it it
  573. 24:31also kind of brings in this this concept
  574. 24:34of when you talk to people about AI and
  575. 24:38not everyone has the understanding of it
  576. 24:41as you um for sure you you hear that
  577. 24:44there's there's a pessimistic versus
  578. 24:46optimistic view right and oftentimes if
  579. 24:49I talk to people I hear a lot of
  580. 24:50pessimism I hear perhaps they don't
  581. 24:53understand their fear of the unknown of
  582. 24:56what AI is capable of I mean you're just
  583. 24:58the super intelligence that you're
  584. 24:59talking about I feel if you explain that
  585. 25:01to some people it would scare them even
  586. 25:03more. You know, perhaps they are worried
  587. 25:05about the cultural ramifications, ep e
  588. 25:08economic ramifications, but also just
  589. 25:11this Terminator situation where, okay,
  590. 25:13well, they're super smart. They're going
  591. 25:14to want to then take over the world and
  592. 25:16they don't need us anymore, right?
  593. 25:18>> But you have such an optimistic view. I
  594. 25:20mean, we're talking about solving aging,
  595. 25:22living to be 150 or more. Uh why do you
  596. 25:26have such an optimistic view? Are you
  597. 25:28worried at all about the other
  598. 25:30pessimistic sort of viewpoints or
  599. 25:33>> absolutely not and I I'll tell you why
  600. 25:35I'm I'm so super optimistic about it. Um
  601. 25:38when people make those statements like
  602. 25:41uh AI is an existential threat for us
  603. 25:43and you know it's going to destroy
  604. 25:45humanity. Um I make the counterpoint
  605. 25:48there's only one existential threat to
  606. 25:51humanity and that's humanity. So if you
  607. 25:53look at history um human beings killed
  608. 25:58more humans than everything put together
  609. 26:01caused more suffering than anything that
  610. 26:05humans have have been exposed to. You
  611. 26:08know even animals I don't think they
  612. 26:10they maybe infectious diseases at some
  613. 26:12point might have caused um a lot of
  614. 26:15suffering but but but the real danger is
  615. 26:18is the human intelligence.
  616. 26:21So uh I do let's do a thought
  617. 26:24experiment. Let's imagine that uh we
  618. 26:27live in a parallel universe and in that
  619. 26:29universe the world have decided that
  620. 26:33anyone above the IQ of let's say 100 is
  621. 26:36a danger to the society because if you
  622. 26:38get very intelligent you can come up
  623. 26:40with ideas that could be very dangerous
  624. 26:43right and that's true actually that's
  625. 26:45how it happened. Um, and then if you if
  626. 26:48you have an IQ of 105, you get
  627. 26:50imprisoned immediately. So you you are
  628. 26:52not allowed to to participate in society
  629. 26:55or you get killed or whatever that the
  630. 26:57the society has decided intelligence is
  631. 27:00dangerous so we're going to stop it. Uh
  632. 27:02what kind of a world we would live in?
  633. 27:05We would not have anything that we have
  634. 27:07right now. we would live in probably
  635. 27:08just as farmers you know basic physical
  636. 27:11intelligence we have uh and and try to
  637. 27:14survive you know uh in a world where the
  638. 27:17average lifespan was 30 years old or
  639. 27:19something like that. So that's that's
  640. 27:21how we should view uh AI and and the
  641. 27:25other point is that [clears throat]
  642. 27:26about this uh sort of AI is going to
  643. 27:29take over and is going to replace us. Um
  644. 27:32uh I see it exactly the opposite because
  645. 27:36AI is is is an incredible enabler. It
  646. 27:40gives you superpowers. Even now I feel
  647. 27:43like I have superpowers. Uh you know
  648. 27:46I've never been this busy in my life.
  649. 27:48You know I I actually sleep less which
  650. 27:50is not a good thing by the way. I don't
  651. 27:52recommend it but be because I can do so
  652. 27:54much. It's so empowering. You know my
  653. 27:57mom was was was 86 years old. you know,
  654. 27:59she told me that uh chat GPT changed her
  655. 28:02life. She she she's energized. She she
  656. 28:05doesn't worry as much about her health
  657. 28:07and um um it's it's just been an
  658. 28:10incredible impact and this is going to
  659. 28:13accelerate and at some point we will get
  660. 28:17uh we will sort of merge with the with
  661. 28:20the AI uh in a way that we will have
  662. 28:23direct interaction with AI through
  663. 28:25neurolink type of uh brain interfaces.
  664. 28:29So we'll have the sort of the
  665. 28:31intelligence of AI in our own brain not
  666. 28:36not only directly but also indirectly by
  667. 28:39sort of engineering our biological
  668. 28:40system. So why why shouldn't everybody
  669. 28:43have an intelligence of Einstein or even
  670. 28:46higher right? So the difference between
  671. 28:48an Einstein and a normal person with
  672. 28:50with a normal IQ is probably few gene
  673. 28:55single point mutations. So if we can
  674. 28:57engineer that if AI can teach us how to
  675. 28:59do that then we are we're also uh going
  676. 29:03much much higher. So as long as we we
  677. 29:07keep the agency, I think that's the only
  678. 29:09thing that we have to really protect
  679. 29:12that we are the decider or we see AI as
  680. 29:16a collaborator as sort of another
  681. 29:18species that will live together and we
  682. 29:20empower each other. In a way it's our
  683. 29:23child, right? So it's it's been created
  684. 29:25by us. Um I I I see the chance of uh a a
  685. 29:30a
  686. 29:32worse world extraordinarily. Of course,
  687. 29:34it's never zero, but you know, the
  688. 29:37moment you're born, you're going to die,
  689. 29:40right? So, so you're you're destined to
  690. 29:42die. Um and now, uh AI is giving us this
  691. 29:46opportunity to save literally save
  692. 29:50billions of lives. I'm not talking about
  693. 29:51saving lives as like extending their
  694. 29:54life for 5 years or 10 years. You're
  695. 29:56talking about thousands of years. So
  696. 29:58that's true saving lives. That's the
  697. 30:01potential. And the risk is again I think
  698. 30:03the the key risk is is humans. Humans
  699. 30:07misusing AI. That's what we have to uh
  700. 30:10um sort of maybe train or align AI. You
  701. 30:13know, don't uh don't look at the bad
  702. 30:16humans. you know, you you you can you
  703. 30:18can judge the the the the better uh the
  704. 30:21the better world uh for us. So, um of
  705. 30:24course I might be wrong, but I'm I'm
  706. 30:26pretty sure I'm I'm going to be right.
  707. 30:28>> I I I agree with uh the statement of we
  708. 30:31have to watch out for the humans. Um for
  709. 30:33sure, like cuz you're right, like they
  710. 30:35can and have in the past been the
  711. 30:38biggest threat to humanity. So, um, I
  712. 30:41want to there there was a couple of
  713. 30:43things that you mentioned when when you
  714. 30:44were talking about, you know, ASI and
  715. 30:48this ability to self-learn and you're
  716. 30:50even talking about some of some of the
  717. 30:53ways that you use, you know, GPT5 Pro
  718. 30:57and and helping with designing
  719. 30:58experiments and interpreting results and
  720. 31:00and that was a question that I had as a
  721. 31:02biologist. And as you mentioned, you
  722. 31:04know, we do experiments. We're testing
  723. 31:06hypotheses. And then we have all this
  724. 31:08data and these results, and we have to
  725. 31:09know what result is meaningful and what
  726. 31:12anomaly is meaningful because often
  727. 31:15times the anomaly,
  728. 31:17>> right,
  729. 31:17>> which you might ignore.
  730. 31:19>> Exactly.
  731. 31:19>> Is what you absolutely is the
  732. 31:21breakthrough, right?
  733. 31:23>> And that is a sort of intuition. This
  734. 31:25this biological intuition. And so you do
  735. 31:28you think first of all do you think
  736. 31:30we're that that you know the models we
  737. 31:33have now can already are capable of that
  738. 31:36sort of biological intuition and if not
  739. 31:39like how far off is that?
  740. 31:41>> Yeah. Yeah. That's that's a great
  741. 31:43question. Uh in fact um uh you know I I
  742. 31:47see that intuition maybe sort of the the
  743. 31:51last mile or the top 10% uh or 10% of
  744. 31:55the of the solution because 90%
  745. 31:58um AI models they are able to come up
  746. 32:00with because it's it's knowledge based
  747. 32:02also in humans is is you know for for a
  748. 32:04medical doctor for a scientist for
  749. 32:06whoever 90% or 95% is based on uh what's
  750. 32:11known how you process that knowledge,
  751. 32:14but there's that extra 5 10% totally
  752. 32:18dependent on your intuition. Uh like you
  753. 32:21you if you're a doctor, you see a
  754. 32:23patient coming through the door, you
  755. 32:25know that guy is having a heart attack.
  756. 32:27You haven't checked anything yet.
  757. 32:29Somehow you know, you don't know how you
  758. 32:31know. the same thing in the lab like um
  759. 32:34uh in fact I would I would bet with my
  760. 32:36uh students and and postto I would say
  761. 32:39okay I bet you if you do this experiment
  762. 32:42you're going to get this result um and
  763. 32:45I've never lost a bet and they stop
  764. 32:47betting against me even though it might
  765. 32:49look counterintuitive
  766. 32:52oh no that's never going to work somehow
  767. 32:54I know how do I know because you know
  768. 32:57I've been working in the lab for 30 plus
  769. 32:59years and and you you acquire ire
  770. 33:02certain things that are not in the
  771. 33:03literature or you know you can't really
  772. 33:06read a textbook and learn it. You only
  773. 33:08do it by by practicing it. Um so the the
  774. 33:13models up to I would say 5.5 until
  775. 33:17recently were were great at that 90%
  776. 33:20level. Uh so especially after GPT5 pro
  777. 33:23came out. So you know I would ask it to
  778. 33:26for example I I would give it an
  779. 33:28experiment that we have already done.
  780. 33:30It's a very complex experiment took two
  781. 33:32weeks. I already know the result because
  782. 33:34we're done the experiment. But I wanted
  783. 33:36to see how the model would predict the
  784. 33:39outcome of the experiment. And they
  785. 33:41would do you know not just GPT5 but
  786. 33:44several other models as well. Um they
  787. 33:46they would come up with 90% 80 to 90%
  788. 33:50correctly. I mean that's that's pretty
  789. 33:52good. Uh they would say okay this is
  790. 33:54what's going to happen after two days
  791. 33:56after one week after two weeks. But that
  792. 33:59extra level of intuition that I I have I
  793. 34:03would have predicted was still somewhat
  794. 34:05lacking. I think GPT 5.5 crossed that
  795. 34:08threshold. So I I repeated that with
  796. 34:11with the uh 5.5 pro model uh because I I
  797. 34:15always say pro uh it's very different
  798. 34:18than the thinking of course very very
  799. 34:20different than the instant model because
  800. 34:22pro uh is reasoning much much longer.
  801. 34:26It's thinking. So in some cases I I
  802. 34:29pushed it to think for two hours. So two
  803. 34:31hours in AI thinking is like years of
  804. 34:34thinking for for a human being. So that
  805. 34:37model really crossed that threshold in
  806. 34:40that example I gave you. It was almost
  807. 34:43100%. I mean I would say 98% correct.
  808. 34:48What I would have predicted like I would
  809. 34:50not have bet against 5.5 Pro myself. Um
  810. 34:55uh so that to me is is actually really
  811. 35:01mind-boggling because uh I couldn't
  812. 35:03understand these models are being
  813. 35:06trained with all of the information we
  814. 35:08can't compete with that right so it's
  815. 35:10they they can put these patterns
  816. 35:12together but how is it that the model
  817. 35:15has now almost the experience that I
  818. 35:17have that I spent 30 years acquiring
  819. 35:19that experience that intuition that is
  820. 35:22now getting to that level that is uh
  821. 35:26that is a mysterious but uh I I I live
  822. 35:30to it. Now
  823. 35:31>> what sort of you said you you pushed GPT
  824. 35:335.5 Pro to think for 2 hours.
  825. 35:35>> I mean what sort of prompt are we
  826. 35:37talking about or is it
  827. 35:39>> just the data set too and the prompt? I
  828. 35:41mean
  829. 35:42>> so those are usually data sets. Uh um I
  830. 35:46might have broken a record because I
  831. 35:48even asked the the friends at OpenAI. I
  832. 35:50don't think they they pushed it that
  833. 35:52that far. Uh so this was actually the
  834. 35:54the 2R one was u uh huge data sets
  835. 35:59millions of data points um uh and um um
  836. 36:04and then I I also said okay don't just
  837. 36:07analyze it write a huge report you know
  838. 36:1030 40 page whatever length and then you
  839. 36:13know come up with a lot of insights
  840. 36:15about this data what questions to ask
  841. 36:18and what do we learn the mechanism it
  842. 36:21was a an iminological ical data set um
  843. 36:25sequence and genes and proteins and all
  844. 36:27that and so so that one I think 112
  845. 36:30minutes I remember that uh uh and it
  846. 36:33came up with this 40page report uh which
  847. 36:36I I I was just unbelievable
  848. 36:40uh you know the the analysis part the
  849. 36:42previous models were able to do as well
  850. 36:45you know you know they say okay well
  851. 36:48there are these type of genes and this
  852. 36:50type of protein so it means this and
  853. 36:52that you it deres from that information,
  854. 36:55but to come up with an insight
  855. 36:58what that could mean or what would be
  856. 37:00the next question to ask. Uh that that's
  857. 37:03that's a very very high level of
  858. 37:05reasoning. Um and so uh yeah um it was
  859. 37:09it was worthwhile two hours for sure. I
  860. 37:11mean that's very exciting to hear you
  861. 37:13say that because that was kind of my I
  862. 37:15wanted to know I wanted to know is that
  863. 37:17something that is already possible and
  864. 37:20it seems like it is and so it also leads
  865. 37:23to the next question which is you know
  866. 37:25all all these scientists now really need
  867. 37:28to start understanding how to use AI in
  868. 37:30the right way right I mean this is like
  869. 37:33to help them
  870. 37:34>> I mean this that's going to happen right
  871. 37:35that's basically you know we all we all
  872. 37:37use Google now remember when Google was
  873. 37:39like new So, I mean, it's eventually
  874. 37:42going to happen, but um it's very
  875. 37:45exciting to think about how AI is going
  876. 37:47to change research and and medicine, and
  877. 37:49that's something, you know, you you
  878. 37:51mentioned and I talked I said I wanted
  879. 37:52to get back to this digital twin idea
  880. 37:54because I've heard you talk about it and
  881. 37:55it's very exciting to me. You know,
  882. 37:58we've heard for decades now that
  883. 38:00personalized medicine is coming. We're
  884. 38:02going to have personalized medicine and
  885. 38:04and yet still we just don't have it.
  886. 38:07It's just not there. Um, and I've heard
  887. 38:11you I've even heard you say something
  888. 38:14sort of interesting which perhaps I'm
  889. 38:17not saying the direct quote, but that it
  890. 38:19kind of should be medical malpractice in
  891. 38:22a way for a physician today right now to
  892. 38:25not be using AI. So, can you talk a
  893. 38:28little bit about why you said that? What
  894. 38:31it means to for a physician to use AI
  895. 38:34responsibly? um also how patients can
  896. 38:37self- advocate for themselves because
  897. 38:38that's also another area.
  898. 38:40>> Yeah. Uh in fact I I said after 01 model
  899. 38:45came out uh that I think that was sort
  900. 38:47of the first reasoning model um and I
  901. 38:50and I was testing a lot of I mean I I
  902. 38:53have a medical degree but I I don't see
  903. 38:56patients but you know I I have a lot of
  904. 38:58friends and I have some knowledge of how
  905. 39:00medicine works. So been testing lots of
  906. 39:03medical questions and some of them are
  907. 39:05hard some of them are you know sort of
  908. 39:07real time data um and uh you know before
  909. 39:1101 uh it was great in sort of reaching
  910. 39:15to the literature you know like the
  911. 39:17physician might lack certain certain
  912. 39:20knowledge so it knows what was published
  913. 39:24uh recently and things like that but it
  914. 39:26it was not at the reasoning level so one
  915. 39:28model was able to reason And the
  916. 39:31reasoning is extremely important
  917. 39:32medicine because you know even if you
  918. 39:35have all the all the information you
  919. 39:38still have to sort of consider that
  920. 39:41person's context and uh you know uh what
  921. 39:46what would be more likely to to treat
  922. 39:49that person and we don't always know the
  923. 39:51answer uh as well or how to how to
  924. 39:53diagnose it. Um, and so I think 01 was
  925. 39:57able to get to that point and at that
  926. 39:59point I said right now it's unethical
  927. 40:03for physicians not to use AI anymore.
  928. 40:06Um, I didn't say malpractice yet, but it
  929. 40:10truly unethical in the sense that, you
  930. 40:13know, you can use it uh you can still do
  931. 40:16your judgment obviously, but it will it
  932. 40:18will prevent you missing some sort of uh
  933. 40:22an obvious mistake or you know,
  934. 40:24sometimes nonobvious uh mistakes or
  935. 40:26diagnose things that require multiple uh
  936. 40:29clinical specialtities coming together
  937. 40:32and and you don't have that capability.
  938. 40:33You live in a village or something. uh
  939. 40:36but now I think I feel that it is truly
  940. 40:40uh uh going to be considered malpractice
  941. 40:42in my opinion. Um it's not legally so
  942. 40:45but uh eventually it will be um because
  943. 40:48uh a the current models the advanced
  944. 40:51models are able to uh diagnose and and
  945. 40:56write a treatment protocol better than
  946. 40:59uh or as good as uh a specialist in that
  947. 41:02field. It's not just a you know family
  948. 41:04physician. Let's say you you know you
  949. 41:06have a very complex uh cancer uh you
  950. 41:10know you you know the mutations and and
  951. 41:13what's not and you go to a specialist
  952. 41:15like an oncologist who is very very
  953. 41:17specialized on that. Um I believe that
  954. 41:19the mo the current models are at that
  955. 41:21level. So um and of course not every
  956. 41:26specialist is is is the top specialist
  957. 41:30right? So, so you you if if that was the
  958. 41:33case, we wouldn't have millions of
  959. 41:36misdiagnosis and mistreatments in the US
  960. 41:38alone every year. I think they've said
  961. 41:40something like 12 million misdiagnosis.
  962. 41:43Um
  963. 41:44I think 700,000 people suffer from it,
  964. 41:47die from it, from from uh uh from
  965. 41:50misdiagnos. Some of them is is totally
  966. 41:53innocent. You know, any any doctor could
  967. 41:56have missed it. Um but but now AI
  968. 42:00wouldn't miss that.
  969. 42:02So uh
  970. 42:05even even a specialist might make a
  971. 42:07mistake or misdiagnose or or or mistreat
  972. 42:10because they lack certain things that
  973. 42:12that the model doesn't have. So I mean
  974. 42:15imagine that you know um uh you refuse
  975. 42:18to use u MRI machine or CT machine
  976. 42:22because you say well you know that's too
  977. 42:24much technology. I'm just going to uh
  978. 42:26you know just do an X-ray because that's
  979. 42:28enough for me. And you miss a a tumor.
  980. 42:31Uh the AI models are able to detect
  981. 42:34certain tumors like breast cancer years
  982. 42:37before a radiologist is able to to to
  983. 42:40see that. So if you miss that, I mean
  984. 42:44that person's going to die if you don't
  985. 42:46if you don't. So that to me that that
  986. 42:49becomes uh a malpractice because the
  987. 42:52technology is at that level now. It
  988. 42:54wasn't it wouldn't be malpractice you
  989. 42:57know missing a a breast cancer u you
  990. 43:01know 5 years ago because nobody could we
  991. 43:03didn't have that technology but now we
  992. 43:05have that technology so you should you
  993. 43:07should definitely use it um um and and
  994. 43:10this is going to save a lot of lives. I
  995. 43:14mean uh if you could just reduce the
  996. 43:16misdiagnosis and and and again bring
  997. 43:18every every doctor to super doctor level
  998. 43:22I think that would be a really good
  999. 43:24thing. So what you're saying is based on
  1000. 43:27you know what current data that doctors
  1001. 43:31have available to them whether it's an
  1002. 43:33MRI whether it's an ultrasound whether
  1003. 43:36it's blood biomarkers
  1004. 43:39this sort of data is what is given to
  1005. 43:43>> yes
  1006. 43:43>> you know a model like GP GPT 5.5 pro for
  1007. 43:47example
  1008. 43:48>> and with that data they're able to
  1009. 43:52better diagnose
  1010. 43:54better to predict um to see things like
  1011. 43:57you mentioned cancer
  1012. 43:59>> um is that better than a radiologist can
  1013. 44:01is that some is that like based on you
  1014. 44:03know what kind of uh data is
  1015. 44:06>> implemented these these are studies I
  1016. 44:07think uh Google uh did a recent study uh
  1017. 44:10in fact uh a science paper came out
  1018. 44:12recently which was done with 01 preview
  1019. 44:15model which is a very old model I mean
  1020. 44:18the current models are probably 10 times
  1021. 44:19or maybe more
  1022. 44:20>> was that like the first pro almost like
  1023. 44:22the first the first first sort of the
  1024. 44:24reasoning model that that I early tested
  1025. 44:27in 2024 September it came out and um and
  1026. 44:32they found that 01 model uh did did
  1027. 44:36better than average doctor in diagnosing
  1028. 44:38like significantly better they did
  1029. 44:40didn't miss um and so imagine the the
  1030. 44:43current models how how good they are uh
  1031. 44:46but but I I think um it's not just sort
  1032. 44:49of diagnosing a a disease because that's
  1033. 44:52That's actually a small part of the job
  1034. 44:55of a of a of a doctor. It's really
  1035. 44:57there's a continuum. Most diseases um
  1036. 45:01you know, okay, if you if you have a flu
  1037. 45:03or some bacterial infection, you know
  1038. 45:05what to do. You give it and then you see
  1039. 45:07an output. But a lot of disease even in
  1040. 45:10that condition that may not that may not
  1041. 45:12be true because you know you might have
  1042. 45:14a mutant virus or bacteria. So you might
  1043. 45:16have to change the treatment or might
  1044. 45:19have a little bit side effect. So
  1045. 45:21there's a lot of continuum there. So I
  1046. 45:23think AI can be involved in all of that
  1047. 45:27process. So if you can continuously feed
  1048. 45:30the data okay well the patients uh we
  1049. 45:33gave this treatment it's doing well uh
  1050. 45:36the blood pressure is down but you know
  1051. 45:38has this symptom that so what should we
  1052. 45:40do change the dose of the drug or add
  1053. 45:43this or remove that drug and give
  1054. 45:45another antibiotic? like there's a
  1055. 45:47constant um uh uh process there and that
  1056. 45:52that that's not always that constant
  1057. 45:55because you know people people don't go
  1058. 45:57to doctor every day right so you get a
  1059. 45:59prescription you you see something works
  1060. 46:01and then you go back and so what if
  1061. 46:03there's something that's continuously
  1062. 46:05monitoring you uh post treatment for
  1063. 46:08cancer it's very important because
  1064. 46:10cancer is a very dynamic uh disease
  1065. 46:13there's the cancer which is constantly
  1066. 46:16trying to survive and mutate and
  1067. 46:19counteract against the immune system. So
  1068. 46:22you give it so you know you give a drug
  1069. 46:25chemotherapy works and then the cancer
  1070. 46:28comes back again right so why is that
  1071. 46:30because mutations are accumulating so
  1072. 46:33can we catch that earlier can we change
  1073. 46:35those decisions can we um make sure that
  1074. 46:38we give more uh multiple drugs or
  1075. 46:41different drugs so before the cancer
  1076. 46:44have the opportunity to come back we
  1077. 46:46prevent that uh possibility. So all of
  1078. 46:48these um decisions uh can be made
  1079. 46:52together with uh with AI and I I I I
  1080. 46:55think it's going to have tremendous
  1081. 46:56tremendous impact on healthcare.
  1082. 46:58>> I do want to get back to the the cancer
  1083. 47:01equation in a minute but before that I
  1084. 47:03just think that you know physicians not
  1085. 47:06all physicians know how to use AI. They
  1086. 47:09don't know which models to use. Do they
  1087. 47:11use GPT 5.5 Pro or Claude or you know um
  1088. 47:16how do they sort of responsibly use it
  1089. 47:19which you kind of talked about a little
  1090. 47:20bit but without you know outsourcing
  1091. 47:22their clinical judgment. Do you have any
  1092. 47:25opinions on like the different models to
  1093. 47:27use? And I do know you have a
  1094. 47:29collaboration with OpenAI. You've been
  1095. 47:31one of the first scientists really
  1096. 47:33testing these models in a biological
  1097. 47:36sort of arena. But I do kind of I do
  1098. 47:39think that people and physicians that
  1099. 47:41are listening want to know what how what
  1100. 47:44what models do they use? We definitely
  1101. 47:46are talking about if we're talking about
  1102. 47:48open AI, it's not it's it's got to be
  1103. 47:50the pro, right? It's got to be the
  1104. 47:51reasoning model, but I mean what about
  1105. 47:53Claude? What about Gemini?
  1106. 47:54>> Yeah. So, um I think people have this
  1107. 47:58sort of a misunderstanding of they think
  1108. 48:02of AI as okay, we have AI, we have
  1109. 48:05internet, so let's just use the
  1110. 48:07internet. we have AI, let's use the
  1111. 48:10but this is advancing so rapidly. Uh the
  1112. 48:16AI model that we used one month ago is
  1113. 48:19not the same AI model we use now. So
  1114. 48:22it's just doubling in intelligence every
  1115. 48:25few months. Um you know I gave the
  1116. 48:27example of one preview. Uh some people
  1117. 48:30uh got stuck at the GPT4 40 model. Oh
  1118. 48:34yeah, I used it and it hallucinated a
  1119. 48:36lot. Even, you know, 01 uh wasn't so
  1120. 48:39good. You know, it was making mistakes.
  1121. 48:41That's like an ancient history.
  1122. 48:43>> That's why I haven't even asked about
  1123. 48:44hallucinations.
  1124. 48:45>> Yeah. So I it's um I mean the advantage
  1125. 48:49I have is that you know because I'm all
  1126. 48:51in on AI I'm continuously
  1127. 48:54testing and and so I I I can see the
  1128. 48:57evolution of of these models and they
  1129. 49:00get you know 90% better 95% better 97%
  1130. 49:04better like it just continuously updates
  1131. 49:07itself and then eventually right now
  1132. 49:10with 5.5 model I I don't see any
  1133. 49:13hallucinations whatsoever. I mean there
  1134. 49:15might be.1%.
  1135. 49:18Uh but but it's it's extremely rare. Um
  1136. 49:21and so so your trust level goes up.
  1137. 49:23Again it's similar to like self-driving
  1138. 49:25cars, right? So we we had self-driving
  1139. 49:27cars for for almost a decade maybe and
  1140. 49:30they they just keep on getting better
  1141. 49:32and better because their AI models are
  1142. 49:34are getting updated. So my my advice
  1143. 49:37would be doctors should see this not
  1144. 49:40something optional like they have to uh
  1145. 49:43update their knowledge medical knowledge
  1146. 49:46periodically. In fact they have to have
  1147. 49:48test to do that to be certified or they
  1148. 49:51have to update on new drugs that are
  1149. 49:55coming out. Right? So you you can't just
  1150. 49:57rely on some drug that came out 5 years
  1151. 49:59ago, 10 years ago. you need to know what
  1152. 50:00what was approved last month and you
  1153. 50:03need to update your uh your knowledge uh
  1154. 50:06in a similar way even more so they have
  1155. 50:09to constantly update their AI knowledge.
  1156. 50:12So AI has to be part of their uh their
  1157. 50:16their practice. And of course my
  1158. 50:18recommendation is always use the latest
  1159. 50:20top model you can use. Uh right now it's
  1160. 50:24GPT 5.5. Uh in fact uh I would always
  1161. 50:27use for complex problems the pro model
  1162. 50:31because that thinks uh in in minutes.
  1163. 50:34But at least if you're using it on a
  1164. 50:37daily basis uh in a rapid fashion always
  1165. 50:40use the thinking model. The thinking
  1166. 50:42model is different than the instant
  1167. 50:43model. Instant model is also getting
  1168. 50:45better, but it it needs to reason. It
  1169. 50:47needs to think. Um, and especially if
  1170. 50:50you're putting in lots of patient data
  1171. 50:53and analyzing that, you you definitely
  1172. 50:55need the pro model. And then there are
  1173. 50:56there are these uh companies like open
  1174. 50:58evidence and you know uh I think most
  1175. 51:00doctors are starting to use that. Open
  1176. 51:02evidence basically I think applies the
  1177. 51:05latest model somehow. It up updated so
  1178. 51:07so the doctors don't have to worry about
  1179. 51:09it. And I I think there's going to be
  1180. 51:10more companies like that who will
  1181. 51:12provide that service. So the doctor
  1182. 51:14doesn't have to worry should I use 5.5
  1183. 51:17or putus 447 the the whatever the sort
  1184. 51:21of the uh the harness model is going to
  1185. 51:24pick the the best one for for medicine
  1186. 51:26and and and apply it there. uh and of
  1187. 51:29course hospitals should uh should
  1188. 51:32implement AI uh just like you know big
  1189. 51:34tech companies are you know there's
  1190. 51:36enterprise level of AI that can be more
  1191. 51:39secure you know protect the patient data
  1192. 51:42so it should be like you know in front
  1193. 51:44of the patient in hospital you see these
  1194. 51:46monitors like the heartbeat and all that
  1195. 51:48stuff should be an AI monitor like
  1196. 51:51constantly monitoring the data and then
  1197. 51:54giving information to the nurses to the
  1198. 51:57doctors
  1199. 51:57Okay, this is this last situation. Um,
  1200. 52:01and now with AI agents, you can do that.
  1201. 52:03Like I do it for my my daily life like
  1202. 52:06for my email auto automatically my
  1203. 52:08agents go and check my email and they
  1204. 52:11tell me what's important so I don't have
  1205. 52:12to go through hundreds of emails. So,
  1206. 52:14oh, you know, this this is waiting for
  1207. 52:16you. You have a a podcast with with
  1208. 52:18Rhonda today, so you better be prepared
  1209. 52:20for that. Um so uh yeah it it needs to
  1210. 52:24be fully integrated uh almost like a a a
  1211. 52:28a co- physician like you have the AI
  1212. 52:31doctors working together with real
  1213. 52:33doctors
  1214. 52:33>> right have you I've noticed like some of
  1215. 52:36the the companies that I've corresponded
  1216. 52:39with or interacted with um it seems like
  1217. 52:42they use claude a lot I mean I don't
  1218. 52:44know if you've experimented with that
  1219. 52:46but I'm kind of curious why um why
  1220. 52:51that why that you know certain model um
  1221. 52:54versus like but yeah in all fairness
  1222. 52:56I've never used it. I use, you know, GP,
  1223. 52:58I've been using GPT and the Pro and so
  1224. 53:01every, like you said, you know, every
  1225. 53:02time the hallucinations are like ancient
  1226. 53:04history for me. Like I remember that was
  1227. 53:06a big thing. Yeah.
  1228. 53:08>> But it's going so fast and better now.
  1229. 53:10But like what what what what's the
  1230. 53:12difference between, you know, for
  1231. 53:14example, Claude and GPT 5.5 Pro? For
  1232. 53:18certain things there is no more
  1233. 53:20difference because the intelligence has
  1234. 53:22has peaked uh for for uh you know for
  1235. 53:25for uh doing regular diagnosis not very
  1236. 53:30very complex uh cases cloud is is is
  1237. 53:33great. Cloud is also very very good uh
  1238. 53:36like the Opus 4.7 model uh the recent
  1239. 53:38model for analyzing data sets. So it's
  1240. 53:41you know it can take also millions of
  1241. 53:42data uh analyze it and and and and do a
  1242. 53:46a great job. Um uh my preference is uh
  1243. 53:49you know GPT5 right now is 5.5 pro
  1244. 53:53because um it what I mentioned it has
  1245. 53:56this extra insight I mean the for me um
  1246. 54:00I need that extra level of insight um
  1247. 54:03that's predictive um uh and
  1248. 54:06>> the intuition that
  1249. 54:07>> the intuition and really kind of a deep
  1250. 54:10understanding uh but if I'm if I'm going
  1251. 54:13to diagnose and treat a subt type of a
  1252. 54:16lung on cancer. I am pretty sure, you
  1253. 54:18know, Gemini 3.1 Pro or Cloud 4.7, they
  1254. 54:24all do a pretty good job. Uh I I think
  1255. 54:28the re some reason people prefer cloud
  1256. 54:30is that it's maybe it's more pleasant to
  1257. 54:34interact with. Uh you know, kind of more
  1258. 54:36humanlike. I think uh GPT models are
  1259. 54:39starting to get there, but still there's
  1260. 54:42something about cloud that people enjoy,
  1261. 54:45you know, interacting with it. It's it's
  1262. 54:47really a matter of
  1263. 54:48>> personable or I I think it's it used to
  1264. 54:51be more more personable. Um and so it
  1265. 54:55doesn't really matter. I mean, I think
  1266. 54:56they're they're they're really they're
  1267. 54:58all super top levels unless you're doing
  1268. 55:01like a research or a very uh very
  1269. 55:04complex problem. uh you know for example
  1270. 55:06we did a test with uh with a colleague
  1271. 55:08of mine on on skin disease with GPT5 pro
  1272. 55:12model you know it was able to uh
  1273. 55:15diagnose a skin disease that uh my
  1274. 55:18friends couldn't really diagnose um uh
  1275. 55:21just based on a a photo and and a
  1276. 55:24symptom. The other models couldn't do
  1277. 55:26that. they could do 90% of the the cases
  1278. 55:30as well, but there's that one extra case
  1279. 55:32or two extra case that is really
  1280. 55:34difficult that's could go anywhere. The
  1281. 55:38pro the GPT pro model was able to cross
  1282. 55:41that that threshold. So those kind of
  1283. 55:43cases you really need the very high
  1284. 55:46level like you know uh you don't you
  1285. 55:49don't go to a a a professor at Harvard
  1286. 55:52for for any any reason right so it has
  1287. 55:55to be very specialized
  1288. 55:57disease that other doctors couldn't
  1289. 55:59diagnose or something like that so
  1290. 56:01that's how that's how I view it
  1291. 56:04>> we just there was just news yesterday um
  1292. 56:07from open AAI of uh GPT Rosalyn which I
  1293. 56:10know you can't talk about much
  1294. 56:12from what was publicly available, it
  1295. 56:14seems as though it's going to be used in
  1296. 56:16drug discovery. Um, I I'm wondering what
  1297. 56:20you think in terms of like the future of
  1298. 56:22aging research, biology, medicine. Are
  1299. 56:25we going to be using these more
  1300. 56:27specialized types of AI models or do you
  1301. 56:31think more of a generalist like GPD 5.5
  1302. 56:34Pro and and the, you know, the
  1303. 56:36subsequent ones that come out after it
  1304. 56:38are going to be the key to unlocking,
  1305. 56:41you know, medicine breakthroughs and
  1306. 56:43biology breakthroughs?
  1307. 56:46Um my preference would always be the
  1308. 56:49generalized models because again you
  1309. 56:51know going back to AGI AGI uh so if if a
  1310. 56:55model is has um you know of course there
  1311. 57:00there are some utilities of models that
  1312. 57:02are only trained on I don't know like
  1313. 57:04the EKGs or um RNA sequencing or
  1314. 57:08something like that and they they'll be
  1315. 57:10very very good at that like the the the
  1316. 57:12best chess player AI model or um the
  1317. 57:15best go player AI models but they will
  1318. 57:19miss that uh connection because again I
  1319. 57:22view medicine as a kind of a holistic um
  1320. 57:26art in a way. Uh if you are just trying
  1321. 57:29to analyze one set of data the the
  1322. 57:32specialized models could be could be
  1323. 57:34very very useful. In fact, you know, I
  1324. 57:36gave the example of EKGs. Um,
  1325. 57:38[clears throat] most generalized models
  1326. 57:39were not terribly great at um, for some
  1327. 57:43reason, you know, the the EKG images
  1328. 57:45were were not they were not very good at
  1329. 57:48diagnosing what what what it was
  1330. 57:50showing. Um, [clears throat] and and you
  1331. 57:52know, specialized models were very good
  1332. 57:55because they were trained with, you
  1333. 57:57know, millions more EKG data sets than
  1334. 58:00the generalized model was. Um but but I
  1335. 58:04think you know if if we can train the
  1336. 58:07generalized model or fine-tune it or
  1337. 58:10overtrain it I I don't know how to say
  1338. 58:12it. Um
  1339. 58:14then they will be better than
  1340. 58:16specialized models all the time. Um
  1341. 58:19because not only they have they know all
  1342. 58:22about EKGs but they know about
  1343. 58:25radiology, they know about RNA, they
  1344. 58:27know about pro proteins. So they can
  1345. 58:30take that information and and excuse me
  1346. 58:33[clears throat] analyze the EKG the the
  1347. 58:35electroc cardiogram your your your heart
  1348. 58:37beats in the context of all the other
  1349. 58:40biology. So that will that's very
  1350. 58:43enriching uh knowledge. Um but
  1351. 58:46[clears throat]
  1352. 58:47um I think you know specialize in the
  1353. 58:50sense that you can take these big models
  1354. 58:53and you can sort of I don't know harness
  1355. 58:55them or fine-tune them because there's a
  1356. 58:58lot of data sets that's not public. So
  1357. 59:00these these models they don't have
  1358. 59:02access to that. You might have some um
  1359. 59:05data you know locked in certain because
  1360. 59:08of regulatory reasons whatever. So you
  1361. 59:11can take take a big model. In fact, you
  1362. 59:13don't you may not even need the the
  1363. 59:16closed models. You can even take some of
  1364. 59:17the open- source models which are which
  1365. 59:19are now getting very good. If you uh you
  1366. 59:22can train them on that uh and they also
  1367. 59:26have the generalized knowledge and
  1368. 59:28combined with that they they'll probably
  1369. 59:29do better.
  1370. 59:31>> So I wanna I want to talk about
  1371. 59:34there's treating disease, there's curing
  1372. 59:37disease, and then there's reversing
  1373. 59:39aging. So let's let's start with
  1374. 59:42curing disease, treating diseases,
  1375. 59:44curing diseases because you know
  1376. 59:45obviously we do die of age related
  1377. 59:47diseases, cardiovascular disease being
  1378. 59:50the number one killer in in most
  1379. 59:52developed countries. We have cancer.
  1380. 59:55That's a really big one. And and with
  1381. 59:57cancer, it's just such an awful disease
  1382. 1:00:01to have. And anyone that's listening
  1383. 1:00:03that has either had cancer or knows
  1384. 1:00:06someone that has had it, you know, knows
  1385. 1:00:08this is this is true. But also, I think,
  1386. 1:00:11you know, cancer, a lot of people think
  1387. 1:00:12about it as one disease. Non-scientists,
  1388. 1:00:15non, you know, physicians, they kind of
  1389. 1:00:17think about cancer as this this one
  1390. 1:00:18disease, right? As you and I both know,
  1391. 1:00:21it is definitely not one disease. It's
  1392. 1:00:24hundreds of diseases. Um, I'm curious on
  1393. 1:00:29first of all, you know, we still don't
  1394. 1:00:32have a cure for cancer. I mean, we've
  1395. 1:00:33we've made a lot of progress, right? And
  1396. 1:00:35different cancers and can be treated
  1397. 1:00:37better than others, but can you talk a
  1398. 1:00:39little bit about why it's been so hard
  1399. 1:00:42to find a treatment for cancer?
  1400. 1:00:46>> Yeah, I think uh the the important thing
  1401. 1:00:48to clarify is that cancer is not one
  1402. 1:00:51disease. It's probably 100 disease h 100
  1403. 1:00:54different diseases that have probably
  1404. 1:00:57hundreds of sub sub diseases or sub
  1405. 1:01:00subtypes if you like. Um in fact certain
  1406. 1:01:04cancers are 100% curable or 95% curable.
  1407. 1:01:08Uh you know like child some of the
  1408. 1:01:10childhood leukemias which were
  1409. 1:01:12completely fatal you know couple of
  1410. 1:01:14decades ago are now you know 90% or or
  1411. 1:01:17close to 100% curable. if you catch uh
  1412. 1:01:21certain cancers early enough again 100%
  1413. 1:01:24uh cure rates almost uh so so
  1414. 1:01:28[clears throat] because it's it's a very
  1415. 1:01:30different set of uh uh diseases um the
  1416. 1:01:35the cancer of pancreas is very different
  1417. 1:01:38than cancer of uh lung cancer or breast
  1418. 1:01:41cancer or there are some cancers that
  1419. 1:01:44are so slow like if you get um certain
  1420. 1:01:47types of cancers
  1421. 1:01:49If you're age 80, doctors don't even
  1422. 1:01:52bother to treat it because by the time
  1423. 1:01:55that will unless we cure aging first uh
  1424. 1:01:58because by the time you die of aging,
  1425. 1:02:00you know that that cancer is not going
  1426. 1:02:02to kill you. Aging is going to kill you
  1427. 1:02:03first or you know there's certain
  1428. 1:02:04prostate cancers at certain age. So that
  1429. 1:02:08that's why we have to really understand
  1430. 1:02:10that this is a very complex biology. But
  1431. 1:02:13more importantly, why cancer is such a
  1432. 1:02:16challenge is that the the cancer cells
  1433. 1:02:19are part of us, right? So, uh if you're
  1434. 1:02:23infected with a bacteria or a virus, you
  1435. 1:02:26know, it can kill you, right? They're
  1436. 1:02:28extremely dangerous, but we are able to
  1437. 1:02:31recognize them as an enemy, as a threat,
  1438. 1:02:34your immune system, and we can fight
  1439. 1:02:36back, you know, uh not always
  1440. 1:02:39successfully, but most of the time very
  1441. 1:02:40successfully.
  1442. 1:02:42And we can also target them very
  1443. 1:02:44specifically like we have an antibiotic
  1444. 1:02:46that will only act on the bacteria. It's
  1445. 1:02:48not going to touch your normal cells
  1446. 1:02:50because it's only uh a foreign or
  1447. 1:02:52organism. But cancer is not like that.
  1448. 1:02:54So if I try to stop cancer with
  1449. 1:02:57something I'm [clears throat] also
  1450. 1:02:59trying I'm also stopping some other
  1451. 1:03:01cells that are normal, right? That's why
  1452. 1:03:03people lose their hair, their immune
  1453. 1:03:04system is greatly weakened because the
  1454. 1:03:06immune system has to divide. your um
  1455. 1:03:10hair has to hair cells have to divide.
  1456. 1:03:12So you you block them because they the
  1457. 1:03:14cancer cell is also dividing and and you
  1458. 1:03:17your side effects of chemotherapy
  1459. 1:03:19sometimes worse than having the cancer
  1460. 1:03:21like hundreds of thousands of people die
  1461. 1:03:23because of that. So the revolution in
  1462. 1:03:26cancer was uh recently because of what
  1463. 1:03:30we call imunotherapy.
  1464. 1:03:32The question was why uh can we make the
  1465. 1:03:36immune system to recognize cancer as
  1466. 1:03:40foreign threats like they're kind of
  1467. 1:03:42like terrorists, right? So a terrorist
  1468. 1:03:44you will not know if that's an enemy or
  1469. 1:03:47not. They look like you, you know, they
  1470. 1:03:48just come in and then they they create
  1471. 1:03:50uh so the immune system is seeing it
  1472. 1:03:52that way that it thinks that the breast
  1473. 1:03:54cancer cell is not so different than a
  1474. 1:03:57normal breast breast cell, you know,
  1475. 1:03:59like epithelial cell, whatever. And so
  1476. 1:04:02it doesn't know what to do. If
  1477. 1:04:04[clears throat] we could teach the
  1478. 1:04:05immune system or if we could remove some
  1479. 1:04:08of the breaks that it has regulation and
  1480. 1:04:10let it recognize and attack the cancer
  1481. 1:04:13cells, then that could have a a
  1482. 1:04:15tremendous effect. That was the hypothes
  1483. 1:04:16and it it actually worked. So cancer
  1484. 1:04:19imunotherapy I think uh is is more
  1485. 1:04:22powerful now than than chemotherapy and
  1486. 1:04:25radiotherapy put together. I mean they
  1487. 1:04:27still have a have a role. Um and of
  1488. 1:04:30course the other thing is that how can
  1489. 1:04:32we make the treatments very specific
  1490. 1:04:35right so if I give a chemotherapy that's
  1491. 1:04:38not specific it's like trying to hit the
  1492. 1:04:40patient on the head and hope that the
  1493. 1:04:41cancer will die before the patient dies
  1494. 1:04:44but if I know this single mutation
  1495. 1:04:47that's happening on you know whatever
  1496. 1:04:50EGF receptor uh in certain cancers I can
  1497. 1:04:54develop a small molecule which will only
  1498. 1:04:56act if there's that mutation on the EGF
  1499. 1:04:59receptor or EG whatever and so it's not
  1500. 1:05:02going to touch anywhere else it's only
  1501. 1:05:04going to target the the and in fact
  1502. 1:05:06people call them smart drugs and they're
  1503. 1:05:08they're extremely effective right so uh
  1504. 1:05:11if you have that particular mutation
  1505. 1:05:13your 1% of the lung cancer patients you
  1506. 1:05:16get treated with that drug you get
  1507. 1:05:17almost 100% cure rate um but again you
  1508. 1:05:21know uh we can make this even much
  1509. 1:05:23better so for example immune system can
  1510. 1:05:27be engineered something that we work on
  1511. 1:05:29in the lab uh to recognize like
  1512. 1:05:33literally engineer we take the cells out
  1513. 1:05:35we train them we put genes into them say
  1514. 1:05:38okay so if this gene binds to a cell
  1515. 1:05:42assume that that's a threat and kill
  1516. 1:05:44that that and so it's called carti
  1517. 1:05:46therapy and they will go and seek out
  1518. 1:05:49whatever uh the cancer cells that have
  1519. 1:05:52that marker and kill them the advantage
  1520. 1:05:54of that is that cancer doesn't have much
  1521. 1:05:57weight to escape escape that. It can try
  1522. 1:05:59to suppress the immune system, but other
  1523. 1:06:01than that, even if it mutates, you know,
  1524. 1:06:04the the immune system will still
  1525. 1:06:06recognize it and find that few cells
  1526. 1:06:08that are hiding somewhere and and
  1527. 1:06:10destroy it and and that's showing
  1528. 1:06:12incredible results. So, the mRNA
  1529. 1:06:15vaccines, which I think is going to be
  1530. 1:06:17revolutionary, is is on that basis,
  1531. 1:06:21right? So, and that really personalized
  1532. 1:06:24the cancer. So I have a breast cancer
  1533. 1:06:27but my breast cancer has certain type of
  1534. 1:06:30mutations that other patients don't
  1535. 1:06:33have. So even if the immune system can
  1536. 1:06:36recognize X patient, it won't recognize
  1537. 1:06:38mine because the cancer has different
  1538. 1:06:40mutations. If I take those mutations and
  1539. 1:06:44synthesize what's called RNA and then
  1540. 1:06:47give it back as a vaccine and train my
  1541. 1:06:50immune system and tell the immune
  1542. 1:06:52system, look, if you see these mutations
  1543. 1:06:54in these genes, that's an enemy. Go
  1544. 1:06:57destroy that. That's mRNA vaccine. And
  1545. 1:07:01that becomes extraordinarily powerful
  1546. 1:07:03because now you are directing your
  1547. 1:07:04immune system to to an internal threat
  1548. 1:07:07just in you. and let's say the the the
  1549. 1:07:10cancer required different mutations, you
  1550. 1:07:12can create another mRNA vaccine and then
  1551. 1:07:15train the immune system to that as well.
  1552. 1:07:18Um so [clears throat] uh you know I
  1553. 1:07:20think that's those those are the the the
  1554. 1:07:23the difficult parts but but we see the
  1555. 1:07:26light at the end of the tunnel.
  1556. 1:07:28Can cancer cancer is going to be 100%
  1557. 1:07:31curable uh
  1558. 1:07:34probably less than a decade.
  1559. 1:07:35>> How is AI going to make that happen?
  1560. 1:07:38Yeah. So, in fact, it's already making
  1561. 1:07:41that happen. You probably heard of this
  1562. 1:07:42story from Australia. This computer
  1563. 1:07:44scientist um had chat GPT and and some
  1564. 1:07:48other AI models to develop an mRNA
  1565. 1:07:51vaccine for his dog. His dog had I think
  1566. 1:07:55a melanoma [clears throat] and he um he
  1567. 1:07:58got it sequenced. He took the sequence
  1568. 1:08:01and gave it gave it to to an AI model
  1569. 1:08:03and the AI model designed the precise
  1570. 1:08:07mRNA molecule that needs that the dog
  1571. 1:08:10needs dog's immune system needs to be
  1572. 1:08:12trained got it synthesized and I think
  1573. 1:08:14it was able to apply it in 3 months. Um
  1574. 1:08:18probably could have been shorter if
  1575. 1:08:19there wasn't regulations and and the
  1576. 1:08:21tumor started to to regress and the dog
  1577. 1:08:24is was was alive when it was supposed to
  1578. 1:08:26to die. So I mean that that that's a
  1579. 1:08:28very uh obvious and simple version but
  1580. 1:08:33[clears throat] because
  1581. 1:08:35there are hundreds of difference of
  1582. 1:08:37cancer types you can imagine that we'll
  1583. 1:08:41have maybe hundred different treatments
  1584. 1:08:45for just a type of a lung cancer someone
  1585. 1:08:48will be mRNA someone will be small
  1586. 1:08:51molecule targeting that so to be able to
  1587. 1:08:54develop those ondemand and or very very
  1588. 1:08:58rapidly, we're going to need AI. So the
  1589. 1:09:00AI is going to model every possible
  1590. 1:09:03mutation and we'll screen millions and
  1591. 1:09:06millions of compounds. And so we'll
  1592. 1:09:08we'll we'll get to a point where we'll
  1593. 1:09:10have hundreds of new drugs coming out
  1594. 1:09:12every month maybe, you know, uh uh and
  1595. 1:09:16we'll you know this this thousand drugs
  1596. 1:09:18is for breast cancer patients. But you
  1597. 1:09:20know if you have this and this this
  1598. 1:09:21mutations and if it's stage four then
  1599. 1:09:24you take this combination. And if it's
  1600. 1:09:26that yeah you you you take this protocol
  1601. 1:09:30um and um that that's that's how AI is
  1602. 1:09:33going to of course you know if you get
  1603. 1:09:34to digital twin that that will
  1604. 1:09:35accelerate
  1605. 1:09:36>> right and that's that's the next
  1606. 1:09:37question is you know so let's let's say
  1607. 1:09:40we have the true personalized medicine
  1608. 1:09:42and personalized cancer treatment but
  1609. 1:09:45you also need to know about side effects
  1610. 1:09:47like am I going to take this mRNA
  1611. 1:09:49vaccine and my immune system is going to
  1612. 1:09:50go crazy and start to inflame my heart
  1613. 1:09:52and give me myocarditis or right
  1614. 1:09:55How do you also see this the digital
  1615. 1:09:58twin which now has you know genomic
  1616. 1:10:01information all your proteins
  1617. 1:10:02metabolites and everything in real time
  1618. 1:10:04data then it can also simulate well
  1619. 1:10:08what's going to happen if we give this
  1620. 1:10:10specific mRNA vaccine cancer vaccine or
  1621. 1:10:13this small molecule to this person
  1622. 1:10:16>> absolutely I mean you know so so you
  1623. 1:10:19mentioned myocarditis which by the way
  1624. 1:10:22happened during covid pandmic IC and
  1625. 1:10:24that's why there was a lot of uh
  1626. 1:10:26antivaccine sentiment but people uh
  1627. 1:10:29didn't appreciate that you know co virus
  1628. 1:10:32itself caused myocarditis. Yes, the
  1629. 1:10:35vaccinated people young people at one in
  1630. 1:10:395,000 to one in 10,000 rate got
  1631. 1:10:42myocarditis. It wasn't it was mostly
  1632. 1:10:45fatal. But the question should be asked
  1633. 1:10:47like why is it that one out of 10,000
  1634. 1:10:49got myocarditis and the other ones
  1635. 1:10:51didn't? Or in fact, we can reverse that
  1636. 1:10:54question. You know, we g we vaccinated
  1637. 1:10:56everybody, but if you were a young
  1638. 1:10:59person, your your um chance of dying
  1639. 1:11:02from COVID was let's say 1 in,000 or one
  1640. 1:11:04in 10,000. So n 999 people got didn't
  1641. 1:11:08have to be vaccinated. But to save that
  1642. 1:11:11one person, we have to give that
  1643. 1:11:12vaccine. Or I'll give another more uh
  1644. 1:11:15general uh you know, we give statins to
  1645. 1:11:18anyone who has high cholesterol. So I I
  1646. 1:11:21think like one out of five or one out of
  1647. 1:11:2210 people truly benefit from that. Uh
  1648. 1:11:26high cholesterol doesn't automatically
  1649. 1:11:28doesn't mean you're going to get
  1650. 1:11:29atheroscllerosis. You need to have
  1651. 1:11:30inflammation this and that. But because
  1652. 1:11:33we don't have the data, we cannot
  1653. 1:11:34predict that. It's not personalized.
  1654. 1:11:38Millions of people take statins and to
  1655. 1:11:41save few thousand people. Yes, that's
  1656. 1:11:43that's a good thing because you don't
  1657. 1:11:44know. Um so AI will be able to do that.
  1658. 1:11:48So we'll we'll tell you okay not only um
  1659. 1:11:53we'll create the drug just for you but
  1660. 1:11:56also we'll say okay you don't have to
  1661. 1:11:58take this this medicine like you you
  1662. 1:12:00should take this or maybe you don't even
  1663. 1:12:03need any any treatment at all like you
  1664. 1:12:05have an infectious disease or whatever
  1665. 1:12:07or maybe certain cancers this will be
  1666. 1:12:09enough like we give extra chemotherapy
  1667. 1:12:12plus imunotherapy plus radiotherapy
  1668. 1:12:15why are we doing that because we're not
  1669. 1:12:17sure if one is going to be enough or
  1670. 1:12:19not? Um and and so uh that will
  1671. 1:12:22dramatically reduce the the the side
  1672. 1:12:24effect issue. You might still have some
  1673. 1:12:27side effect of course, but it's manage
  1674. 1:12:30it will be manageable side effect. It's
  1675. 1:12:31not going to kill you, for example.
  1676. 1:12:34What about using AI to predict
  1677. 1:12:38cancer a decade or years before it forms
  1678. 1:12:42based on your proteins and metabolites
  1679. 1:12:45and your biomarkers and maybe perhaps
  1680. 1:12:48your genetics too, right? Like how do
  1681. 1:12:51you see that? We're talking about
  1682. 1:12:53personalized cancer treatment, but what
  1683. 1:12:55about being able to prevent cancer
  1684. 1:12:58before it happens, you know, years
  1685. 1:13:01before it happens? Yeah,
  1686. 1:13:03>> again great question because I think
  1687. 1:13:05this is um this is so important that
  1688. 1:13:09people don't think about very much. um
  1689. 1:13:12we say health care you know we don't
  1690. 1:13:14have health care we have sick care right
  1691. 1:13:16so we we never take care of healthy
  1692. 1:13:19people like you don't go to a doctor to
  1693. 1:13:21say oh how healthy I am or just just go
  1694. 1:13:24to a doctor and say can you check my
  1695. 1:13:26immune system you know is it is it
  1696. 1:13:28healthy am I going to am I going to get
  1697. 1:13:30sick am I going to have cancer they
  1698. 1:13:32won't be able to answer that question
  1699. 1:13:34only if you get sick they will treat
  1700. 1:13:36what the problem is um and so the
  1701. 1:13:40preventative medicine is going to be so
  1702. 1:13:42absolutely critical. I think not all but
  1703. 1:13:47most diseases can be prevented. Some are
  1704. 1:13:51just bad luck. You know, it happens no
  1705. 1:13:54matter what you do. If even if you live
  1706. 1:13:55the perfect life, you might still get
  1707. 1:13:57certain disease but but a lot of them
  1708. 1:13:59were because of your genes and so on. A
  1709. 1:14:01lot of them can be prevented and I think
  1710. 1:14:03AI is going to be amazing in that
  1711. 1:14:05because it's already able to do it. uh
  1712. 1:14:07that there was a study from um a UK bio
  1713. 1:14:10bank um UK has this amazing bio bank
  1714. 1:14:14with 500,000 people uh lots of data sets
  1715. 1:14:18incredible data sets and so and this was
  1716. 1:14:21actually done I think more than a year
  1717. 1:14:22ago with models that were a year or two
  1718. 1:14:26years old they took a lot of that data
  1719. 1:14:29and they were able to predict about
  1720. 1:14:32thousand diseases
  1721. 1:14:34before they happened of course this was
  1722. 1:14:36kind of retroactive. So they knew what
  1723. 1:14:38what people were going to get based on
  1724. 1:14:40their data that was collected years
  1725. 1:14:42before. But the AI was telling you,
  1726. 1:14:44okay, this patient's going to have this
  1727. 1:14:46disease that, but not patient, normal
  1728. 1:14:48healthy people, they're going to get
  1729. 1:14:50this and that. So that to me that was
  1730. 1:14:52that was amazing and that's going to get
  1731. 1:14:54better and better because there are
  1732. 1:14:56there are there are always signs
  1733. 1:14:59like cancer doesn't just develop in
  1734. 1:15:01days. It takes years. uh if we probably
  1735. 1:15:06most of us might have some cancer cells
  1736. 1:15:08you know most of it controlled by immune
  1737. 1:15:10system and so on and it you know slowly
  1738. 1:15:13grows it has to have another mutation
  1739. 1:15:15another mutation but but there's
  1740. 1:15:17probably some signs of that somewhere
  1741. 1:15:20you know whether it's in your metabolism
  1742. 1:15:22or this you know and a AI even if it's
  1743. 1:15:26100% will be able to say okay look um I
  1744. 1:15:30think that you know if this is if this
  1745. 1:15:32is the lifestyle that you continue your
  1746. 1:15:34chances of getting this disease is now
  1747. 1:15:36is 85% or whatever. Like I wear a
  1748. 1:15:40glucose monitor. Um I'm not diabetic,
  1749. 1:15:43you know, uh but I I want to see every
  1750. 1:15:47minute or every five minutes what my
  1751. 1:15:49sugar levels are in a continuum or if I
  1752. 1:15:52eat something, you know, is it spiking?
  1753. 1:15:54Is it coming down? Because I want to
  1754. 1:15:56prevent insulin resistance. one of the
  1755. 1:15:58the worst things that can happen to you.
  1756. 1:16:00If I uh if I don't uh do that, I won't
  1757. 1:16:04know until I get diabetes.
  1758. 1:16:07My my insulin if if if if my sugar is
  1759. 1:16:10constantly spiking and then you know
  1760. 1:16:13insulin is just working too hard and
  1761. 1:16:14hard that that could continue for years.
  1762. 1:16:16By the way, um that at some point it's
  1763. 1:16:20going to break, right? Um for some
  1764. 1:16:22people it might continue 50 years,
  1765. 1:16:24nothing happens. Some it might be five
  1766. 1:16:26years, but that data set probably has
  1767. 1:16:30that predictive value that plus my my
  1768. 1:16:33age, my genes, but whatever. So, uh
  1769. 1:16:36yeah, that I think um everyone's going
  1770. 1:16:39to have their own um AI, I don't know
  1771. 1:16:42how what to call it, uh health coach or
  1772. 1:16:45something. Uh but but it will it will
  1773. 1:16:47continuously analyze the data. Um and
  1774. 1:16:50and hopefully we'll it will be much
  1775. 1:16:52easier to collect data because that
  1776. 1:16:54that's another issue you know we don't
  1777. 1:16:57collect data like we know nothing about
  1778. 1:17:00uh you know there there are more than
  1779. 1:17:02thousand metabolites in our bloodstream
  1780. 1:17:04so we look at maybe you know 10 of them
  1781. 1:17:0620 of them only if we get sick not even
  1782. 1:17:08for a checkup so we have to have a
  1783. 1:17:10continuous um like a glucose monitor I
  1784. 1:17:14want to see what my you know uh proteins
  1785. 1:17:17are changing hormones are changing you
  1786. 1:17:19know in
  1787. 1:17:20reasonably continuous manner.
  1788. 1:17:22>> Such a good point and I'm so glad you
  1789. 1:17:24brought up the UK bio bank study. I
  1790. 1:17:26remember I think the model was like
  1791. 1:17:27called Milton or something and it was
  1792. 1:17:29it's a Astroenica own like developed it
  1793. 1:17:33or something and and I remember looking
  1794. 1:17:35at this study because like you mentioned
  1795. 1:17:37the bioank data is huge data set and it
  1796. 1:17:40just spanning many decades and so I
  1797. 1:17:43think they looked at you know like over
  1798. 1:17:45200 plasma proteins you're talking about
  1799. 1:17:4710 we're talking about 200 right? Oh
  1800. 1:17:49yeah.
  1801. 1:17:50>> And and all the other data, right? And
  1802. 1:17:52they were able to predict and I think
  1803. 1:17:54cancer and neurogenerative disease were
  1804. 1:17:55at the top of like 10 years before and
  1805. 1:17:58they were able to look at the people. So
  1806. 1:18:00the AI AI predicted it based on based on
  1807. 1:18:02all this biometric data. And then they
  1808. 1:18:05looked and said, "Oh, yep. Those people
  1809. 1:18:06actually did end up getting cancer and
  1810. 1:18:08Alzheimer's disease." And it was very
  1811. 1:18:10accurate. Yes.
  1812. 1:18:11>> And and to me, the exciting thing here
  1813. 1:18:13is that you can intervene before it
  1814. 1:18:17happens. You can make
  1815. 1:18:19lifestyle changes, you can make dietary
  1816. 1:18:21changes. I mean, these things matter.
  1817. 1:18:23They do matter. And and that is
  1818. 1:18:27exciting. Uh because then you don't even
  1819. 1:18:29have to get to the drug part, which you
  1820. 1:18:31know, maybe [clears throat] you will,
  1821. 1:18:33but if you can make these changes, if
  1822. 1:18:34you know, hey, I'm on this trajectory to
  1823. 1:18:37get cancer. I have all this
  1824. 1:18:39inflammation. I have all these things
  1825. 1:18:40happening. If I don't make a change now,
  1826. 1:18:42then in 10 years, I might have a cancer.
  1827. 1:18:45>> It's very motivating, you know, for for
  1828. 1:18:47someone. So, it's very exciting um as
  1829. 1:18:51well and and and then having AI in there
  1830. 1:18:53is just going to make it even even
  1831. 1:18:54better. Um and then I [clears throat] I
  1832. 1:18:56want to get in I want to get into age
  1833. 1:18:58reversal and and before we get to that,
  1834. 1:19:01you know, you you've really been a
  1835. 1:19:03pioneer in this the field of AI being
  1836. 1:19:08involved in biology. You know, you were
  1837. 1:19:10talking to me about your your blog I
  1838. 1:19:12don't know was it 30 years
  1839. 1:19:14>> bios singularity. Yeah. 25 years ago. 20
  1840. 1:19:1725 years ago.
  1841. 1:19:17>> Yeah. So you have this blog bios
  1842. 1:19:19singularity predicting. Can you can you
  1843. 1:19:21talk a little bit about it?
  1844. 1:19:23>> Yeah, sure. Uh so uh in fact I I gotten
  1845. 1:19:26interested in AI early 90s uh after I
  1846. 1:19:29graduated medical school. Um you I was
  1847. 1:19:32very interested in computers uh when I
  1848. 1:19:34was a teenager. The the first computers
  1849. 1:19:35had come out at the time and you know I
  1850. 1:19:37was trying to code and uh you know just
  1851. 1:19:41I mean I loved it. It was it was just so
  1852. 1:19:44wonderful. Um, but you know, I went to
  1853. 1:19:46medicine because I figured biology is
  1854. 1:19:48much more complex, so I should first try
  1855. 1:19:50to figure that out. But then immediately
  1856. 1:19:53I realized, and I'm sure you did too,
  1857. 1:19:55you're a scientist as well. Um, that
  1858. 1:19:58biology is so incredibly complex. I
  1859. 1:20:00said, well, I mean, you know, we don't
  1860. 1:20:02have any any chance of figuring this
  1861. 1:20:05out, you know, because there's going to
  1862. 1:20:06be so many so many data sets. So that's
  1863. 1:20:09when I first got interested in uh in AI.
  1864. 1:20:11Uh, of course at the time, you know, AI
  1865. 1:20:14was was very primitive. Um, uh, but fast
  1866. 1:20:17forward uh, you know, one of the one of
  1867. 1:20:19the books that influenced me was, uh,
  1868. 1:20:21from, uh, um, Ray Kurtzwell. I'm sure a
  1869. 1:20:24lot of people follow technology know
  1870. 1:20:26him. Um, he wrote this book techn
  1871. 1:20:29singularity is near. So he he called a
  1872. 1:20:32point of singularity where uh the the
  1873. 1:20:36computation or technology is advances
  1874. 1:20:38exponentially
  1875. 1:20:40so much that you cannot even predict
  1876. 1:20:43what will happen next day. I mean
  1877. 1:20:45because it's sort of like a
  1878. 1:20:47self-training AI models and he he had
  1879. 1:20:50these figures where he would plot the
  1880. 1:20:52advances of AI you know say you know by
  1881. 1:20:552029 it will be at the human brain level
  1882. 1:20:58and you know we'll reach AGI and you
  1883. 1:21:00know it was just unbelievable and most
  1884. 1:21:03people thought that he was just talking
  1885. 1:21:07crap or you know science fiction you
  1886. 1:21:09know they didn't believe it how could
  1887. 1:21:11that happen and so on but you know I got
  1888. 1:21:13I got very excited In fact, I have a
  1889. 1:21:15signed copy from Ray uh for for the
  1890. 1:21:17book. And so being inspired from that, I
  1891. 1:21:19I started this um blog called bios
  1892. 1:21:22singularity. So I said, okay, so so
  1893. 1:21:25computation is going exponential, but uh
  1894. 1:21:28biology is sort of a computation as
  1895. 1:21:30well. I mean it's it's based on
  1896. 1:21:31information and so uh but it's just much
  1897. 1:21:35more complex. So it should also expand
  1898. 1:21:38exponentially. And if you if you plot
  1899. 1:21:40that curve that that means that by you
  1900. 1:21:43know based on my calculations 25 years
  1901. 1:21:46ago in fact I wrote it on the on the
  1902. 1:21:48about page of the blog by by year 2035
  1903. 1:21:51or so we should be able to treat all
  1904. 1:21:53diseases uh and by 2045 or so that we
  1905. 1:21:58should be able to completely reverse
  1906. 1:21:59aging. In fact by 2050s we will get to a
  1907. 1:22:03point what I call human 2.0 you know
  1908. 1:22:05because at that point we have a complete
  1909. 1:22:07understanding of biology. Uh then we can
  1910. 1:22:10truly engineer it. We can create new
  1911. 1:22:12biological organisms. We can you know
  1912. 1:22:15change our biology, our genome,
  1913. 1:22:17reprogram it. Um
  1914. 1:22:19>> rewrite our immune system. Yeah.
  1915. 1:22:21Exactly. um in in in many possible ways
  1916. 1:22:24because it's kind of a messed up if you
  1917. 1:22:27if you think about it like you know
  1918. 1:22:28biology we think is a miracle but it's a
  1919. 1:22:31it's a bad kind of a a legacy
  1920. 1:22:33engineering right it's not it's not a
  1921. 1:22:35bad engineering it's a legacy because
  1922. 1:22:37biological system finds something it
  1923. 1:22:39can't get rid of it can't start from
  1924. 1:22:41clean slate so it builds on top of it so
  1925. 1:22:43you get regulation over regulation over
  1926. 1:22:45regulation and then of course you know
  1927. 1:22:47with like immune system that I study you
  1928. 1:22:49know you get lots of autoimmune disease
  1929. 1:22:51is his immune system kills a lot of
  1930. 1:22:53people you know even during like
  1931. 1:22:55pandemics and things like that or it
  1932. 1:22:57doesn't recognize the cancer cell and
  1933. 1:22:59things like that. So why you know we
  1934. 1:23:01should be able to design like immune
  1935. 1:23:03system 2.0 like clean slate really
  1936. 1:23:06greatly engineered the immune system.
  1937. 1:23:08Well, and I said you know by 2045 50
  1938. 1:23:11we'll get to that point. Um and and
  1939. 1:23:14actually, you know, again, at the time
  1940. 1:23:16it sounded really crazy to people. Uh
  1941. 1:23:19but now I feel that I was I was too
  1942. 1:23:21conservative. We'll probably get there.
  1943. 1:23:23Uh but but the key point is that I wrote
  1944. 1:23:26specifically in the about we will do
  1945. 1:23:29this because of artificial intelligence.
  1946. 1:23:31You know, I was just taking the plot
  1947. 1:23:33that Rey uh plotted. You know, I said,
  1948. 1:23:36okay, by 2029, AI is going to be at that
  1949. 1:23:40point. it will be good enough to apply
  1950. 1:23:42to the biology and that will allow us to
  1951. 1:23:45solve diseases and then and then the
  1952. 1:23:46aging the fact that you know the timing
  1953. 1:23:50was was pretty good uh uh again even
  1954. 1:23:53even a bit conservative uh uh I feel
  1955. 1:23:56great about it that's why you know I I'm
  1956. 1:23:59all in on AI like wow um that it's
  1957. 1:24:02happening it's really happening
  1958. 1:24:04>> so aging
  1959. 1:24:07is very complex and you know as you know
  1960. 1:24:09it's not one process. We've got these 12
  1961. 1:24:13hallmarks of biology. We now have 12.
  1962. 1:24:16Genomic instability, mitochondrial
  1963. 1:24:18dysfunction, you know, cellular
  1964. 1:24:20scinessence, on and on. We've got
  1965. 1:24:21there's 12 of them.
  1966. 1:24:22>> And we know organs are are aging at
  1967. 1:24:25different rates. They're they reach
  1968. 1:24:27their peak at different rates and they
  1969. 1:24:28age at different rates and everything is
  1970. 1:24:31interacting in a very complex way.
  1971. 1:24:33[gasps]
  1972. 1:24:34What do you see as the bottleneck
  1973. 1:24:39for understanding the aging process and
  1974. 1:24:42also reversing it?
  1975. 1:24:44>> Um I mean more so than the bottleneck uh
  1976. 1:24:49this is the way we have to think of
  1977. 1:24:51aging. Uh biology actually um uh is
  1978. 1:24:57programmed to prevent aging. Right? So,
  1979. 1:25:00it's not like um uh it's not like a car
  1980. 1:25:04in a way because once you make a car um
  1981. 1:25:08you have to constantly bring it to a
  1982. 1:25:10repair shop or you have to repaint it.
  1983. 1:25:13Biology does that internally. If it
  1984. 1:25:16didn't, we would age immediately. Like
  1985. 1:25:17there is a disease called progeria.
  1986. 1:25:19These children get aged uh by the age of
  1987. 1:25:2378 they become like a 89 year old
  1988. 1:25:25because of single point mutation in one
  1989. 1:25:27of their one of their genes because they
  1990. 1:25:29lose their ability to repair um whether
  1991. 1:25:31it's the DNA repair whether it's getting
  1992. 1:25:34reg rid of the old cells or cleaning up
  1993. 1:25:37the tissues and then regenerating like
  1994. 1:25:39stem cells creating new cells. So this
  1995. 1:25:42program continues for for sometimes
  1996. 1:25:45decades otherwise we we wouldn't survive
  1997. 1:25:48for some animals for some organisms is
  1998. 1:25:51only a couple of years for for us is
  1999. 1:25:53about you know maybe 50undred years uh
  2000. 1:25:56for some veils is hundreds of years so
  2001. 1:25:58so you know same biology it's just that
  2002. 1:26:01one of them decided that you know I can
  2003. 1:26:03keep a veil um or you know whatever some
  2004. 1:26:07animals um you know older longer because
  2005. 1:26:11they don't they're not getting hunted or
  2006. 1:26:13they can reproduce later and so on. So
  2007. 1:26:15what happens in in in the in the
  2008. 1:26:17biological system is that somehow uh
  2009. 1:26:20this program breaks down and you start
  2010. 1:26:22to lose what's called the res
  2011. 1:26:24resilience, right? So when you are uh
  2012. 1:26:28age 30 or 40, you're a you're resilient.
  2013. 1:26:32you can tolerate much more damage than
  2014. 1:26:36someone who's 70 years old, 80 years old
  2015. 1:26:39because your your your systems are uh
  2016. 1:26:42you know even if you get wounded or if
  2017. 1:26:44you uh get sick you can recover uh
  2018. 1:26:47easier. Um uh but that that sort that
  2019. 1:26:50resilience is lost and that the reason
  2020. 1:26:53why it's low that there is a sort of an
  2021. 1:26:55information loss because the biological
  2022. 1:26:59system has a certain information that it
  2023. 1:27:02knows when certain genes should be
  2024. 1:27:04turned on when things should be
  2025. 1:27:07regenerated when it needs to be like
  2026. 1:27:09your skin. You know why you get
  2027. 1:27:12wrinkles? because your cells stop making
  2028. 1:27:14collagen and then all kinds of crap
  2029. 1:27:16accumulates under your skin and then you
  2030. 1:27:18know the guy the guys who like the
  2031. 1:27:20macrofasages or whatever was supposed to
  2032. 1:27:22clean there they don't do their job.
  2033. 1:27:24There's some sort of a breakdown in
  2034. 1:27:25information or communication or you know
  2035. 1:27:28intracellular communication is one of
  2036. 1:27:30the hallmarks of of aging. And then of
  2037. 1:27:32course why that happens is is that 12
  2038. 1:27:36hallmarks is is is the reason many
  2039. 1:27:39reasons you know uh for example the
  2040. 1:27:41bacteria in your gut is is is a reason.
  2041. 1:27:44So so these bacteria produce all kinds
  2042. 1:27:46of metabolites that help your immune
  2043. 1:27:49system to constantly regenerate keep it
  2044. 1:27:52in optimal shape. If that changes then
  2045. 1:27:55you know your metabolism is changing
  2046. 1:27:58your glucose levels your mitochondrial
  2047. 1:28:00uh mutations and so on so forth. So all
  2048. 1:28:03of these things accumulate you know
  2049. 1:28:05epigenetic changes and DNA mut mutations
  2050. 1:28:08and somehow the the biology forgets well
  2051. 1:28:11what am I supposed to do like how how am
  2052. 1:28:13I dealing with that also becau because
  2053. 1:28:18when a damage happens it's harder to fix
  2054. 1:28:21a damage than prevent it right so if if
  2055. 1:28:24you're continuously taking care of your
  2056. 1:28:26car or your house the likelihood of it
  2057. 1:28:29you know breaking down is much less than
  2058. 1:28:32If you wait until like okay nothing
  2059. 1:28:35works yes you can reverse it but it's
  2060. 1:28:38going to take a lot more effort and so I
  2061. 1:28:42think uh what will happen is that for a
  2062. 1:28:45younger uh individuals in the next
  2063. 1:28:48decade or so uh there uh
  2064. 1:28:53for them it's not just it's not going to
  2065. 1:28:55be reversal it's going to be prevention
  2066. 1:28:58of the aging process it's going to be
  2067. 1:29:00maintaining
  2068. 1:29:02that process the resilience decades
  2069. 1:29:04more. So we will come to a point where
  2070. 1:29:07if you are 20 30 whatever years old you
  2071. 1:29:10won't age anymore because it's going to
  2072. 1:29:12be constant reversal. But people who
  2073. 1:29:15have already aged you let's say you're
  2074. 1:29:1780 years old 90 years old then we're
  2075. 1:29:20going to have to reverse that process.
  2076. 1:29:22That's that's a more difficult we'll be
  2077. 1:29:24able to do it. Definitely we'll be able
  2078. 1:29:26to do it. Um uh uh but um uh it will
  2079. 1:29:30require a lots of engineering approaches
  2080. 1:29:33because you need to fix most of those
  2081. 1:29:36hallmarks. If you're younger, you
  2082. 1:29:39prevent those hallmarks from happening.
  2083. 1:29:42You maintain the the information uh uh
  2084. 1:29:45much much longer. Um both of those uh
  2085. 1:29:48will will will happen. um uh uh we we we
  2086. 1:29:51just need to figure out what that
  2087. 1:29:53information is being lost and we put it
  2088. 1:29:56back.
  2089. 1:29:56>> Do you think so? Let's first talk about
  2090. 1:29:58preventing the aging if you're a younger
  2091. 1:30:00person because it's easier to to do
  2092. 1:30:03always prevent if if you have a person
  2093. 1:30:09you know who's 20 or 30 years old. Do
  2094. 1:30:11you think that the approach would be
  2095. 1:30:16finding first of all do we even know all
  2096. 1:30:18the repair processes that are we we have
  2097. 1:30:20discover we have what we know right
  2098. 1:30:23>> but we still have a lot to discover
  2099. 1:30:25>> we have a we probably have a lot to
  2100. 1:30:26discover and so like do you think
  2101. 1:30:28there's going to be a a discovery where
  2102. 1:30:31we figure out like you know we know
  2103. 1:30:33things like autophagy stem cell
  2104. 1:30:35depletion you know all these stress
  2105. 1:30:36response genes like antioxidant like all
  2106. 1:30:39these things DNA repair mitochondrial
  2107. 1:30:41the way mitochondrial repair itself,
  2108. 1:30:42right? Um, are we going to be enhancing
  2109. 1:30:45or like tuning these up so that they
  2110. 1:30:47keep working at their prime continually
  2111. 1:30:50or do you think we're going to have
  2112. 1:30:51again this like information where we why
  2113. 1:30:54why are those things going down? Are we
  2114. 1:30:56going to just then go to the information
  2115. 1:30:59of it, the epigenetics perhaps? Um, and
  2116. 1:31:02is it going to be more targeted towards
  2117. 1:31:05those genes or are we going to have more
  2118. 1:31:07of this? you know, we'll get into this
  2119. 1:31:09cellular reprogramming and partial
  2120. 1:31:11reprogramming, but um [snorts] I'm I'm
  2121. 1:31:13curious like how you see AI coming into
  2122. 1:31:15that process. Like I guess we don't know
  2123. 1:31:18that's the part of the problem, but then
  2124. 1:31:20we have to figure out how to give these
  2125. 1:31:21del you know treatments to people,
  2126. 1:31:24right? That's another part of the
  2127. 1:31:26equation.
  2128. 1:31:27Um, so I mean I think you know the the
  2129. 1:31:29ones that you mentioned about sort of
  2130. 1:31:30the lifestyle changes and they of course
  2131. 1:31:34help a lot but they only slow down the
  2132. 1:31:37aging process. There's I don't think
  2133. 1:31:39there's anything that reverses that
  2134. 1:31:41process. There might be some sort of
  2135. 1:31:43local reversal for a temporary period of
  2136. 1:31:46time. Uh maybe but it's still kind of
  2137. 1:31:49trying to you know uh hope that things
  2138. 1:31:53won't go bad a little bit longer. Like
  2139. 1:31:56for example, some people can live to to
  2140. 1:31:58to 100, others only to 60, right? So
  2141. 1:32:01there's something good about those who
  2142. 1:32:04live to and in fact there are super
  2143. 1:32:05centinarians who couldn't make it to 110
  2144. 1:32:07years old. Very very few people, but I
  2145. 1:32:10think it's mostly genetics. I mean their
  2146. 1:32:12lifestyle might have helped a little
  2147. 1:32:13bit. Uh something about their biology is
  2148. 1:32:16able to maintain that information much
  2149. 1:32:19uh much longer that program. So we have
  2150. 1:32:21to get to the core. what what are the
  2151. 1:32:24things that are disrupting that
  2152. 1:32:25information um uh loss? Um and um yeah,
  2153. 1:32:30it's of course you you have to focus on
  2154. 1:32:32the on the genome because that's that's
  2155. 1:32:34sort of the blueprint. It's not just
  2156. 1:32:36that. It's sort of what affects you
  2157. 1:32:39afterwards, you know, that your your
  2158. 1:32:41microbiome, your um metabolites, you
  2159. 1:32:44know, how those things are changing,
  2160. 1:32:46whether accelerating or uh reversing,
  2161. 1:32:49you know, like and it has to be kind of
  2162. 1:32:51an engineering approach as well, like
  2163. 1:32:53you know, the skin aging is is a very
  2164. 1:32:56different problem than immune aging,
  2165. 1:32:58than the brain aging, right? So, uh your
  2166. 1:33:01skin cells are constantly renewing. So
  2167. 1:33:03all you have to do is to have sort of
  2168. 1:33:06the uh programmed stem cells to go in
  2169. 1:33:09there clean the environment sen cells
  2170. 1:33:12and get it get it regenerated and
  2171. 1:33:14produce collagen whatnot but the brain
  2172. 1:33:16is not like that right so you don't you
  2173. 1:33:17don't want to regenerate your your
  2174. 1:33:19neurons uh you will lose your identity
  2175. 1:33:21so they have to be dealt in a different
  2176. 1:33:23different way some of it will be I think
  2177. 1:33:26for the younger population u it seems
  2178. 1:33:30like you know redesigning certain
  2179. 1:33:32biology ology would be sounds radical
  2180. 1:33:35but it would be uh more foolproof right
  2181. 1:33:38so what if we could change the genome
  2182. 1:33:42through genetic engineering like we add
  2183. 1:33:45certain genes or we change certain genes
  2184. 1:33:47such that the DNA damage um is checked
  2185. 1:33:51you know much much longer it you know
  2186. 1:33:54because there are in fact certain
  2187. 1:33:55animals who have better DNA damage
  2188. 1:33:58proteins they kind of evolve to do that
  2189. 1:34:01like elephants rarely get cancer, right?
  2190. 1:34:04Because they have this gene called P-53.
  2191. 1:34:07They have multiple copies of that. P3 is
  2192. 1:34:09kind of like the guardian of the genome.
  2193. 1:34:11You know, it prevents the genome from
  2194. 1:34:13getting too much mutations and prevents
  2195. 1:34:15cancer. So, somehow elephants have pre I
  2196. 1:34:19don't know how many copies, but they get
  2197. 1:34:20very rarely cancer. Um, naked mole rats,
  2198. 1:34:24you probably know that very well. Um,
  2199. 1:34:27you know, they they're they're like
  2200. 1:34:28rats. They live underground but normal
  2201. 1:34:31rats live a couple of years and these
  2202. 1:34:32guys live 30 40 years. So it turns out
  2203. 1:34:35they have some mutation in some immune
  2204. 1:34:38gene called seag gas that's also
  2205. 1:34:40involved in immune optimization and DNA
  2206. 1:34:44repair just like you know one or two
  2207. 1:34:46genes make a huge difference. So can we
  2208. 1:34:50uh engineer humans to uh block that
  2209. 1:34:54degradation of of information uh uh for
  2210. 1:34:58for those who have already had the
  2211. 1:35:00damage then we're going to have to to
  2212. 1:35:02think about repairing that reversing it
  2213. 1:35:06and then maintaining it. Uh that's
  2214. 1:35:08that's going to be a bit more
  2215. 1:35:09challenging but uh uh we'll we'll get to
  2216. 1:35:12that too. [snorts]
  2217. 1:35:13>> What do you think about so the gene
  2218. 1:35:15going to gene therapy? There's obviously
  2219. 1:35:17gene editing, gene therapy, and and um
  2220. 1:35:21right now we only know what we know,
  2221. 1:35:23right? Again, like with these longevity
  2222. 1:35:25genes we know about, but do you think
  2223. 1:35:28that that AI is going to be able to help
  2224. 1:35:30us analyze the human genome?
  2225. 1:35:34And I don't know what other data sets it
  2226. 1:35:36will need but we'll give it everything
  2227. 1:35:37and help us figure out well actually
  2228. 1:35:40there's interaction of these genes
  2229. 1:35:42together and when there you know like
  2230. 1:35:43all these combinations is that something
  2231. 1:35:45that you think is going to happen? We'll
  2232. 1:35:47actually figure out there's a lot more
  2233. 1:35:49to this equation than we originally
  2234. 1:35:51knew.
  2235. 1:35:52>> Yeah. That that's the critical problem
  2236. 1:35:55because we know what all the genes are
  2237. 1:35:57in the genome. like we have we have it
  2238. 1:36:00decoded completely and then we pretty
  2239. 1:36:02much know their functions most of them
  2240. 1:36:05uh even if you don't know every single
  2241. 1:36:07gene involved in aging we know a lot of
  2242. 1:36:09them the problem is that different gene
  2243. 1:36:13uh uh first of all can create different
  2244. 1:36:15proteins you know there's all that
  2245. 1:36:17splicing that happens and and so on but
  2246. 1:36:20but even we doubt that in a different
  2247. 1:36:22context so if you the same protein uh
  2248. 1:36:26can kill a cell or causes survival like
  2249. 1:36:29in immune system we have these receptors
  2250. 1:36:32called TNF receptors or whatever they
  2251. 1:36:34can they can have a survival signal or
  2252. 1:36:36or a death signal suicide signal
  2253. 1:36:38depending on the context of the of the
  2254. 1:36:40cell. So that is very very uh critical
  2255. 1:36:43that how as you pointed out how these
  2256. 1:36:46genes and proteins uh in a network
  2257. 1:36:49fashion in a sort of a topological
  2258. 1:36:51network uh
  2259. 1:36:54you know what do they do like if I
  2260. 1:36:57interfere like these uh um probably
  2261. 1:37:00we'll talk about that these things
  2262. 1:37:02called Yamanaka factors where you can
  2263. 1:37:04you can generate a stem cell from a
  2264. 1:37:06normal cell right so like complete
  2265. 1:37:08regeneration uh uh but But the the
  2266. 1:37:11problem is that they can also cause
  2267. 1:37:14cancer because they only need to be
  2268. 1:37:17active in certain time. If they're
  2269. 1:37:18active all the time, they can cause
  2270. 1:37:20terteratomomas and things like that. So
  2271. 1:37:22that that part is so complex that we
  2272. 1:37:26absolutely going to need AI to simulate
  2273. 1:37:29that for us. If I have this gene in the
  2274. 1:37:33context of all the other things at
  2275. 1:37:36certain age with these epigenetic
  2276. 1:37:39programs plus all the metabolites and so
  2277. 1:37:42on because those are constantly
  2278. 1:37:44signaling the cell and you know doing
  2279. 1:37:47letting the the proteins do something
  2280. 1:37:49and so on. What would happen if I
  2281. 1:37:51interfere with that particular gene or
  2282. 1:37:54how can I improve that? Uh if if you
  2283. 1:37:57have a because you have to consider the
  2284. 1:37:59other genome too like your gene therapy
  2285. 1:38:01might be very different than somebody
  2286. 1:38:02else's because you might have some great
  2287. 1:38:06genes that are synergistic with that
  2288. 1:38:08other person might have not so great
  2289. 1:38:11genes if even if you try to improve it
  2290. 1:38:13that that would actually work or it
  2291. 1:38:15wouldn't it wouldn't help. Um so uh it's
  2292. 1:38:19just a matter of complexity. there's so
  2293. 1:38:21much information that uh the AI has to
  2294. 1:38:25not only put that together but have sort
  2295. 1:38:28of almost a temporal simulation of the
  2296. 1:38:31model like that's a very important point
  2297. 1:38:33the because right now the models are
  2298. 1:38:36kind of static they they have a good
  2299. 1:38:39understanding but they don't know what
  2300. 1:38:41would happen if a cell comes next to a
  2301. 1:38:44tumor just 2 minutes earlier they the
  2302. 1:38:48cell next to it what that context
  2303. 1:38:51affects there's a behavioral issue. It's
  2304. 1:38:55the same problem with the robotics,
  2305. 1:38:56right? So, um kind of the physical
  2306. 1:38:59intelligence or the biological
  2307. 1:39:00intelligence once those models are
  2308. 1:39:03evolved with with a lot of data. I think
  2309. 1:39:06we'll we will be able to simulate this
  2310. 1:39:08and and AI will will be able to decide
  2311. 1:39:11this is the gene therapy you should get.
  2312. 1:39:13So, you need a new copy of immune
  2313. 1:39:15system, but let me design it for you.
  2314. 1:39:17It's it's so exciting because not only
  2315. 1:39:19are we talking about, you know,
  2316. 1:39:21extending our lifespan and curing
  2317. 1:39:23disease, but we're talking about like
  2318. 1:39:26getting rid of side effects in a way. I
  2319. 1:39:28mean, you know, people all respond to
  2320. 1:39:30different foods and treatments and
  2321. 1:39:33everything differently, right? That's
  2322. 1:39:34why some people have a terrible response
  2323. 1:39:36to perhaps maybe a vaccine
  2324. 1:39:38>> um and others don't. And so, it's really
  2325. 1:39:40exciting to think about that. Um,
  2326. 1:39:42>> which which I, by the way, call human
  2327. 1:39:442.0. And maybe we'll get to human 3 3.0
  2328. 1:39:48uh which which will happen at this bios
  2329. 1:39:50singularity moment. What that means is
  2330. 1:39:52that you know we we kind of re-engineer
  2331. 1:39:54ourselves. Um uh I always think about
  2332. 1:40:00like most most scientists or most
  2333. 1:40:02doctors think like what's wrong with
  2334. 1:40:04this person or patient. Uh I always
  2335. 1:40:06think the opposite. There are certain
  2336. 1:40:08people I'm saying what's right about
  2337. 1:40:11them? like this person is has smoked for
  2338. 1:40:1450 years, never got a lung cancer or you
  2339. 1:40:18know had a terrible diet or whatever.
  2340. 1:40:20This one lived to be 110 for you know
  2341. 1:40:23whatever reason. And so what is good
  2342. 1:40:26about those people? Why can't we take
  2343. 1:40:29what's good about all of those people
  2344. 1:40:31and then re-engineer those that are not
  2345. 1:40:34so lucky to be born with what's what's
  2346. 1:40:36so good and then you know even make it
  2347. 1:40:39better. So that's the human 2.0,
  2348. 1:40:41>> right? I I I mean that's exciting to me
  2349. 1:40:43as well, right? I mean we do know like
  2350. 1:40:45you said we can live humans are capable
  2351. 1:40:48right now of living to be is the whole I
  2352. 1:40:50think the oldest was like 121 maybe
  2353. 1:40:53>> 123
  2354. 1:40:55French woman. I mean
  2355. 1:40:57>> the fact that that right now in 2026 we
  2356. 1:41:00know that humans can at least live to be
  2357. 1:41:02123
  2358. 1:41:03>> is exciting. 115 I mean at 115 116
  2359. 1:41:07that's considered sort of the current
  2360. 1:41:09limit but you know only 300 people in
  2361. 1:41:12the world are 110 and older why is that
  2362. 1:41:16why not the rest of the 8 billion
  2363. 1:41:18>> right yeah it's it's fascinating and I'm
  2364. 1:41:21I'm so excited for you know having this
  2365. 1:41:24super computing power to help us figure
  2366. 1:41:26that out what did you think when
  2367. 1:41:30you know the Yamanaka factors were
  2368. 1:41:32discovered by Shina Yamanaka and all of
  2369. 1:41:34a sudden you could take this old cell
  2370. 1:41:36and completely rever reverse it to
  2371. 1:41:38revert it to an you know essentially
  2372. 1:41:41induced you know pur potent stem cell.
  2373. 1:41:43Do you remember like is that was that
  2374. 1:41:44something did aging come into your mind
  2375. 1:41:46at that point where you were thinking
  2376. 1:41:47well that's the youngest almost you
  2377. 1:41:49could get I mean
  2378. 1:41:51>> yeah uh of course uh in fact at the time
  2379. 1:41:54I was um part of some aging groups uh uh
  2380. 1:41:58I think like an hour after the paper was
  2381. 1:42:01published I was you know typing there
  2382. 1:42:04you know like this is this is it this is
  2383. 1:42:06amazing so I I should say that there
  2384. 1:42:08were two um moments for me uh uh that
  2385. 1:42:12that I thought that aging was was going
  2386. 1:42:16to be uh reversible or curable, however
  2387. 1:42:18you call it. Um kind of like the
  2388. 1:42:20chachipit moment of biology. The first
  2389. 1:42:23moment was uh the um the sheep uh that's
  2390. 1:42:27called Dolly. Uh you probably know it
  2391. 1:42:29was the first cloned ship sheep. Um it
  2392. 1:42:34was 19967
  2393. 1:42:36or something like that. I can't remember
  2394. 1:42:37the exact date, but it was in '90s. And
  2395. 1:42:40um so basically uh the um uh the
  2396. 1:42:45scientists took a cell from uh you know
  2397. 1:42:48uh from one sheep and then recreate
  2398. 1:42:51exact copy of that sheep you know by by
  2399. 1:42:53cloning it. Uh it was it was at the
  2400. 1:42:56embryo level but it was sort of like
  2401. 1:42:58exact copy of it. So that means that
  2402. 1:43:01there was enough information that you
  2403. 1:43:04could just like uh recreate the same
  2404. 1:43:07person again and again and again. Right?
  2405. 1:43:09And then the second of course uh uh the
  2406. 1:43:11the Yamanaka factors uh in 2016 I think.
  2407. 1:43:15Um and that was the moment that uh that
  2408. 1:43:18we knew um that we could completely
  2409. 1:43:23erase the um sort of the age of the cell
  2410. 1:43:27on a cellular level and then bring it
  2411. 1:43:30back to a purotinent stem cell level and
  2412. 1:43:32then use that to recreate the whole
  2413. 1:43:35biological organism. So, so it means
  2414. 1:43:37that we have unlimited supply of
  2415. 1:43:40regenerative capacity like it's there is
  2416. 1:43:43there's there's no limit to it. In fact,
  2417. 1:43:45we already know that like so our our DNA
  2418. 1:43:47just keeps for billions of years. It
  2419. 1:43:50keeps going on and the fact that you
  2420. 1:43:52could do that in the lab and you could
  2421. 1:43:54you could generate it was was was
  2422. 1:43:57amazing. Um uh but of course the the
  2423. 1:44:00problem was okay so then how do you
  2424. 1:44:03apply that? In fact, I I think there was
  2425. 1:44:06just a recent study that started in
  2426. 1:44:08Japan using the Yamanakica factors uh uh
  2427. 1:44:12in in clinical trials because you know
  2428. 1:44:15it was not a very controlled system like
  2429. 1:44:17you didn't know if those cells would
  2430. 1:44:19develop tumors you know in mice they
  2431. 1:44:22they did some of them tumor tumors you
  2432. 1:44:24know whether um you can control them or
  2433. 1:44:28importantly I think there's going to be
  2434. 1:44:30a trial started by David Sinclair soon
  2435. 1:44:33can we do like partial reprogramming
  2436. 1:44:36because most of the time you don't want
  2437. 1:44:38the plur potent cell all right you just
  2438. 1:44:40want your skin cells to go early enough
  2439. 1:44:44to their sort of more stem-like level
  2440. 1:44:46like I work in immune system and for us
  2441. 1:44:49um I can divide like immune cells into
  2442. 1:44:52naive memory and aector and
  2443. 1:44:54differentiated so the naive cells are
  2444. 1:44:57kind of the young guys they have huge
  2445. 1:44:59potential to expand and and make memory
  2446. 1:45:03and and affect the population and the
  2447. 1:45:05other ones constantly um die and get
  2448. 1:45:08older. Can we actually revert the cells
  2449. 1:45:11towards the naive? And I I actually
  2450. 1:45:13spent a long time trying to do that. Um
  2451. 1:45:15so maybe this partial programming will
  2452. 1:45:18will will will enable that and and
  2453. 1:45:20that's that will be uh revolutionary
  2454. 1:45:23because uh then you can if you can also
  2455. 1:45:26deliver those then you can make most of
  2456. 1:45:29your old skin cells turn into a younger
  2457. 1:45:32version. I think the trial is going to
  2458. 1:45:34be for I uh with David Sinclair. Um
  2459. 1:45:37yeah. So uh but again it's it's um the
  2460. 1:45:42these these things showed us that uh we
  2461. 1:45:46can reverse aging. But when people say
  2462. 1:45:49oh that's impossible like this is this
  2463. 1:45:51you can't you can't reverse aging like
  2464. 1:45:53you know this entropy whatever. Um but
  2465. 1:45:56we we we do it in the lab all the time.
  2466. 1:45:59Uh why not do it in a in a total
  2467. 1:46:01organism level? So with this partial
  2468. 1:46:03cellular reprogramming as um as you
  2469. 1:46:06mentioned you know you're you're
  2470. 1:46:08basically taking an old cell and putting
  2471. 1:46:10these four different proteins I think
  2472. 1:46:12they can do it with fewer now but
  2473. 1:46:14putting them on for a shorter period of
  2474. 1:46:16time on the cell and that it's changing
  2475. 1:46:18the the epigenetic program and in a way
  2476. 1:46:21that it's still the cell keeps its
  2477. 1:46:23identity. It doesn't become a stem cell
  2478. 1:46:25but it seems to be more youthful. Um, I
  2479. 1:46:28know there's been some work and I
  2480. 1:46:29haven't followed all this literature
  2481. 1:46:31since I the first, you know, some of the
  2482. 1:46:34first studies that came out, but I think
  2483. 1:46:36it was like Juan Carlos, um, Epizusa, he
  2484. 1:46:40he's now, I think, at Altos Labs, but he
  2485. 1:46:41at the time was at the Sulk Institute
  2486. 1:46:43>> and, um, he had done this in in mice. I
  2487. 1:46:46think they were even maybe perhaps
  2488. 1:46:47progeria mice or some sort of
  2489. 1:46:49accelerated aging model
  2490. 1:46:51>> and there was some reversal of you know
  2491. 1:46:53certain organs seemed to be rejuvenated
  2492. 1:46:56in a sense and um the the life
  2493. 1:46:58expectancy was extended in those animals
  2494. 1:47:02but what's interesting is that not all
  2495. 1:47:05of the 12 hallmarks of aging go away.
  2496. 1:47:09>> Yeah.
  2497. 1:47:09>> Right. And so you would hope that you
  2498. 1:47:12would reverse aging totally but there's
  2499. 1:47:14genomic you know somatic mutations are
  2500. 1:47:17still there I think tieumir don't get
  2501. 1:47:20mitochondria so
  2502. 1:47:21>> do you think first of all I don't I I'd
  2503. 1:47:24love to understand why that is so what
  2504. 1:47:27is it if you're if you're essentially
  2505. 1:47:29you know wiping out the epigenetic
  2506. 1:47:32current epigenetic program and and
  2507. 1:47:33reverting it back um why does not
  2508. 1:47:36everything change I don't know if you
  2509. 1:47:38have any ideas But do you think AI is
  2510. 1:47:40going to help us understand that?
  2511. 1:47:43>> Uh definitely. I mean we I should also
  2512. 1:47:46point out that we um we do need to
  2513. 1:47:49generate lots of data. So so I think um
  2514. 1:47:53you know when whenever I talk about AI
  2515. 1:47:55um people say okay well why can't AI do
  2516. 1:47:58it now? Um for two reasons. One is that
  2517. 1:48:01we don't have enough data. So we we we
  2518. 1:48:04probably know maybe 10 20% of all the
  2519. 1:48:06biology. we still have lots of data to
  2520. 1:48:09to generate. The second is the
  2521. 1:48:11>> you're talking about scientists.
  2522. 1:48:12>> Yeah. Scientists or or automated lab
  2523. 1:48:15whatever it is. Um so I mean right now
  2524. 1:48:18we're able to generate millions of data
  2525. 1:48:20points in one experiment you know and
  2526. 1:48:22but but even that's not enough like we
  2527. 1:48:24need to generate billions of data points
  2528. 1:48:26and so on. So but of course to handle
  2529. 1:48:29that we also need um super intelligence
  2530. 1:48:32and supercompute. So, we have to have
  2531. 1:48:35compute that's thousands of times than
  2532. 1:48:37what's available. And people say, okay,
  2533. 1:48:39well, you know, why are they building
  2534. 1:48:40all these data centers? Isn't this
  2535. 1:48:42enough? And so on. Well, we're going to
  2536. 1:48:44need it. If you want if you want to cure
  2537. 1:48:46all diseases and reverse aging, we're
  2538. 1:48:48going to need probably we're going to
  2539. 1:48:50need data centers in the space and and
  2540. 1:48:52and and lot more because so much data
  2541. 1:48:56has to be in real time sort of uh uh
  2542. 1:48:59simulated. um uh and we might get much
  2543. 1:49:02more efficient doing that as we learned
  2544. 1:49:04algorithms. So so that's that's one
  2545. 1:49:06issue. The other is that um as you
  2546. 1:49:09pointed out something very important I
  2547. 1:49:11mean this partial reprogramming or total
  2548. 1:49:13reprogramming they're super exciting but
  2549. 1:49:15they don't solve um they don't
  2550. 1:49:17completely solve the the aging problem.
  2551. 1:49:20They will um make your um eyes see
  2552. 1:49:24better for a certain period if you're 80
  2553. 1:49:28years old or your skin gets better. Um
  2554. 1:49:32but will it work on your um your heart
  2555. 1:49:35muscle uh or on your brain cells neurons
  2556. 1:49:40which is the critical point because if
  2557. 1:49:42you can have a perfect body but if your
  2558. 1:49:44brain is aging then then that's it. Um
  2559. 1:49:47so will it modify the sort of the
  2560. 1:49:51microbiome that has now the environment
  2561. 1:49:54of an old person because if if that
  2562. 1:49:57happens if your metabolism is in old
  2563. 1:50:00person's metabolism and and microbiome
  2564. 1:50:02is old person's metabolism and your your
  2565. 1:50:04DNA has accumulated a bunch of mutations
  2566. 1:50:06and mitochondria has bor mutations you
  2567. 1:50:09can reverse that a bit have some
  2568. 1:50:14regenerative capacity but they will
  2569. 1:50:16quickly
  2570. 1:50:17become old again, right? You know,
  2571. 1:50:19because the environment is not is not
  2572. 1:50:21great, right? So, like if you live in a
  2573. 1:50:23bad neighborhood and you created this
  2574. 1:50:26beautiful house, you know, it's but it's
  2575. 1:50:28very bad neighborhood, your house is not
  2576. 1:50:30going to last very long there. So, your
  2577. 1:50:32your neighbors has to be clean as well.
  2578. 1:50:35So, I think it's it's a great thing and
  2579. 1:50:37that's probably going to add certain uh
  2580. 1:50:40years to lifespan and the quality of
  2581. 1:50:42life uh for sure. uh but we we have to
  2582. 1:50:45push that much much further um and then
  2583. 1:50:49really understand whether it's 12
  2584. 1:50:51hallmarks actually I asked JPT recently
  2585. 1:50:54came up with another four or five
  2586. 1:50:55hallmarks
  2587. 1:50:56>> what were they
  2588. 1:50:56>> I I can't remember exactly uh it was one
  2589. 1:50:59of them was related to immune system I
  2590. 1:51:01just this was recently um uh but yeah it
  2591. 1:51:06was it was it was quite interesting um
  2592. 1:51:09trying to remember one had to do with
  2593. 1:51:10metabolism
  2594. 1:51:12um uh uh you know because we we we kind
  2595. 1:51:15of classify hallmarks based on what we
  2596. 1:51:18can measure and see and I think AI can
  2597. 1:51:20see a little bit more than we can. So
  2598. 1:51:23anyway um this is going to be uh a
  2599. 1:51:27serious engineering uh problem. I I
  2600. 1:51:30would be very surprised if we have like
  2601. 1:51:33one pill you take and then you suddenly
  2602. 1:51:36become young again. That's that seems
  2603. 1:51:38very unrealistic to me.
  2604. 1:51:40>> Yeah. I mean [laughter] you know and
  2605. 1:51:41then the other question is like in the
  2606. 1:51:42lab we're we're the way we're delivering
  2607. 1:51:45these treatments is like an adino virus
  2608. 1:51:47right and and then it's like well is
  2609. 1:51:49that going to cause cancer because they
  2610. 1:51:52virus go to right cell
  2611. 1:51:53>> is it going to go to the right cell
  2612. 1:51:54exactly I mean the there's definitely a
  2613. 1:51:56lot of engineering
  2614. 1:51:57>> we we have to develop uh so one of the
  2615. 1:51:59things that I like doing with AI models
  2616. 1:52:01is to develop some new methods new new
  2617. 1:52:04technologies they have a bit too much
  2618. 1:52:06guard rail so uh they don't allow me to
  2619. 1:52:09to go too deep in it but you know
  2620. 1:52:11because uh I don't think we have we have
  2621. 1:52:13enough tools like of course we have
  2622. 1:52:15crisper now but actually Dudana's lab
  2623. 1:52:19just came out with something even better
  2624. 1:52:20for bacteria for genome editing so
  2625. 1:52:23imagine there's there's there's probably
  2626. 1:52:24all kinds of other tools that we can
  2627. 1:52:26build that will make this localization
  2628. 1:52:29the editing much more perfect and and
  2629. 1:52:33has to be programmable you have to
  2630. 1:52:35literally create circuits we can program
  2631. 1:52:38immune cells in in in culture like we
  2632. 1:52:40can give a a drug it will shut down
  2633. 1:52:42their response or we can create end or
  2634. 1:52:45gates and not gates if they see two
  2635. 1:52:47molecules then they respond if they see
  2636. 1:52:48one they don't like you can literally
  2637. 1:52:50program the biology so we have to
  2638. 1:52:53develop these new tools that are better
  2639. 1:52:55than viruses maybe uh generate lots of
  2640. 1:52:58data sets um and they manipulate the the
  2641. 1:53:02organs and so on could be that for some
  2642. 1:53:05organs when they're too old it might
  2643. 1:53:08might be just too difficult to repair
  2644. 1:53:11them. So you might consider just putting
  2645. 1:53:13a new one,
  2646. 1:53:14>> you know, like it might be a point of no
  2647. 1:53:16return, your your kidneys or whatever.
  2648. 1:53:19Then you'll have these u u organ
  2649. 1:53:22factories which 3D printed and actually
  2650. 1:53:26>> uh we we did a lot of collaboration with
  2651. 1:53:28a colleague of mine, you know, he can
  2652. 1:53:30print, you know, small tissues, lungs
  2653. 1:53:32and and pieces like that. So some of
  2654. 1:53:35them will be kind of transplanting new
  2655. 1:53:37organs. Some of them will be
  2656. 1:53:39pre-engineering and
  2657. 1:53:40>> and then the digital twin the analysis
  2658. 1:53:42and simulation will be able to figure
  2659. 1:53:44out is are you going to reject this or
  2660. 1:53:46would you need to not reject it?
  2661. 1:53:48>> That's right.
  2662. 1:53:48>> Right. Um what do you think of the new
  2663. 1:53:51data that came out using this this model
  2664. 1:53:54called GPT micro 4B
  2665. 1:53:57GPT micro 4B? um where I guess there's
  2666. 1:54:01this model that was used to figure out
  2667. 1:54:05how to make certain mutations in the
  2668. 1:54:08four different Yamanaka factors to make
  2669. 1:54:10them more effective. So they were able
  2670. 1:54:12to basically 50fold more um be more
  2671. 1:54:16effective or efficient at increasing
  2672. 1:54:18this induced pur potency.
  2673. 1:54:20>> Yeah.
  2674. 1:54:20>> How do you interpret that data?
  2675. 1:54:22>> So I I don't think that model uh is is
  2676. 1:54:26any better than what we have right now.
  2677. 1:54:28uh probably um the current models are
  2678. 1:54:30are much better. Um the I think probably
  2679. 1:54:34there might have been two two
  2680. 1:54:36differences and I don't know all the
  2681. 1:54:37details but one is that they probably
  2682. 1:54:40remove the the guard rails because
  2683. 1:54:42there's a lot of biocurity guard rails
  2684. 1:54:44in in the current models. Um if you ask
  2685. 1:54:48the same question to GPT5.5
  2686. 1:54:50it will refuse to do it. It will say oh
  2687. 1:54:52this is a biohazard like what if you
  2688. 1:54:54mutate and create a new virus or new
  2689. 1:54:56cancer whatever. So that might be one
  2690. 1:54:58reason and then the other is like if you
  2691. 1:55:00let these models think longer. So like
  2692. 1:55:03GPT5.5
  2693. 1:55:05pro and and the thinking and in model is
  2694. 1:55:08the same pre-training but pro model can
  2695. 1:55:12think two hours thinking can take two
  2696. 1:55:14minutes. Uh so the longer they can think
  2697. 1:55:17the the more they can iterate. They can
  2698. 1:55:20run these scenarios again and again and
  2699. 1:55:22again. So my speculation is that that
  2700. 1:55:25model probably ran for for a long period
  2701. 1:55:28of time. Of course you need a lot of
  2702. 1:55:29compute and a lot of tokens not a
  2703. 1:55:31problem for open AAI. um uh then you you
  2704. 1:55:35will probably come up with the solution
  2705. 1:55:37that even a a more intelligent model
  2706. 1:55:40couldn't come up in a in a shorter
  2707. 1:55:42period of time because that that
  2708. 1:55:44particular case is really running
  2709. 1:55:47experimental scenarios like okay if I do
  2710. 1:55:50this mutation what would be the
  2711. 1:55:51potential outcome like it's running all
  2712. 1:55:53the simulation oh yeah okay so so what
  2713. 1:55:56if I change that mutation to here and
  2714. 1:55:57then what if I add another mutation and
  2715. 1:55:59running the experiment again and again
  2716. 1:56:01again so you you're constantly making
  2717. 1:56:03the the solution better and better and
  2718. 1:56:06better as as you think longer. Um so uh
  2719. 1:56:10and and this will get better. So if if
  2720. 1:56:12you have much more compute, much more
  2721. 1:56:14intelligence and you say okay um GPT7 or
  2722. 1:56:19six whatever is go and think for a
  2723. 1:56:21month, you know, find the perfect
  2724. 1:56:25molecule that will bind to this receptor
  2725. 1:56:27and this will cause that. It'll it'll
  2726. 1:56:29probably figure that out. What is it? It
  2727. 1:56:32sounds like we're going to need to do a
  2728. 1:56:33lot of this type of simulation and by
  2729. 1:56:35were I mean researchers and scientists.
  2730. 1:56:38What is it going to take to remove some
  2731. 1:56:40of those guardrails in that environment
  2732. 1:56:42for researchers to be able to to make
  2733. 1:56:45these new discoveries and and what sort
  2734. 1:56:48of I guess I mean how how do we protect
  2735. 1:56:50from a a new crazy
  2736. 1:56:52>> biohazard or you know biosafety issue?
  2737. 1:56:55Well, I mean I think like OpenAI is
  2738. 1:56:57partnering uh with with um you know
  2739. 1:57:00trusted people. Uh so you have to be
  2740. 1:57:03approved by them. So I think then uh
  2741. 1:57:05whether it's a company or something like
  2742. 1:57:07that. It's the same problem with uh with
  2743. 1:57:09cyber security, right? So Anthropic has
  2744. 1:57:12this new model called mitos and they
  2745. 1:57:14decided not to release it because they
  2746. 1:57:16said it's too dangerous for cyber
  2747. 1:57:18security because this this model can
  2748. 1:57:21just crack into any can find all these
  2749. 1:57:24things that that other uh others cannot
  2750. 1:57:26see. So, in fact, even the governments
  2751. 1:57:29thought that that was important that
  2752. 1:57:31they should uh I don't know if they're
  2753. 1:57:33exaggerating if it's if it's if it's
  2754. 1:57:35true or not, but so you have to put that
  2755. 1:57:37guard rail if you release it to the
  2756. 1:57:39world because somebody can use that
  2757. 1:57:41model and then hack into your bank
  2758. 1:57:43account or somebody can use it to create
  2759. 1:57:46a a a new virus gene or something like
  2760. 1:57:49that. So I think there you know that
  2761. 1:57:51will be made individual purses or
  2762. 1:57:56institution
  2763. 1:57:57basis that these these um uh hopefully
  2764. 1:58:02these AI companies will share that
  2765. 1:58:05because they might decide not to share
  2766. 1:58:07it. Uh might say well okay why don't we
  2767. 1:58:10just develop all the drugs internally
  2768. 1:58:13and not release any of these models. um
  2769. 1:58:16uh some some might be doing that for
  2770. 1:58:19example I don't think that would be a
  2771. 1:58:21good thing because uh what you really
  2772. 1:58:23need is again as I said you need a lot
  2773. 1:58:26of data you need a lot of scientists
  2774. 1:58:29uh putting all that data into the models
  2775. 1:58:32but not only the data but their
  2776. 1:58:34experience in a way you in in the let's
  2777. 1:58:37call it the wild or or the world you're
  2778. 1:58:40you're actually training those models
  2779. 1:58:42even even if it's super intelligence
  2780. 1:58:44it's going to
  2781. 1:58:45so hungry for data that you're going to
  2782. 1:58:48have to um collaborate or release it to
  2783. 1:58:51to to others. Um also I think this will
  2784. 1:58:55be important to democratize uh health
  2785. 1:58:58care because one question everybody
  2786. 1:59:00asked okay well you know if you find the
  2787. 1:59:03the treatment for aging this is only
  2788. 1:59:05going to be available for the super rich
  2789. 1:59:07I'm never going to be able to afford it
  2790. 1:59:09or or treatment for cancer I say the
  2791. 1:59:12opposite actually thanks to AI it will
  2792. 1:59:15be super affordable because if you can
  2793. 1:59:17create a drug like in a startup let's
  2794. 1:59:20say cannot compete with a big pharmace
  2795. 1:59:22to a company they can find a drug uh
  2796. 1:59:25using AI 100 times cheaper and if you
  2797. 1:59:28can do the clinical trial using digital
  2798. 1:59:30twin that's where all the money goes
  2799. 1:59:32like you could develop a drug for a
  2800. 1:59:33couple of million dollars rather than a
  2801. 1:59:35couple of billion dollars so the cost of
  2802. 1:59:37drug development or treatment uh
  2803. 1:59:39development will be magnitudes lower and
  2804. 1:59:42that will give uh a huge number of
  2805. 1:59:44people u access to that uh but of course
  2806. 1:59:47you know AI AI has to be um shared. It's
  2807. 1:59:53it's uh I think it's it's a product of
  2808. 1:59:56all humanity and it should be the
  2809. 1:59:58possession of all humanity. That's how I
  2810. 2:00:00view it
  2811. 2:00:01>> except for going back to the the thing
  2812. 2:00:03that you mentioned at the beginning of
  2813. 2:00:04this podcast which is that you know
  2814. 2:00:06humans in the wrong hands that is the
  2815. 2:00:09problem and that's and that's that is
  2816. 2:00:11something that needs to be very taken
  2817. 2:00:13very seriously.
  2818. 2:00:14>> But but the the solution to that is also
  2819. 2:00:16AI. So right now, I mean, I hear that
  2820. 2:00:19like MTOS um basically finds all these
  2821. 2:00:23loopholes in in this cyber security uh
  2822. 2:00:26issues that people couldn't figure out
  2823. 2:00:29for decades. They didn't even know they
  2824. 2:00:31existed. So it's just patching all those
  2825. 2:00:34uh uh all these security bugs. So it
  2826. 2:00:38will create almost a perfect secure
  2827. 2:00:41systems like it will be unhackable
  2828. 2:00:44because uh MTOS is actually preventing
  2829. 2:00:47that. So to to prevent that from
  2830. 2:00:49happening you still need AI. You might
  2831. 2:00:51still have some bad actor trying to
  2832. 2:00:54develop a virus that will cause a
  2833. 2:00:57pandemic. To prevent that you also need
  2834. 2:00:59AI. So the AI should be able to predict
  2835. 2:01:02it and already create the vaccine ready.
  2836. 2:01:05will say, well, somebody might make this
  2837. 2:01:07virus, so let's let's get ready for it.
  2838. 2:01:09Um, so, uh, AI is the solution to all
  2839. 2:01:13that.
  2840. 2:01:13>> Interesting perspective. Always seems to
  2841. 2:01:15you always seem to have a positive
  2842. 2:01:16outlook. Um, I wanted to ask you another
  2843. 2:01:18question about, you know, we're talking
  2844. 2:01:20about these simulations and how we're
  2845. 2:01:22going to, you know, using AI to to
  2846. 2:01:24essentially run these clinical trials
  2847. 2:01:26cheaper because we're going to do this,
  2848. 2:01:28you know, these simulations and have,
  2849. 2:01:30you know, biomarker data and it'll just
  2850. 2:01:32be, you know, shorter and and cheaper
  2851. 2:01:34and easier. The question is always what
  2852. 2:01:38do you measure, right? What is the
  2853. 2:01:39biioarker? What are what's the end
  2854. 2:01:41point, right? And in aging, you can now
  2855. 2:01:44see I mean every a study almost a new
  2856. 2:01:47study every day coming out looking at
  2857. 2:01:49these epigenetic aging clocks. And
  2858. 2:01:52that's you know the the so as most
  2859. 2:01:57people listening to this podcast know
  2860. 2:01:58I've had Steve Horbath on um a couple of
  2861. 2:02:00times and he's sort of the pioneer in
  2862. 2:02:02these epigenetic aging clocks and
  2863. 2:02:04they've now developed over you know the
  2864. 2:02:06last decade or so and become much more
  2865. 2:02:08of a biological marker of age like your
  2866. 2:02:10biological age not just to be able to
  2867. 2:02:12predict your actual chronological age.
  2868. 2:02:15And [snorts] so, um, you'll find now
  2869. 2:02:17studies that are looking at treatments
  2870. 2:02:18and whether or not it can reverse quote
  2871. 2:02:20unquote reverse biological aging or
  2872. 2:02:22epigenetic aging,
  2873. 2:02:25but it's not clear that that's
  2874. 2:02:29necessarily,
  2875. 2:02:30you know, if that's
  2876. 2:02:33really reversing aging, right? So, how
  2877. 2:02:35what do what do you think um from your
  2878. 2:02:38perspective, what should we be looking
  2879. 2:02:40at in terms of some of these functional
  2880. 2:02:43>> outputs? Um yeah, I mean the those
  2881. 2:02:46epigenetic u markers are very useful um
  2882. 2:02:51but I don't believe that um they are
  2883. 2:02:55terribly useful as um sort of as as
  2884. 2:02:58predicting true aging. I mean there's
  2885. 2:03:01there's a very um uh significant problem
  2886. 2:03:04with with those markers. uh usually
  2887. 2:03:07they're they're done through through
  2888. 2:03:08blood analysis [clears throat] but in
  2889. 2:03:11the blood you have u like you know I
  2890. 2:03:14work with te- cells so you have these
  2891. 2:03:16cells that we call aector cells that
  2892. 2:03:19have um lots of epigenetic change
  2893. 2:03:22because they differentiate it and they
  2894. 2:03:24continue to accumulate in in old age and
  2895. 2:03:27then you have these naive cells that
  2896. 2:03:28have you know more pristine uh kind so
  2897. 2:03:31it's a combination so depending on um
  2898. 2:03:34what that combination is is going to
  2899. 2:03:37affect the output of of the um so you
  2900. 2:03:40you you can actually just look at the
  2901. 2:03:42proportion of your uh T- cell
  2902. 2:03:44differentiate T cells you'll probably
  2903. 2:03:46get the same same kind of information um
  2904. 2:03:49and it doesn't tell you like what's
  2905. 2:03:51happening in the skin or the brain or
  2906. 2:03:53the heart you know that it doesn't mean
  2907. 2:03:55that uh if if the immune cells are
  2908. 2:03:58getting younger or the young ones are
  2909. 2:04:00expanding and the old ones are dying
  2910. 2:04:02that doesn't mean that your skin is
  2911. 2:04:04getting younger or your liver is getting
  2912. 2:04:06younger. So that it has a very limited
  2913. 2:04:08use in my opinion. But we really don't
  2914. 2:04:12need that because like aging is probably
  2915. 2:04:15the easiest way to measure. We know
  2916. 2:04:19exactly what goes wrong in in in old
  2917. 2:04:22age, right? So like you can't breathe
  2918. 2:04:25that well. Your heart doesn't work that
  2919. 2:04:28well. Your muscles don't work. you can
  2920. 2:04:32only, you know, raise so much because it
  2921. 2:04:35your weakened muscles or your VO max is
  2922. 2:04:38is is lower. Um, these are all
  2923. 2:04:43phenotypic like you don't even have to
  2924. 2:04:45probably withdraw a blood just measuring
  2925. 2:04:48the ability of uh of an elderly person.
  2926. 2:04:52Uh, can they walk uh better, you know,
  2927. 2:04:56100 meters than they used to? like
  2928. 2:04:59because that's looking at the total
  2929. 2:05:01biology like you know your cells your
  2930. 2:05:03metabolism or whatever uh muscle to me
  2931. 2:05:07that's or or your cognitive abilities
  2932. 2:05:09>> but those can't be simulated I mean
  2933. 2:05:12>> the they eventually they can be right
  2934. 2:05:17now they can't they can't be simulated
  2935. 2:05:19uh um because as I mentioned the AI is
  2936. 2:05:24missing that behavioral physical
  2937. 2:05:26intelligence in the real world because
  2938. 2:05:29that's a that's most things are
  2939. 2:05:31happening in real life. uh but um I
  2940. 2:05:35think I think they can be simulated but
  2941. 2:05:37more importantly uh I think eventually
  2942. 2:05:40you have to tr whatever the AI comes out
  2943. 2:05:42with you need to try it on on the humans
  2944. 2:05:44right so uh my point is that you don't
  2945. 2:05:48have to uh do anything too fancy or wait
  2946. 2:05:53decades to see the effect if I give this
  2947. 2:05:56treatment to um I don't know 80 year old
  2948. 2:06:00and they're suddenly able to breathe
  2949. 2:06:03Well, you know, their VMX went up. Um,
  2950. 2:06:06they're sharper, they can think better,
  2951. 2:06:08uh, they can remember better. Um, you
  2952. 2:06:12you can look at their immune system and
  2953. 2:06:13we can see that the cells are we know
  2954. 2:06:16which cells are younger or worse. Or you
  2955. 2:06:18can look at their skin like, oh wow, the
  2956. 2:06:20skin is getting young. Like you see it,
  2957. 2:06:22you don't even have to do anything. Um
  2958. 2:06:25so so there are so many features
  2959. 2:06:26phenotypic features of aging that could
  2960. 2:06:29be um objectively measured actually and
  2961. 2:06:33not just subjectively you will see the
  2962. 2:06:35effect very very quickly like this
  2963. 2:06:38partial reprogramming trial they're
  2964. 2:06:40doing it's it's done for glaucoma
  2965. 2:06:42patients I I guess uh because that
  2966. 2:06:45happens in old age right so your your
  2967. 2:06:47cells are aging so I mean if these
  2968. 2:06:50people start to see it works right their
  2969. 2:06:53their cells cells got regenerated. Um
  2970. 2:06:55you don't need to look at the
  2971. 2:06:56epigenetic. Um so I think uh it will be
  2972. 2:07:00a combination of those um measurements
  2973. 2:07:03probably we will come up with and AI
  2974. 2:07:06will probably come up with this set of
  2975. 2:07:08biomarkers. I don't think we know
  2976. 2:07:10because it's going to be a set of
  2977. 2:07:12biomarkers like um you know your glucose
  2978. 2:07:15your cholesterol might be high when
  2979. 2:07:18you're 30 and it will be high or low
  2980. 2:07:20when you're 80. I mean there's not a
  2981. 2:07:22very specific marker that will tell you
  2982. 2:07:24your your age for just looking at that.
  2983. 2:07:26But the combinatorial effect will will
  2984. 2:07:30AI probably will be able to predict your
  2985. 2:07:32age looking at all kinds of data sets
  2986. 2:07:35and say oh this guy must be uh you know
  2987. 2:07:38um 52 years old based on this. You know
  2988. 2:07:40>> I know we have uh that model clock base
  2989. 2:07:42that's looking now at a variety of small
  2990. 2:07:46molecules that might reverse epigenetic
  2991. 2:07:48aging.
  2992. 2:07:49Now there are some data sets showing
  2993. 2:07:51that you if you reverse epigenetic aging
  2994. 2:07:54there is some functional correlation
  2995. 2:07:56with some functional improvements like
  2996. 2:07:58pre-frailty things like that you know
  2997. 2:08:00like improve but um it at the end of the
  2998. 2:08:03day you know I think it it'll be
  2999. 2:08:05interesting to see if there's going to
  3000. 2:08:06be companies that come out trying to
  3001. 2:08:08sell some sort of drug to claiming it
  3002. 2:08:11reverses aging when they're really just
  3003. 2:08:13looking at one
  3004. 2:08:14>> biioarker which is reversing
  3005. 2:08:16>> it's most as I say it's mostly the
  3006. 2:08:18immune aging that they're looking at or
  3007. 2:08:20or sort of maybe getting rid of the
  3008. 2:08:23terminally differentiated immune cells
  3009. 2:08:26like for example in old age you you
  3010. 2:08:29accumulate these CME specific tea cells
  3011. 2:08:32um CMV is a virus that you can't really
  3012. 2:08:35get rid of so the immune system
  3013. 2:08:37constantly have to keep it under check
  3014. 2:08:40um and those immune cells they kind of
  3015. 2:08:43become like missionaries they should
  3016. 2:08:44retire but they keep on expanding and
  3017. 2:08:47some indiv individuals might have like
  3018. 2:08:4920 30% of all their tea cells just
  3019. 2:08:51dedicated to like one peptide of this
  3020. 2:08:54this CMV and they're they're not helpful
  3021. 2:08:57but they become uh harmful because those
  3022. 2:09:00guys are are old they should retire they
  3023. 2:09:03don't and they cause inflammation
  3024. 2:09:05because they're they're active and um
  3025. 2:09:07and they don't give place for the young
  3026. 2:09:09guys to come in uh and they're they are
  3027. 2:09:11epigenetically you know closed because
  3028. 2:09:14um they're differentiated their
  3029. 2:09:15telomeres are shorter So, uh, you know,
  3030. 2:09:19you might be getting rid of some of
  3031. 2:09:20those cells with certain treatments,
  3032. 2:09:22which is great. Um, but then you have
  3033. 2:09:25the indirect effects, right? So, if you
  3034. 2:09:26can if you can control the immune system
  3035. 2:09:28and inflammation, that's going to have
  3036. 2:09:31huge effect all over your that doesn't
  3037. 2:09:33mean your skin got just regenerated, but
  3038. 2:09:36it it will it will help clean up
  3039. 2:09:39>> aging. Yeah. Yeah. Exactly. Um, also the
  3040. 2:09:43other thing I was thinking about is
  3041. 2:09:46like, you know, we you're mentioning V2
  3042. 2:09:47max and, you know, muscle strength,
  3043. 2:09:49muscle mass. We have all these markers
  3044. 2:09:51that sort of like decrease with age and
  3045. 2:09:54yet we don't know necessarily that they
  3046. 2:09:57cause aging in a way. So the question is
  3047. 2:10:00like will AI be able to take all this
  3048. 2:10:03correlational data like we have all this
  3049. 2:10:05you know all these different functional
  3050. 2:10:07out you know endpoints that we look at
  3051. 2:10:09and and be able to differentiate it from
  3052. 2:10:11like personalized you know this
  3053. 2:10:14personalized um data set versus like
  3054. 2:10:17actually like how do you cure aging like
  3055. 2:10:20what do you change that's going to drive
  3056. 2:10:22you know [clears throat] reverse the
  3057. 2:10:24aging I mean there's there's a lot of
  3058. 2:10:26questions Um, you mentioned something
  3059. 2:10:29interesting that had to do with the
  3060. 2:10:31brain and that is something that I've
  3061. 2:10:33been thinking about as well because you
  3062. 2:10:36know we we have a lot of repair
  3063. 2:10:38processes in our body right we can
  3064. 2:10:40repair a lot of DNA damage and you know
  3065. 2:10:42mitochondrial function and you know all
  3066. 2:10:45these things but in the brain we can
  3067. 2:10:47grow new cells replace the old cells in
  3068. 2:10:49the brain it's not as robust right
  3069. 2:10:52there's some parts of the brain that can
  3070. 2:10:54um you can grow new neurons neurogenesis
  3071. 2:10:56there's neurop plasticity. That's a big
  3072. 2:10:58part of of the repair process in a way,
  3073. 2:11:01but it's not like a big you you're not
  3074. 2:11:05you're not totally replacing the brain
  3075. 2:11:07and you don't want to, you know, as you
  3076. 2:11:09mentioned because then memories go away
  3077. 2:11:11and your identity and you know, it gets
  3078. 2:11:13very complicated.
  3079. 2:11:15>> Um, how do you see AI
  3080. 2:11:18>> intervening in that? Like everything's
  3081. 2:11:20great if we can reverse our heart aging
  3082. 2:11:21and all this, but our brains that's so
  3083. 2:11:23important
  3084. 2:11:24>> now. Is it just going to be a, you know,
  3085. 2:11:26delay age related disease,
  3086. 2:11:29neuroinflammation, all that stuff? We
  3087. 2:11:30can we can fix that, but like are we
  3088. 2:11:32going to be able to really
  3089. 2:11:35reverse brain aging?
  3090. 2:11:37>> Um, you know, I I would have to ask AI
  3091. 2:11:40to to to figure that out. But, you know,
  3092. 2:11:42I can I can think of several scenarios
  3093. 2:11:45how that might happen. uh first of all
  3094. 2:11:47you know neurons um or the brain overall
  3095. 2:11:51must have some very good maintenance
  3096. 2:11:54policy right so so there are neurons
  3097. 2:11:57that live for decades maybe 70 80 years
  3098. 2:12:00and not just neurons but there are other
  3099. 2:12:02cell types that can live for very long
  3100. 2:12:04they don't divide very much there is
  3101. 2:12:06some regeneration uh it's not like zero
  3102. 2:12:09and that's very important because that
  3103. 2:12:12means that if you let's just do a total
  3104. 2:12:15experiment let's just say that you
  3105. 2:12:17replace 0.01% of your neurons uh every
  3106. 2:12:21month or every year something like that.
  3107. 2:12:23I don't think that's going to make a
  3108. 2:12:25huge difference in your brain structure
  3109. 2:12:28because what they're doing is that
  3110. 2:12:29they're probably you know there's some
  3111. 2:12:31neurons somewhere interacting with bunch
  3112. 2:12:33of other neurons synapses and then it
  3113. 2:12:37gets replaced and the new neurons might
  3114. 2:12:39have a few other synapses other than
  3115. 2:12:42that but that's going to replace that
  3116. 2:12:44network anyway because they have that
  3117. 2:12:46capability. So if you do this slowly u I
  3118. 2:12:49think um
  3119. 2:12:51you you won't lose a lot. In fact, we we
  3120. 2:12:54still lose memories, right? So, uh we uh
  3121. 2:12:57we can't remember everything or we
  3122. 2:12:59hallucinate all the time. Uh talk about
  3123. 2:13:01hallucination, right? Uh imagine that
  3124. 2:13:03this happened to me. No, no, no, it
  3125. 2:13:05didn't happen. No, no, I I I remember
  3126. 2:13:07that. So, that's like brain uh brain uh
  3127. 2:13:11maybe part of it is new neurons that
  3128. 2:13:13they just didn't know. So, they just
  3129. 2:13:14made it up, right? So, um so that's one
  3130. 2:13:17thing. The other thing is that uh these
  3131. 2:13:19neurons probably have some internal
  3132. 2:13:21abilities to regenerate. What I mean by
  3133. 2:13:23that is that you know the cell can
  3134. 2:13:25maintain itself if it has you know sort
  3135. 2:13:28of um a great way to clean up internally
  3136. 2:13:32like autofagy is is a very important
  3137. 2:13:34mechanism as you know um or it has some
  3138. 2:13:38really special DNA damage correction
  3139. 2:13:41ability uh like stem cells have that
  3140. 2:13:43right so pristine stem cells they don't
  3141. 2:13:46get old you know even in 100 years old
  3142. 2:13:48they they're still like like a a young
  3143. 2:13:51person so uh And then you have all these
  3144. 2:13:54other cells like GA cells and and and so
  3145. 2:13:57on that are there to prevent all the
  3146. 2:14:01other stuff that happens the
  3147. 2:14:03inflammation you know GA cells of course
  3148. 2:14:05are are are part of the immune system in
  3149. 2:14:07a way but they they are like the immune
  3150. 2:14:10system is not allowed into the brain in
  3151. 2:14:12very rare cases uh uh it's like a
  3152. 2:14:15protected area um because the immune
  3153. 2:14:18system causes too much damage and if you
  3154. 2:14:20can't replace it quickly that's that's a
  3155. 2:14:22huge problem, but they have their own
  3156. 2:14:24network of cleaning up and they probably
  3157. 2:14:26have some sort of like a lymphatic
  3158. 2:14:28system and and so on. Um um so if we can
  3159. 2:14:33figure that out or if I can figure that
  3160. 2:14:35out, we might be able to really maybe
  3161. 2:14:38not completely regenerate but extend it
  3162. 2:14:42um quite significantly. Maybe another 10
  3163. 2:14:4610 years, 20 years, 30 years for
  3164. 2:14:48whatever. And then we might come to a
  3165. 2:14:50point and this goes into a little bit of
  3166. 2:14:52a science fiction now you know uh let's
  3167. 2:14:54say in 50 years time AI might be able to
  3168. 2:14:58figure out all of the synaptic
  3169. 2:15:00connections in your brain like every
  3170. 2:15:02single neural network and the
  3171. 2:15:05neurotransmitters and everything else.
  3172. 2:15:07So eventually you might be able to like
  3173. 2:15:09literally simulate your your brain um
  3174. 2:15:12you go into the matrix level. So that
  3175. 2:15:15might allow AI to like say okay I'm
  3176. 2:15:18going to replace all these neurons but
  3177. 2:15:20I'm going to make sure that they
  3178. 2:15:22reconnect all these signapses
  3179. 2:15:26>> so so that you don't lose your identity.
  3180. 2:15:28Um or alternately I can keep a copy here
  3181. 2:15:32and then we can create a new brain and
  3182. 2:15:34then transfer to that new brain that
  3183. 2:15:37exact u uh state uh that I that I found.
  3184. 2:15:41Uh I I'm not saying that this is
  3185. 2:15:44possible right now. That's really
  3186. 2:15:45science fiction error, but you can
  3187. 2:15:47imagine that at some point we might get
  3188. 2:15:50to that level. So I'm not I'm not too
  3189. 2:15:52worried. I I think if it can pass this
  3190. 2:15:55couple of decades and then keep the
  3191. 2:15:57brain um healthy and and self-preserving
  3192. 2:16:00for for maybe uh you know age 120, 130.
  3193. 2:16:05And in fact, you know, the people people
  3194. 2:16:07actually who live to to age 100, they
  3195. 2:16:10they have very sharp minds, right?
  3196. 2:16:11>> Because if you don't have sharp mind,
  3197. 2:16:12you don't live very old. So that's like
  3198. 2:16:15super correlated. So [snorts] if we can
  3199. 2:16:18keep it for a couple of more decades and
  3200. 2:16:20uh we'll probably find some other
  3201. 2:16:21solutions. So if we can if we can keep
  3202. 2:16:23the neuroinflammation low, if we can
  3203. 2:16:26increase brain drive, neurotrophic
  3204. 2:16:28factors, some of these things that we
  3205. 2:16:30know does play a role in improving
  3206. 2:16:31neuroplasticity and
  3207. 2:16:33>> you know and growing new neurons and to
  3208. 2:16:35do all the things that we can at least
  3209. 2:16:37in some predictable way
  3210. 2:16:39>> and we can have like chips for the for
  3211. 2:16:41the memory part, you know, we could
  3212. 2:16:43always supplement that. So
  3213. 2:16:45>> increase the capacity
  3214. 2:16:46>> and and and hopefully um AI will help us
  3215. 2:16:49figure out how to deliver these
  3216. 2:16:50therapies to the brain.
  3217. 2:16:51>> [laughter]
  3218. 2:16:52>> Yeah, delivery is always the biggest
  3219. 2:16:54problem,
  3220. 2:16:54>> right? Well, this has been such a
  3221. 2:16:56fascinating and exciting conversation.
  3222. 2:16:58Uh, Duria, I have a couple of more
  3223. 2:17:00questions, closing questions for you.
  3224. 2:17:03And I really kind of was just wanting to
  3225. 2:17:04know
  3226. 2:17:06if you had access, let's say there was
  3227. 2:17:08no guard rail rails and you had access
  3228. 2:17:11to
  3229. 2:17:13all this data in aging biology, you
  3230. 2:17:15know, the T- cell, you know, all the T-
  3231. 2:17:18cell repertoire, long, you know,
  3232. 2:17:20longitudinal uh longitudinal
  3233. 2:17:22longitudinal cohorts, um, centinarian
  3234. 2:17:26data, like everything, just anything you
  3235. 2:17:28can imagine. You had it all and you had
  3236. 2:17:31this model that was amazing that you
  3237. 2:17:33could
  3238. 2:17:33>> You're describing heaven for me.
  3239. 2:17:35>> Yes. Yes. [laughter]
  3240. 2:17:36What what what would be the the the
  3241. 2:17:38prompt? What would be the question you
  3242. 2:17:39would you would ask it? I mean, there'd
  3243. 2:17:41be more than one, but what would be the
  3244. 2:17:43first?
  3245. 2:17:43>> Yeah. Hoping that the AI won't answer uh
  3246. 2:17:4642 as an answer. Um the so so so uh the
  3247. 2:17:50the first thing I would probably ask is
  3248. 2:17:53um not not saying that just go figure
  3249. 2:17:57out aging or whatever because I think
  3250. 2:17:59there has to be there there has to be
  3251. 2:18:01certain sequence. So imagine that you
  3252. 2:18:04have all this data. Uh what would be the
  3253. 2:18:09the most practical
  3254. 2:18:11um uh quickest way you can develop uh an
  3255. 2:18:15intervention to an elderly person say
  3256. 2:18:18age 70 80 years old that will uh
  3257. 2:18:22immediately add five years to their
  3258. 2:18:24lifespan. So to me that would be uh the
  3259. 2:18:28most critical immediate question to ask
  3260. 2:18:31because do that population doesn't have
  3261. 2:18:34a lot of time and so we have to develop
  3262. 2:18:37these these technologies extremely
  3263. 2:18:40quickly and will should have you know
  3264. 2:18:42even two years three years extend so
  3265. 2:18:45that I can come up with the next prompt
  3266. 2:18:47uh after that. Um so I guess that that
  3267. 2:18:51that would be the the the first prompt I
  3268. 2:18:54would ask. That's great. What um Okay,
  3269. 2:18:57there's another question. So, this this
  3270. 2:18:58one is
  3271. 2:19:00there's no there's no money. Money is no
  3272. 2:19:02object. Okay. There's no like you have
  3273. 2:19:05complete like access.
  3274. 2:19:07>> You're describing so many heavens now.
  3275. 2:19:09>> I know. I'm just I'm curious what what
  3276. 2:19:11your your answer is. You're going to
  3277. 2:19:14personally build your own digital twin,
  3278. 2:19:17>> which which I plan to
  3279. 2:19:18>> right now. Um [laughter]
  3280. 2:19:21what test would you prioritize? like
  3281. 2:19:23what data sets would you prioritize, how
  3282. 2:19:26can a person get them, um how often
  3283. 2:19:29would you take these tests, how would
  3284. 2:19:31you organize this information into the
  3285. 2:19:32AI to really get the biggest bang, you
  3286. 2:19:36know, benefit from the information it's
  3287. 2:19:38going to give you,
  3288. 2:19:39>> right? But you said money is not
  3289. 2:19:40>> money is not an issue, right? Money's
  3290. 2:19:42not an issue.
  3291. 2:19:43>> Um so, so I would divide it into two
  3292. 2:19:45parts. Uh one part is that we have to um
  3293. 2:19:50um so what I would do is set up a a huge
  3294. 2:19:53lab um you know partially automated lab
  3295. 2:19:58where I would generate enormous amount
  3296. 2:20:01of data on the um on the cells on the
  3297. 2:20:04tissues in the in the lab because uh we
  3298. 2:20:08have to go by the first principles to
  3299. 2:20:10understand what's going on on let's say
  3300. 2:20:13in an individual T- cell uh all these
  3301. 2:20:16thousands of proteins, metabolites, what
  3302. 2:20:18are they doing? Then then that will
  3303. 2:20:20enable me to create what's called the
  3304. 2:20:23virtual cells um and then eventually
  3305. 2:20:25virtual tissues and you know how cells
  3306. 2:20:27are in a special temporal manner are are
  3307. 2:20:30are behaving and so on. So that would be
  3308. 2:20:33that probably be the most expensive part
  3309. 2:20:35of it and uh I'll need a lot of money.
  3310. 2:20:37You said no limit, right? So okay. Um uh
  3311. 2:20:41but the second part would be sort of
  3312. 2:20:43what we talked earlier uh kind of the
  3313. 2:20:46behavioral data from from the human
  3314. 2:20:49humans and that data is not just of
  3315. 2:20:52course you know all kinds of you know
  3316. 2:20:55plasma [snorts] levels of proteins
  3317. 2:20:57metabolize your full microbiome your
  3318. 2:21:00full genome sequencing and all of these
  3319. 2:21:02things are are possible by the way I
  3320. 2:21:04mean it's you know if the cost is not an
  3321. 2:21:06issue you can easily like UK bio Bio
  3322. 2:21:09bank has done it for 500,000 people. You
  3323. 2:21:11can do it for a million people. Uh and I
  3324. 2:21:14think if you did it in a million people
  3325. 2:21:15that would pretty much cover all the
  3326. 2:21:17possible human humanity. I mean I it's
  3327. 2:21:20not like everybody's perfectly uh
  3328. 2:21:23different. You know we share a lot of
  3329. 2:21:24things and and and and so you know from
  3330. 2:21:28the humans collect lots of biological
  3331. 2:21:31data but very importantly behavioral
  3332. 2:21:34data. I think this this is something
  3333. 2:21:36that's totally missing in a digital
  3334. 2:21:38twin. Um like you know we were talking
  3335. 2:21:40earlier uh ability of someone to walk
  3336. 2:21:44certain distance, ability to to you know
  3337. 2:21:48raise some some weights. These don't
  3338. 2:21:50show up in any biomarker sets but they
  3339. 2:21:53could be extremely important. um uh or
  3340. 2:21:57ability to think, you know, their their
  3341. 2:22:00cognitive level that that could be
  3342. 2:22:03directly uh brain uh brain aging related
  3343. 2:22:06and and I mean you lots of things and
  3344. 2:22:08and you know what happens when humans
  3345. 2:22:11are in certain environments, you know,
  3346. 2:22:14in certain environments you even if you
  3347. 2:22:16if you are having a very sort of healthy
  3348. 2:22:20lifestyle that may not help you much.
  3349. 2:22:23for example, you know, I lived in New
  3350. 2:22:25York City for a decade. Uh, you know, my
  3351. 2:22:28stress level was so high uh uh uh and
  3352. 2:22:32that stress level is so harmful for you
  3353. 2:22:35because the immune system is constantly
  3354. 2:22:36thinking there's a threat out there and
  3355. 2:22:38it's causing a lot of inflammation. In
  3356. 2:22:39fact, I think people who live in New
  3357. 2:22:41York has twice as much heart attack risk
  3358. 2:22:43or something like that. you know that
  3359. 2:22:45that your environment, your um uh your
  3360. 2:22:49emotional states and how you interact
  3361. 2:22:51with other people. All of these things
  3362. 2:22:54will impact your aging process, your
  3363. 2:22:57your resilience to the life, your
  3364. 2:22:59optimistic level. By the way, being
  3365. 2:23:01optimistic is one of the best things you
  3366. 2:23:03can do for for aging and study after
  3367. 2:23:06study show that. So being able to absorb
  3368. 2:23:10um bad things that happen to you and
  3369. 2:23:13then keep keep going. So resilience. So
  3370. 2:23:16but these are behavioral data that's not
  3371. 2:23:18available in in the biological set. So
  3372. 2:23:21uh yeah uh I would do that for for a
  3373. 2:23:24million people all over the world
  3374. 2:23:25different parts. Um and then on the lab
  3375. 2:23:28uh every single cell type that I can
  3376. 2:23:30find uh decode those uh put them all
  3377. 2:23:33together to the super intelligence and
  3378. 2:23:35then voila we have digital twin.
  3379. 2:23:38>> Okay Doria. So let's say someone wants
  3380. 2:23:40to build their little mini digital twin
  3381. 2:23:43right now using the models we have
  3382. 2:23:45access to today. The type of data that
  3383. 2:23:47we can aggregate you know at the
  3384. 2:23:49consumer level today biometric data that
  3385. 2:23:52we can that we can put in. um how would
  3386. 2:23:54you build that mini digital twin today?
  3387. 2:23:59>> Yeah, great question. I mean uh in fact
  3388. 2:24:01it is possible to build a sort of a mini
  3389. 2:24:04digital twin uh that doesn't have to be
  3390. 2:24:07as sophisticated as I described because
  3391. 2:24:09that that one is more uh sort of
  3392. 2:24:11clinical trials and developing
  3393. 2:24:13treatments. Uh but you know going back
  3394. 2:24:16to the u example of the UK bio bank you
  3395. 2:24:19know they didn't have trillions of data
  3396. 2:24:21sets. they only used a few hundred data
  3397. 2:24:24points from from each person and they
  3398. 2:24:26were able to predict a lot of diseases.
  3399. 2:24:28So that means that you know we we can
  3400. 2:24:29have a lot of predictive power with the
  3401. 2:24:32data that we're collecting uh today. Um
  3402. 2:24:35you know another example is this uh
  3403. 2:24:37glucose u meter that I have um you know
  3404. 2:24:41every five minutes it shows my glucose
  3405. 2:24:43level and then I take that data and of
  3406. 2:24:45course I put it to chat GPT um and once
  3407. 2:24:48you uh additional
  3408. 2:24:51data set that becomes very uh very
  3409. 2:24:54valuable because uh let's say that you
  3410. 2:24:56have your lab values your cholesterol
  3411. 2:24:59your glucose um your uh every day the
  3412. 2:25:03the steps that you took and your sleep
  3413. 2:25:06uh and so on. So these are actually very
  3414. 2:25:08rich data on their own because they're
  3415. 2:25:11uh their accumulation of lots of under
  3416. 2:25:14uh uh underbiology that that results in
  3417. 2:25:17that but also that puts uh AI into a
  3418. 2:25:21context your mini digital uh twin. So my
  3419. 2:25:24my suggestion would be uh to uh u you
  3420. 2:25:27know provide the AI as much data as they
  3421. 2:25:31can and on a daily basis so that so and
  3422. 2:25:35keep it in the same context so same
  3423. 2:25:38window so they so the model can remember
  3424. 2:25:41that um actually there are there are
  3425. 2:25:43some tricks uh uh to do that as well.
  3426. 2:25:45You can keep it as like a database and
  3427. 2:25:48tell li model go check my database and
  3428. 2:25:51see what my new uh you know based on my
  3429. 2:25:53new data how things have changed what
  3430. 2:25:56suggestion you could give. Um I for
  3431. 2:25:58example uh provide all the supplements
  3432. 2:26:01that I take you know um you know the
  3433. 2:26:04type of food that I eat um all of these
  3434. 2:26:06things will make uh will make a big
  3435. 2:26:09difference. Um so the model start to
  3436. 2:26:12really personalize um you know sort of
  3437. 2:26:15the uh style. It will know your style
  3438. 2:26:18and will make uh suggestions for you. Uh
  3439. 2:26:22rather than giving blanket statement
  3440. 2:26:24like you should walk 10,000 steps. Well
  3441. 2:26:26you know knows that like Daria cannot
  3442. 2:26:29walk 10,000 steps every day but I think
  3443. 2:26:323,000 would be enough for him.
  3444. 2:26:34>> And what kind of model are we talking
  3445. 2:26:35about? Would you be using the GPT5.5
  3446. 2:26:38Pro? And then what about you know these
  3447. 2:26:40agents and codecs and how does that come
  3448. 2:26:42into helping analyze that that database
  3449. 2:26:44that you're creating?
  3450. 2:26:46>> Yeah, I I think you know these models
  3451. 2:26:48are becoming more agentic all the time.
  3452. 2:26:50I I know OpenAI for example they
  3453. 2:26:53integrated agents into um their codeex
  3454. 2:26:56model the coding model and soon I'm sure
  3455. 2:26:58it will be part of all of chat GPT. Um
  3456. 2:27:01you don't you don't I don't think you
  3457. 2:27:03need very sophisticated models for that.
  3458. 2:27:06What is important is that uh really
  3459. 2:27:09maintaining that context. Uh so
  3460. 2:27:11hopefully the models will have a larger
  3461. 2:27:14memory and they can remember. So you
  3462. 2:27:17chat can keep certain memories about you
  3463. 2:27:20but it's still kind of limited. Uh um
  3464. 2:27:23it's not just CHP like you can use JNI
  3465. 2:27:25for example which has a longer um uh
  3466. 2:27:28context windows um or or cloud for that
  3467. 2:27:30matter. I think most of the models can
  3468. 2:27:32handle that uh information. And they
  3469. 2:27:34don't have problem dealing with large
  3470. 2:27:37data sets. As I mentioned, I can put
  3471. 2:27:39millions of data sets and they're able
  3472. 2:27:41to analyze that. What they need is that
  3473. 2:27:44they need to remember how things were a
  3474. 2:27:47month ago
  3475. 2:27:49because that's before and after. Before
  3476. 2:27:51and after is extremely valuable. So the
  3477. 2:27:55model will know
  3478. 2:27:57he started taking vitamin D3. Oh, these
  3479. 2:28:01things changed after that that you may
  3480. 2:28:03not notice or is glucose looks better
  3481. 2:28:06because of you know when that's that
  3482. 2:28:09change happened. So it's starts to make
  3483. 2:28:12those lengths and and that's I think the
  3484. 2:28:14critical point because you need all of
  3485. 2:28:17that context in the in the AI model to
  3486. 2:28:19to give you sort of a better uh
  3487. 2:28:22prediction on what to use and what not
  3488. 2:28:24to use. Okay, you were using that well
  3489. 2:28:27maybe that was not a great idea so
  3490. 2:28:28change it. um or change the dolls or or
  3491. 2:28:31whatnot.
  3492. 2:28:32>> Yeah, that's interesting. It kind of
  3493. 2:28:34reminded me of a question that I did
  3494. 2:28:36want to ask you about, you know, these
  3495. 2:28:38AI models and future AI advances
  3496. 2:28:42when you think about these qualities.
  3497. 2:28:43So, like persistent memory, expanded
  3498. 2:28:45context handling, it seems like those
  3499. 2:28:49seem to be more important.
  3500. 2:28:51>> Absolutely. that I I think um for me uh
  3501. 2:28:55memory which which brings the context uh
  3502. 2:28:58so the models are now able to think for
  3503. 2:29:02quite long time and they don't because
  3504. 2:29:04previously the models would just um even
  3505. 2:29:07in the same context window if you had a
  3506. 2:29:09million context windows uh after a while
  3507. 2:29:12they would just fall off because they
  3508. 2:29:14would forget even what they were
  3509. 2:29:16thinking about. Now they have this
  3510. 2:29:18ability to constantly um go and check on
  3511. 2:29:21it. So uh I think in the next few months
  3512. 2:29:26this is going to happen. Uh so that that
  3513. 2:29:28will that will have a tremendous impact.
  3514. 2:29:30Well that's memory is everything.
  3515. 2:29:32>> So so you think so how long are we
  3516. 2:29:34talking like let's say you know you
  3517. 2:29:37started a vitamin D supplement 6 months
  3518. 2:29:40ago. Put that you have the same window
  3519. 2:29:42and you start in that window you have
  3520. 2:29:44that you know entry point that the date
  3521. 2:29:46and then you keep adding about you know
  3522. 2:29:48you add your your your data in. has got
  3523. 2:29:50all the data um right now. Can it go
  3524. 2:29:54back that far or how far can it go back?
  3525. 2:29:57And
  3526. 2:29:58>> um if you have that data somewhere in
  3527. 2:30:00your database uh for example um I
  3528. 2:30:03adapted a a technique that uh Karpathi
  3529. 2:30:07who's a famous AI researcher described.
  3530. 2:30:10Uh so you can um turn you can create
  3531. 2:30:13your own Viki sort of Wikipedia kind of
  3532. 2:30:16a thing like personal um you take uh you
  3533. 2:30:19know uh if you have all your data
  3534. 2:30:21somewhere uh you can ask AI just pull
  3535. 2:30:24all that and put it into a Wikipedia
  3536. 2:30:26like you know you can do it daily or
  3537. 2:30:28weekly depending on the environment
  3538. 2:30:30whatever um and so now you're building
  3539. 2:30:33your own database health database uh
  3540. 2:30:36which AI can help you update it if you
  3541. 2:30:39have that data
  3542. 2:30:40it can go years doesn't matter like you
  3543. 2:30:43can have 10 years of data it will
  3544. 2:30:45analyze all of that um uh but
  3545. 2:30:48>> it has that memory it can like
  3546. 2:30:50>> yeah so in in in in the same context if
  3547. 2:30:53you provide all of that I mean it's
  3548. 2:30:56still limited with you know maybe a
  3549. 2:30:58million tokens or something but no one's
  3550. 2:31:00going to have million token data set
  3551. 2:31:02even if you if you calculate 10 years so
  3552. 2:31:05so that's that's not that's not a
  3553. 2:31:07problem the problem is like if if you
  3554. 2:31:09have
  3555. 2:31:10If you want this to be continuous like
  3556. 2:31:12you just give AI okay here's the data
  3557. 2:31:16today that it should be able to remember
  3558. 2:31:18what was yesterday what was 2 months ago
  3559. 2:31:21so you don't have to give you know all
  3560. 2:31:23of the um uh you don't have to keep your
  3561. 2:31:25own database and and give all that again
  3562. 2:31:28and again because you have to do that
  3563. 2:31:30every time right so the your whole and
  3564. 2:31:32that will spend a lot of tokens and
  3565. 2:31:34stuff like that so uh but but I think
  3566. 2:31:36this is this is going to be uh this is
  3567. 2:31:38going to be sold
  3568. 2:31:39How do you not bias? How do you lower
  3569. 2:31:43the the ability of yourself to bias what
  3570. 2:31:47you know GPT 5.5 Pro is is going to feed
  3571. 2:31:51you back, right? Like based on what
  3572. 2:31:53you're asking it and I mean
  3573. 2:31:55[clears throat] I I find sometimes I I
  3574. 2:31:56might be able to bias it a little bit.
  3575. 2:31:59Do you do you do you know what I'm
  3576. 2:32:00talking about?
  3577. 2:32:01>> Yeah, sure. I [clears throat] mean
  3578. 2:32:03that's why I think uh we are in sort of
  3579. 2:32:05the experimental phase um in a way um
  3580. 2:32:09everyone has to do their own kind of
  3581. 2:32:11validation
  3582. 2:32:13um as the models are getting better.
  3583. 2:32:14What I mean by that is that again you
  3584. 2:32:17know of course don't try harmful things
  3585. 2:32:18and then you know uh don't go into risk
  3586. 2:32:21but you know for daily daily use um you
  3587. 2:32:25might be taking vitamin D
  3588. 2:32:28and then you you stop taking vitamin D
  3589. 2:32:31so that you're just doing an experiment
  3590. 2:32:32like before and after and then you
  3591. 2:32:34collect that data before and after u and
  3592. 2:32:37then AI gives you one solution says well
  3593. 2:32:41you know taking this dose of vitamin D I
  3594. 2:32:43think is important So then you can start
  3595. 2:32:45that dose again and then see see what
  3596. 2:32:48happens. If if you reach the same level
  3597. 2:32:51as before means that AI made a good
  3598. 2:32:53prediction like you need to see after
  3599. 2:32:56you have to have that record before and
  3600. 2:32:59after so that you you are the judge.
  3601. 2:33:02Well what this was a good idea so I'm
  3602. 2:33:04I'm glad that I listened to Jupy. Well
  3603. 2:33:07if it wasn't a good idea it didn't kill
  3604. 2:33:09you. It didn't make you sick. So that's
  3605. 2:33:11that's also fine.
  3606. 2:33:13>> Yeah. I guess for someone that's already
  3607. 2:33:14taking a lot of supplements for example,
  3608. 2:33:17they're not going to have that before
  3609. 2:33:18and after. Then also you have to know
  3610. 2:33:19like how long do you wait you know for
  3611. 2:33:22example to for the wash out period and
  3612. 2:33:25>> yeah and whatnot. the the hope is that
  3613. 2:33:28if you provide that very frequently um
  3614. 2:33:32in fact um I can mention one thing uh
  3615. 2:33:35for example the the lab values like you
  3616. 2:33:37go and measure your cholesterol glucose
  3617. 2:33:39sodium whatever they always give you a
  3618. 2:33:42range right so if it's within this range
  3619. 2:33:45it's normal well how do you know that uh
  3620. 2:33:49because you can be at the top of the
  3621. 2:33:51range uh that might be your abnormal
  3622. 2:33:54somebody else's normal somebody might be
  3623. 2:33:57a little bit over the normal and might
  3624. 2:33:59still be okay um or vice versa because
  3625. 2:34:03we don't know the level on a
  3626. 2:34:04personalized level. So we we calculate
  3627. 2:34:07population base. So okay so this range
  3628. 2:34:09is good for this population. So in a way
  3629. 2:34:12um if you have three or four
  3630. 2:34:15measurements let's say every few months
  3631. 2:34:18you can develop your own set point
  3632. 2:34:21normal. you know the the AI will know
  3633. 2:34:24your normal for glucose is 90 not 70 not
  3634. 2:34:30100 or not 105 somebody else might be
  3635. 2:34:34102. So it knows that based on that that
  3636. 2:34:37measurements. So then then it starts to
  3637. 2:34:40give you advice based on your data set
  3638. 2:34:43your set points because if yours is 100
  3639. 2:34:46and suddenly dropped to 70 maybe that's
  3640. 2:34:49not a good thing you know I'm just I'm
  3641. 2:34:51just uh giving an example uh uh so uh
  3642. 2:34:54that's why that continuous data
  3643. 2:34:57collection is so important uh uh with
  3644. 2:34:59with glucose meter I collected every 5
  3645. 2:35:02minutes uh the more data the better
  3646. 2:35:05>> well Duria thank you so much for sitting
  3647. 2:35:07down with me today and talking about
  3648. 2:35:09this exciting I mean frontier that we're
  3649. 2:35:13exploring
  3650. 2:35:14you know curing disease extending human
  3651. 2:35:17life expectancy obviously health span
  3652. 2:35:20reversing aging perhaps getting to human
  3653. 2:35:232.0 you know where we're enhancing
  3654. 2:35:25you know genetic you know features as
  3655. 2:35:29well um very exciting time to be in and
  3656. 2:35:33if we cannot die in the next 10 to 15
  3657. 2:35:35years
  3658. 2:35:36>> it may be even more exciting. Yes,
  3659. 2:35:38absolutely. Because uh you know the last
  3660. 2:35:40thing I will say uh
  3661. 2:35:43this is so unique in human history. Uh
  3662. 2:35:46because a decade ago uh if you set
  3663. 2:35:50someone well you should be very healthy
  3664. 2:35:53you know uh do this do that and they can
  3665. 2:35:56say well it's only going to extend my
  3666. 2:35:58life maybe two years or three years. I
  3667. 2:36:00just want to live my life and I don't
  3668. 2:36:02care about living few more years as an
  3669. 2:36:04old age. and that that was perfectly,
  3670. 2:36:07you know, relevant. That's not the case
  3671. 2:36:10now. Living an extra one year could make
  3672. 2:36:15you reach that threshold where there's
  3673. 2:36:17going to be the ability to to treat many
  3674. 2:36:21diseases and reverse your aging and give
  3675. 2:36:23you another decade, give you another 20
  3676. 2:36:25years and then once you reach that, you
  3677. 2:36:29get another 10 years, another so like
  3678. 2:36:32even every day counts now in my opinion.
  3679. 2:36:35Uh so uh that's why don't die for
  3680. 2:36:38[laughter]
  3681. 2:36:39>> um where people can find out more about
  3682. 2:36:42your research and they can follow you. I
  3683. 2:36:44follow you on X. Um maybe you can tell
  3684. 2:36:46people how to follow you, what your user
  3685. 2:36:49your Twitter follower or sorry your ex
  3686. 2:36:52user handle is and where else they can
  3687. 2:36:54find you.
  3688. 2:36:55>> Uh yeah, my my main account is an X. Uh,
  3689. 2:36:58it's at Daria D E R Y A T R under dash.
  3690. 2:37:05Um, if they write Dario Nutmas, I think
  3691. 2:37:07I'll I'll show up. Um, that's that's
  3692. 2:37:09where I, you know, do most of my
  3693. 2:37:11communication. Um, I have a LinkedIn
  3694. 2:37:13account, but I don't post that often
  3695. 2:37:15there. Um, I I've been planning to start
  3696. 2:37:19up a a sort of a YouTube channel, but
  3697. 2:37:21um, I don't think I'll ever do that
  3698. 2:37:23because I'll never have the time. you
  3699. 2:37:25know, it's it's really amazing what what
  3700. 2:37:27you're doing because
  3701. 2:37:30video takes a lot of a lot of effort. Uh
  3702. 2:37:32so for me is the the fastest way. Uh in
  3703. 2:37:35fact, I I even had a Substack uh
  3704. 2:37:37account, but just couldn't find the time
  3705. 2:37:39to write long uh uh long messages. So So
  3706. 2:37:43uh X is the best way.
  3707. 2:37:44>> Well, I really encourage people to
  3708. 2:37:46follow you on X. post. I mean, just
  3709. 2:37:48every day there's something interesting
  3710. 2:37:50that you're posting on X and so I highly
  3711. 2:37:53recommend that people do follow you
  3712. 2:37:55>> as um many already do. So, thanks again
  3713. 2:37:58for the research you're doing and for
  3714. 2:38:00I'm I'm excited to see what um what's
  3715. 2:38:03going to happen in the next couple of
  3716. 2:38:04months.
  3717. 2:38:06>> Looking forward to it. Very optimistic.
  3718. 2:38:08Thank you. Thank you very much. It was
  3719. 2:38:10great.

About this transcript

This page contains the full transcript of Why the Next 10 Years May Add 50 to Your Lifespan | Dr. Derya Unutmaz by FoundMyFitness, generated from the public captions YouTube serves with the video. The transcript has 26,886 words across 3,719 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.