YouTube2Text

Кого ИИ сделает богаче, а кого оставит без работы | Нобелевский лауреат по экономике — Transcript

by Marina Mogilko · 6,456 words · 1,133 segments · language en · Watch on YouTube

Full transcript

  1. 0:01I can easily imagine a grim future where
  2. 0:03many people will not have meaningful
  3. 0:05work.
  4. 0:06>> This is Daron Acemoglu, an economist
  5. 0:09from MIT and a 2024 Nobel Prize winner
  6. 0:11in Economics. He believes we are missing
  7. 0:14the main opportunities offered by
  8. 0:15artificial intelligence. What should be
  9. 0:17done? How can we remain valuable? I
  10. 0:19constantly see posts. Instead of hiring
  11. 0:22a person, I launched an AI agent.
  12. 0:24>> Automation is great, but we shouldn't
  13. 0:26focus solely on it.
  14. 0:28Otherwise, we will destroy too many jobs
  15. 0:30without creating new, meaningful,
  16. 0:32high-paying, productive ones that open
  17. 0:34up opportunities for people.
  18. 0:36>> So, what should people do? What skills
  19. 0:38should they develop?
  20. 0:39>> If you want to know my opinion, go and
  21. 0:42learn.
  22. 0:46>> In 2024, you said that within 10 years,
  23. 0:48AI would add about 1%to the US GDP, and
  24. 0:50only 5%of tasks would be fully automated
  25. 0:53.
  26. 0:54What do you think about that now, in
  27. 0:552026?
  28. 0:57First of all, I am very glad to be here
  29. 1:00.
  30. 1:01And I want to say right away that I do
  31. 1:03not claim to have any unique
  32. 1:05understanding of artificial intelligence
  33. 1:07.
  34. 1:08Artificial intelligence is a huge topic
  35. 1:10.
  36. 1:11Everyone looks at it from their own
  37. 1:13perspective and sees only a part of the
  38. 1:14bigger picture.
  39. 1:16I try to bring some basic economic logic
  40. 1:19into the discussion about productivity.
  41. 1:22I made those calculations more than 2
  42. 1:24years ago.
  43. 1:25There was already a lot of uncertainty
  44. 1:27back then, and today there is even more
  45. 1:30. But there have been very rapid changes
  46. 1:32, especially this year with the
  47. 1:34emergence of agentic AI.
  48. 1:36Perhaps today I would name a figure
  49. 1:38slightly higher than 5%.
  50. 1:41And yet, I still think most
  51. 1:43conversations exaggerate the speed at
  52. 1:45which we will see automation.
  53. 1:48Because automation is complicated.
  54. 1:51Integrating it into the work of
  55. 1:52organizations is difficult.
  56. 1:55The diffusion of new technologies,
  57. 1:57especially when they require large
  58. 1:59additional investments and serious
  59. 2:01changes, is a difficult and slow process
  60. 2:03.
  61. 2:04This does not mean that we are not
  62. 2:06looking at a technology capable of
  63. 2:07changing the world.
  64. 2:09And it does not mean that we cannot
  65. 2:11benefit from it. But when discussing how
  66. 2:14quickly and how deeply the economy will
  67. 2:17change, we must remain realistic.
  68. 2:20And we still need to discuss the same
  69. 2:23questions that arise with any new
  70. 2:24technology.
  71. 2:26How can it be used best? How can it be
  72. 2:30abused? And who will ultimately benefit
  73. 2:33?
  74. 2:33>> You mentioned AI agents. What would have
  75. 2:36to happen for you to say, "Now I am
  76. 2:38ready to raise my estimate of AI's
  77. 2:40impact to 10%or higher"?
  78. 2:42>> In programming, progress is indeed very
  79. 2:46rapid.
  80. 2:48Why is artificial intelligence
  81. 2:50developing faster here than in other
  82. 2:52professions?
  83. 2:54First, programming happens in an office
  84. 2:56.
  85. 2:57There is no need to interact with the
  86. 2:58physical world.
  87. 3:01The people who verify the code work are
  88. 3:03well-versed in both artificial
  89. 3:05intelligence and technology in general.
  90. 3:08But there is one more very important
  91. 3:11thing.
  92. 3:12In programming, like in mathematics or
  93. 3:14chess, there are clear rules.
  94. 3:17A program either works, or it doesn't.
  95. 3:21Reinforcement learning is especially
  96. 3:24effective when there is an unambiguous
  97. 3:26criterion that lets you say: "No, that's
  98. 3:29wrong."
  99. 3:30In many professions, there is no such
  100. 3:32criterion.
  101. 3:34Take customer support, for example.
  102. 3:37Most people probably expected that
  103. 3:39artificial intelligence would be the
  104. 3:41first to fully automate it.
  105. 3:44Something is already happening, but the
  106. 3:46process has proven to be very slow and
  107. 3:48difficult.
  108. 3:50In many companies I deal with, the
  109. 3:52artificial intelligence works just
  110. 3:53terribly.
  111. 3:55You call, you get a bot, and you think:
  112. 3:58"God, another 10 minutes of waiting
  113. 4:00until they connect me to an operator who
  114. 4:03can help, because dealing with people is
  115. 4:05what matters here.
  116. 4:07Their requests change over time and
  117. 4:09depend on the individual."
  118. 4:12People ask the same question in
  119. 4:14completely different ways.
  120. 4:16Plus a multitude of ambiguous and
  121. 4:19non-standard cases.
  122. 4:21All of this is much more complex.
  123. 4:23People must either control artificial
  124. 4:26intelligence or work alongside it.
  125. 4:29Ideally, we need to establish a
  126. 4:31cooperation where artificial
  127. 4:32intelligence complements the human.
  128. 4:36All these options are possible, and
  129. 4:38which one we choose will largely
  130. 4:40determine how many jobs disappear, how
  131. 4:43much productivity grows, and who
  132. 4:45benefits.
  133. 4:47>> In programming, the result is
  134. 4:48unambiguous and easy to check, while in
  135. 4:50support, as you said, everything is much
  136. 4:52more individualized. Can this problem be
  137. 4:54solved by a stronger model, or is there
  138. 4:56a fundamental limitation of artificial
  139. 4:58intelligence meaning it cannot work
  140. 5:00without a human?
  141. 5:02I definitely wouldn't say there's
  142. 5:04anything artificial intelligence will
  143. 5:07never be able to do.
  144. 5:12Over the last 10 years, we've seen
  145. 5:13enough examples.
  146. 5:16If you invest enough resources and data
  147. 5:18, artificial intelligence can advance in
  148. 5:20many things.
  149. 5:23The question is how quickly and in which
  150. 5:25areas.
  151. 5:27If we look broader and ask why
  152. 5:30productivity is growing so weakly and
  153. 5:33what to do about it, my answer fits in
  154. 5:36one word: application. We need more
  155. 5:40applications, and they must be much
  156. 5:42better.
  157. 5:44One could even assume, though I don't
  158. 5:46know for sure and I might be wrong,
  159. 5:48since I don't work at Anthropic or
  160. 5:50OpenAI.
  161. 5:53I don't see the results of their
  162. 5:55experiments, but they have huge
  163. 5:57resources and very talented people.
  164. 6:00But one could say that America and
  165. 6:02Silicon Valley, in particular, have
  166. 6:04focused too much on base models and too
  167. 6:06little on applications and their
  168. 6:08implementation in companies.
  169. 6:11We are constantly told that America is
  170. 6:13ahead of China in large language models
  171. 6:15. And that, by all appearances, is true
  172. 6:17. Chinese models sometimes catch up, but
  173. 6:20largely at the expense of American ones
  174. 6:23. Most likely, they are partly learning
  175. 6:25from their responses.
  176. 6:27However, Chinese companies are doing
  177. 6:29something I see much less of in the USA
  178. 6:32.
  179. 6:33They are actively integrating robots and
  180. 6:36artificial intelligence into production
  181. 6:38and employing a huge number of engineers
  182. 6:41to do it.
  183. 6:42And it seems the U.S. just doesn't have
  184. 6:44that many engineers.
  185. 6:46And all that are available are going
  186. 6:48into model development.
  187. 6:49>> You've brought up a very interesting
  188. 6:50topic.
  189. 6:52I'm an immigrant, and when I see how
  190. 6:53things are set up in the U.S. and ask
  191. 6:54why, they tell me: "Well, that's just
  192. 6:56how it’s always been."
  193. 6:58>> Yes, yes. Or that there’s simply no
  194. 7:01other way.
  195. 7:04Here is something else I would like to
  196. 7:05say.
  197. 7:07I don’t want to overgeneralize, but in
  198. 7:09my view, the American AI industry is
  199. 7:12harmed by a belief in inevitability.
  200. 7:16The conviction that our way is the only
  201. 7:19possible one.
  202. 7:20And if you have doubts, then you're an
  203. 7:23idiot.
  204. 7:26I categorically disagree with that.
  205. 7:29Artificial intelligence is a very
  206. 7:31flexible technology, much like other
  207. 7:33general-purpose technologies.
  208. 7:35That’s what economists call, for example
  209. 7:38, electricity or the printing press—
  210. 7:40technologies that serve as a foundation
  211. 7:42for many other solutions.
  212. 7:45Both history and economic research show
  213. 7:47that they can be developed in completely
  214. 7:49different ways.
  215. 7:51And these paths are not equivalent; they
  216. 7:54yield different benefits and come with
  217. 7:57different costs.
  218. 7:59Therefore, believing that our way is the
  219. 8:01only one or inherently the best is, in
  220. 8:03my opinion, a mistake.
  221. 8:05>> And I think there is another problem
  222. 8:07here. How do you measure it? How do you
  223. 8:09assess the impact of AI on your personal
  224. 8:11productivity or on a company's
  225. 8:12productivity?
  226. 8:14We spoke with Erik Brynjolfsson in May.
  227. 8:16According to their research, the
  228. 8:18economic impact of AI has been virtually
  229. 8:21zero.
  230. 8:22The capabilities of models are growing
  231. 8:25rapidly, but the economic impact remains
  232. 8:27quite modest.
  233. 8:28And this gap between capabilities and
  234. 8:31reality, in my view, opens up huge
  235. 8:33prospects.
  236. 8:34>> Do you think the problem also lies in
  237. 8:36how we measure it?
  238. 8:37>> Erik is a very serious researcher.
  239. 8:40He knows this industry perhaps better
  240. 8:42than any other social scientist, and I
  241. 8:45think he is more optimistic than I am.
  242. 8:48If you asked him about the impact of AI
  243. 8:50on productivity by 2034-2035
  244. 8:54—which is exactly what I was writing
  245. 8:56about then—he would give a bolder
  246. 8:58estimate. But on the whole, I think his
  247. 9:00conclusion is: AI hasn't yet delivered
  248. 9:03the productivity growth we expect, but
  249. 9:05it will. My position is roughly the same
  250. 9:07. It’s not visible yet, but we will see
  251. 9:10it. It’s just, in my view, it will
  252. 9:12happen a little later than Erik thinks.
  253. 9:14I will be very glad if he turns out to
  254. 9:16be right.
  255. 9:17And he has a much better chance of being
  256. 9:19right than many people in the industry
  257. 9:21who view what is happening with
  258. 9:23excessive optimism.
  259. 9:25>> And how, in your opinion, should one
  260. 9:27properly measure the impact on a company
  261. 9:29or on personal productivity?
  262. 9:30>> Productivity has always been difficult
  263. 9:33to measure.
  264. 9:34And one of the problems is quality.
  265. 9:39Erik rightly emphasizes: "In the digital
  266. 9:42age, over the last 40 years, we still
  267. 9:44haven't learned how to measure quality.
  268. 9:47"
  269. 9:48Perhaps that is why we underestimate the
  270. 9:50benefits not only of artificial
  271. 9:52intelligence, but of computers in
  272. 9:54general.
  273. 9:55I agree with that.
  274. 9:58But we never really knew how to properly
  275. 10:00measure improvements in car or road
  276. 10:02quality before.
  277. 10:04So this problem has always existed.
  278. 10:06Maybe happiness levels?
  279. 10:08>> Yes,
  280. 10:09>> we should simply measure happiness
  281. 10:10levels and see how they change.
  282. 10:12>> That is also a good idea.
  283. 10:14Well-being is very important, but here I
  284. 10:17would rate AI and related technologies
  285. 10:20rather negatively.
  286. 10:23In my opinion, social media has caused
  287. 10:25us immense harm.
  288. 10:27It destroyed community and weakened our
  289. 10:29ability to cope with challenges because
  290. 10:32we stopped truly communicating with
  291. 10:35people.
  292. 10:36It brought out our worst instincts
  293. 10:39because algorithms bet on anger, emotion
  294. 10:42, and envy.
  295. 10:44So, I think it's terrible.
  296. 10:48I hope it will be different with AI.
  297. 10:51>> Let's talk about how to ensure that
  298. 10:53workers also benefit from new technology
  299. 10:55.
  300. 10:56>> My favorite topic.
  301. 10:58>> Yes, you talk about that a lot. You said
  302. 11:00the models are strong, but there aren't
  303. 11:02enough applications. What is the problem
  304. 11:03? Is it education, or do we need even
  305. 11:05more powerful models? What needs to
  306. 11:07happen for artificial intelligence to
  307. 11:09positively impact our daily lives?
  308. 11:12>> I think current models are already
  309. 11:14strong enough to build applications on
  310. 11:16top of them.
  311. 11:18You can do a lot with existing
  312. 11:20technology.
  313. 11:23Let me give a few examples.
  314. 11:26You can significantly improve quality
  315. 11:28control. With current models, it’s not
  316. 11:31difficult. Tools for nurses.
  317. 11:35Our country is in a real healthcare
  318. 11:38crisis. Almost every fifth dollar goes
  319. 11:41to medicine.
  320. 11:43And look at the results.
  321. 11:45Serious public health issues, many
  322. 11:48limitations within the system.
  323. 11:51Imagine if nurses could be more involved
  324. 11:54in treatment, helping with diagnostics,
  325. 11:56prescriptions, and patient condition
  326. 11:59decisions.
  327. 12:01One of the main barriers here is access
  328. 12:03to information and quality control.
  329. 12:07All this can be provided with existing
  330. 12:09models if you create suitable
  331. 12:11applications based on them.
  332. 12:13Yes, nurses will probably need to be
  333. 12:15trained a little differently.
  334. 12:17So, investment in human training will be
  335. 12:20needed in parallel.
  336. 12:22But the main obstacle is not training
  337. 12:24nurses. The main obstacle is the lack of
  338. 12:27applications and the fact that tech
  339. 12:29companies aren't building them. It’s the
  340. 12:32same with teachers.
  341. 12:34Electricians could also be given much
  342. 12:37more convenient tools. This is an
  343. 12:39obvious opportunity that is easy to
  344. 12:41capitalize on.
  345. 12:43In many trade professions, AI can
  346. 12:45noticeably increase efficiency.
  347. 12:48We are already seeing some examples in
  348. 12:50China.
  349. 12:51Why isn't this being done in the US?
  350. 12:53It’s not about the base models.
  351. 12:56For most such solutions, last year's
  352. 12:58models would suffice.
  353. 13:00>> Sounds like a great business opportunity
  354. 13:02.
  355. 13:02>> Yes, exactly.
  356. 13:05>> And for electricians, it’s a huge market
  357. 13:07. Only after moving here did I learn how
  358. 13:09much they earn and how much you have to
  359. 13:12pay for any little thing.
  360. 13:14>> And there is such a shortage of
  361. 13:15electricians that you have to wait 3
  362. 13:17weeks for even simple work.
  363. 13:18>> Yeah, yeah. Then the electrician comes
  364. 13:21and says, "No, I can't handle this new
  365. 13:23equipment."
  366. 13:24Why? Because I hadn't seen it before.
  367. 13:26>> That's where artificial intelligence
  368. 13:27comes in handy.
  369. 13:28>> Exactly. That's a very simple task for
  370. 13:31AI.
  371. 13:32>> And in medicine and education, is it a
  372. 13:33state task, or are there opportunities
  373. 13:35for companies there too?
  374. 13:36>> In medicine and education, two or even
  375. 13:39three parties must interact.
  376. 13:42Most healthcare spending is somehow tied
  377. 13:44to the state.
  378. 13:46It’s the same with education.
  379. 13:49So yes, the state can play an important
  380. 13:51role, but electricians, laborers,
  381. 13:53support staff, and scientists—that's
  382. 13:56primarily the private sector.
  383. 13:59Why did I mention three parties? Even
  384. 14:02without the state, there are two: tech
  385. 14:04companies and the business leadership
  386. 14:07itself. And both must participate.
  387. 14:10When I talk to CEOs and top managers of
  388. 14:12large companies about AI that will help
  389. 14:15their employees and make them more
  390. 14:17productive, they tell me: "That's
  391. 14:19exactly what we need.""We are all for it
  392. 14:21."
  393. 14:22Are they being sincere when they say
  394. 14:24that? I don't know. I can't verify it,
  395. 14:26because tech companies don't offer such
  396. 14:29tools yet.
  397. 14:32And when I talk about this to people
  398. 14:33from tech companies, they either nod and
  399. 14:35look away, or they reply: "You live in a
  400. 14:37different world."
  401. 14:39"We are creating AGI."
  402. 14:41"What are you even talking about?"
  403. 14:43>> They care about the distant future, not
  404. 14:44about what can be improved right now. We
  405. 14:46were talking about niches where routine
  406. 14:47tasks haven't been automated by anyone
  407. 14:49yet. For a long time, such a routine in
  408. 14:51my team was onboarding contractors. As
  409. 14:52you know, we have a distributed team all
  410. 14:54over the world. People work from
  411. 14:56different countries. With everyone, you
  412. 14:57need a contract and closing documents.
  413. 14:59And all of this must be stored so that
  414. 15:01before an audit or a meeting with an
  415. 15:02investor, you aren't scrambling to piece
  416. 15:03it together. You can onboard contractors
  417. 15:05in different ways. You can do it
  418. 15:06directly. Then the company manages every
  419. 15:08contract, background check, and document
  420. 15:10itself. You can use tools where
  421. 15:11contracts are just stored, but you
  422. 15:13remain the party to the contract. We
  423. 15:15have been working with Fodev.com for
  424. 15:16several years. The platform becomes a
  425. 15:18party to the contract with the
  426. 15:20contractor and manages their paperwork.
  427. 15:22With Fodev.com, documents for the remote
  428. 15:24team are generated in a unified format.
  429. 15:26Contractors can be hired from more than
  430. 15:28150 countries. And the rights to the
  431. 15:30work result can be secured immediately
  432. 15:31upon onboarding, if the business needs
  433. 15:33it. Contracts, closing documents, and
  434. 15:34the status of all people are stored in
  435. 15:36one place, and they can be shown at any
  436. 15:38moment.
  437. 15:39If you have a distributed team of more
  438. 15:40than twenty people, sign up for a demo
  439. 15:42at Fodev.com using the link in the
  440. 15:43description.
  441. 15:44They will show you how to bring order to
  442. 15:46your contracting process. Now, let's get
  443. 15:49back to our conversation. You mentioned
  444. 15:51solutions for manual labor and China.
  445. 15:53What are some good examples there?
  446. 15:55>> I’ve spoken with several manufacturing
  447. 15:57companies in China.
  448. 16:00Foxconn, BYD, and other electric vehicle
  449. 16:03manufacturers.
  450. 16:05They are all actively implementing
  451. 16:07robots and
  452. 16:10They are technologically very advanced,
  453. 16:12but their experience shows that the
  454. 16:14process is slow and always requires more
  455. 16:17engineers than expected.
  456. 16:20Fortunately for them, there is a huge
  457. 16:22number of engineers in China.
  458. 16:25What advice would you give to
  459. 16:26entrepreneurs watching us who are
  460. 16:28choosing what to do? How can one find a
  461. 16:29suitable niche right now? Lately, on
  462. 16:31podcasts, I’m constantly told: "The
  463. 16:33money is in small niches, especially
  464. 16:35with artificial intelligence."
  465. 16:36Opportunities exist in boring businesses
  466. 16:38that no one has automated for 20 years
  467. 16:40and where everything is still recorded
  468. 16:42on paper. You go there, solve the
  469. 16:44problem, and there’s the opportunity.
  470. 16:45>> Absolutely. I just wouldn’t call it
  471. 16:48boring. In my opinion, it’s very
  472. 16:50interesting. Imagine,
  473. 16:51>> boring from the outside. Does AI sound
  474. 16:53more? exciting.
  475. 16:54>> I understand, but that’s a completely
  476. 16:57wrong perspective. First of all, AGI
  477. 17:00remains a dream for now. No one can even
  478. 17:03agree on its definition. It is always
  479. 17:05somewhere in the future. And honestly,
  480. 17:08we can't really imagine that future.
  481. 17:12If by AGI, and later ASI, artificial
  482. 17:14superintelligence, we mean machines that
  483. 17:17can do absolutely everything humans can
  484. 17:20do, and do it better, then that seems to
  485. 17:23be a long way off.
  486. 17:24And that sounds like a dystopia. Then
  487. 17:27there will be no work left even for the
  488. 17:30best physicists, mathematicians, or
  489. 17:32communication specialists, as everything
  490. 17:34will be done better.
  491. 17:36There are many contradictions in public
  492. 17:38discussions about AGI. People say: "AGI
  493. 17:41is wonderful, it will appear soon, but
  494. 17:43people will still have things to do."
  495. 17:46That’s not how it works. If you
  496. 17:48seriously mean AGI or AI, there won’t be
  497. 17:51many tasks left for humans.
  498. 17:54>> I want to talk more about practical
  499. 17:55applications. And today you mentioned
  500. 17:58opportunities for researchers. How do
  501. 18:01you use it yourself in your work, and
  502. 18:03what else could be done?
  503. 18:05>> I must admit, I’m not at the forefront
  504. 18:09here.
  505. 18:10I use artificial intelligence to gather
  506. 18:13information on a topic, and this year
  507. 18:15I’ve been asking it more and more to
  508. 18:17critique my texts and ideas.
  509. 18:21And it’s very interesting.
  510. 18:24When I tried this with ChatGPT 3.5, and
  511. 18:27later with ChatGPT 4, the results
  512. 18:30greatly disappointed me.
  513. 18:32Now everything is much better.
  514. 18:36I advise all my students and research
  515. 18:39assistants to build a system of AI-based
  516. 18:42tools for research.
  517. 18:45Specifically to help with work, not to
  518. 18:47replace oneself.
  519. 18:50I don’t advise letting artificial
  520. 18:52intelligence write for you, even if
  521. 18:54English is not your native language.
  522. 18:57A person must be responsible for their
  523. 18:59own text.
  524. 19:01However, for students for whom English
  525. 19:03is not their native language, I
  526. 19:04recommend using artificial intelligence
  527. 19:06as a proofreader.
  528. 19:08Check the text for phrases that are
  529. 19:11unclear, too long, or grammatically
  530. 19:13incorrect.
  531. 19:15For such daily tasks, artificial
  532. 19:17intelligence is already very good.
  533. 19:20But I am worried about a different loss
  534. 19:22of control.
  535. 19:25We all, especially students and those
  536. 19:27who are learning, delegate too much to
  537. 19:29artificial intelligence.
  538. 19:31As a result, we don't train our thinking
  539. 19:33and don't engage in the learning process
  540. 19:35.
  541. 19:36In my opinion, this is one of the main
  542. 19:39problems.
  543. 19:40I see it in students.
  544. 19:43And studies show that in school
  545. 19:45education, the use of artificial
  546. 19:48intelligence has led to very serious
  547. 19:50problems. We introduced it blindly.
  548. 19:53Without any protective measures.
  549. 19:56I was talking about this back in 2024 or
  550. 19:59even 2023.
  551. 20:02Simultaneously unleashing generative
  552. 20:04artificial intelligence into all schools
  553. 20:07without any restrictions is a huge and
  554. 20:09dangerous social experiment.
  555. 20:12We didn't know what it would lead to.
  556. 20:14I can't say my worst fears have fully
  557. 20:17materialized, but we've ended up quite
  558. 20:20close to that. And how do you know when
  559. 20:22it's time to tell yourself: "That's it,
  560. 20:24I need to think for myself here." And
  561. 20:25only use artificial intelligence as a
  562. 20:27helper?
  563. 20:28>> For me, things are a bit different.
  564. 20:30Firstly, unfortunately, I am from a
  565. 20:32different generation. And secondly, I
  566. 20:34just love to think. I like to take a pen
  567. 20:36, paper, and sketch something out.
  568. 20:39So, I actually have to remind myself:
  569. 20:41try checking this with AI, for instance
  570. 20:44, I've never used AI for math.
  571. 20:46Although I should, it is much stronger
  572. 20:48than me in that, but I like solving
  573. 20:50things myself. When I write down models
  574. 20:53or do calculations, I do it by hand. I'm
  575. 20:56used to it, I'm fast at it, and in the
  576. 20:59process, I learn, including from my own
  577. 21:01mistakes.
  578. 21:03But looking from the outside, I could
  579. 21:05probably use AI more.
  580. 21:07>> And when you calculate yourself, don't
  581. 21:08you think about the history of
  582. 21:09calculators? Teachers were against them
  583. 21:11, but now we use them everywhere and in
  584. 21:13our spare time solve much more complex
  585. 21:15problems.
  586. 21:16AI might give us the same leap forward,
  587. 21:18or conversely, take away our critical
  588. 21:20thinking and leave us stuck in place.
  589. 21:23I'm worried about the latter.
  590. 21:26The comparison with the calculator is
  591. 21:28useful, but it must be made carefully.
  592. 21:31A calculator took on only a very small
  593. 21:33part of the work.
  594. 21:35I don't know anyone it did the thinking
  595. 21:37for, and I don't know a single school
  596. 21:39that said, "Now you have a calculator,
  597. 21:42so there's no need to learn
  598. 21:43multiplication."
  599. 21:45When I was studying, calculators already
  600. 21:48existed, but we all learned how to
  601. 21:50multiply and divide.
  602. 21:52I can't imagine how one could not know
  603. 21:54how to do that. It is the foundation for
  604. 21:56so many things.
  605. 21:58The same goes for derivatives. If you
  606. 22:01don't know calculus, how can you even
  607. 22:04take the next step?
  608. 22:06>> Yes, a calculator performs one specific
  609. 22:08function, while AI takes on the entire
  610. 22:10thought process,
  611. 22:11>> or a very large part of it.
  612. 22:14And the lines are blurred.
  613. 22:16Where should you stop? How do you work
  614. 22:19with artificial intelligence while still
  615. 22:21going through the key stages yourself?
  616. 22:24Coming up with something new, finding
  617. 22:26creative solutions.
  618. 22:27>> A short pause. Naron says we should
  619. 22:30delegate specific functions to AI, while
  620. 22:32keeping critical and creative thinking
  621. 22:34for ourselves. But how do you know
  622. 22:36exactly what to delegate?
  623. 22:38I compiled a list of thirty of my
  624. 22:40routine tasks—from sorting through
  625. 22:41hundreds of school emails and tracking
  626. 22:43my lab results to a dozen other mundane
  627. 22:45things I’ve handed off to AI.
  628. 22:49Thanks to this, I’ve gained time for
  629. 22:50much more important tasks and for my
  630. 22:52kids.
  631. 22:53This list is in my free newsletter.
  632. 22:55There, I share weekly the tools I
  633. 22:57personally use in my work and daily life
  634. 22:59. What I’ve tried, what I’ve noted, and
  635. 23:01what I’ve been disappointed in.
  636. 23:03Subscribe, the link is in the
  637. 23:04description. The newsletter is
  638. 23:05completely free. Let’s get back to their
  639. 23:08business. On X, I constantly see posts
  640. 23:10saying, "Instead of hiring a person, I
  641. 23:13launched an agent." What are such
  642. 23:15companies and people missing in the long
  643. 23:17run? I know you are against selling
  644. 23:19agents as a replacement for humans.
  645. 23:21>> Let me clarify my position.
  646. 23:24It is based on my research into robots,
  647. 23:2620 th-century technology, the Industrial
  648. 23:29Revolution, and theoretical works.
  649. 23:32I’m not against automation, and I don’t
  650. 23:34think it can be stopped.
  651. 23:36We need it, and I’m against trying to
  652. 23:40stop it. We certainly don’t want to go
  653. 23:42back to the days when people carried
  654. 23:44heavy loads on their backs.
  655. 23:46Automation is great. The problem is that
  656. 23:50we are focusing only on that.
  657. 23:52Firstly, we won’t get all the possible
  658. 23:55growth in productivity that way.
  659. 23:57Automating everything in sight isn't
  660. 23:59that simple. Secondly, we won’t achieve
  661. 24:02prosperity for everyone.
  662. 24:05Too many jobs will disappear, and in
  663. 24:07return, no new, meaningful, high-paying
  664. 24:10, productive roles that open doors for
  665. 24:13people will appear.
  666. 24:15That’s why I’ve always said: we need a
  667. 24:18balance.
  668. 24:19Use new technologies for both automation
  669. 24:22and other tasks.
  670. 24:24Digital technologies gave us many
  671. 24:26opportunities to create new occupations
  672. 24:28for people.
  673. 24:30And this applies twice as much to AI.
  674. 24:33I’m for AI-driven automation, provided
  675. 24:36that in parallel we use it to create new
  676. 24:39opportunities for nurses, electricians,
  677. 24:42and many professions we can't even
  678. 24:45imagine today.
  679. 24:47>> How do you explain this to a small
  680. 24:49business owner whose main goal is profit
  681. 24:51growth? Why is it bad from an economic
  682. 24:53standpoint?
  683. 24:53>> In productivity.
  684. 24:56In my experience, most companies
  685. 24:57understand this, although they might
  686. 24:59just be being polite to me.
  687. 25:01Or maybe I convinced them? Not all, but
  688. 25:04most understand that employees are your
  689. 25:07primary asset. Imagine if they could
  690. 25:09work more productively. For a business,
  691. 25:11that’s wonderful, and by the numbers,
  692. 25:14it’s more profitable. Let’s say you have
  693. 25:16100 employees.
  694. 25:18I offer a choice: automate 5%of the jobs
  695. 25:21and cut costs for them in half, or
  696. 25:23increase the productivity of all
  697. 25:25employees by 5%.
  698. 25:28The second option will bring in much
  699. 25:30more. they will automate the work. If
  700. 25:32you get rid of people, your productivity
  701. 25:34will ultimately decline.
  702. 25:36>> Exactly.
  703. 25:37Employees have a lot of experience, and
  704. 25:40many, if you have been a good manager,
  705. 25:42are also truly loyal to the company.
  706. 25:46If you help them become more inventive,
  707. 25:49better at solving problems, and better
  708. 25:51at doing new things, such a business
  709. 25:53will be successful.
  710. 25:55Let me ask you, which companies from the
  711. 25:58past do you remember?
  712. 26:00You will surely name familiar names:
  713. 26:03Ford, General Motors, General Electric,
  714. 26:07IBM, Microsoft; what unites them all?
  715. 26:10They did something new. They didn't just
  716. 26:13cut costs and repeat after others.
  717. 26:16They invented new markets, technologies
  718. 26:18, and tasks.
  719. 26:20Henry Ford put it this way: "If I had
  720. 26:22asked customers what they wanted, they
  721. 26:25would have asked for faster horses."
  722. 26:28That is what we expect from successful
  723. 26:30companies. And cutting costs by 5%is not
  724. 26:32enough to go down in history.
  725. 26:35>> You said that 5%of tasks can be
  726. 26:37automated. But how do you understand
  727. 26:39which ones are worth automating and
  728. 26:40which are not? Do you have any kind of
  729. 26:42rule?
  730. 26:43>> If you cut costs by 5%, then GDP usually
  731. 26:46grows by those 5%multiplied by your
  732. 26:49company's share of the GDP.
  733. 26:52Everything is simple here. Complications
  734. 26:54begin if the quality changes.
  735. 26:57And even to cut costs by 5%over several
  736. 27:00years, you need to work very hard. I see
  737. 27:03how some companies I know rush into
  738. 27:06automation.
  739. 27:08Some think it will be easy, others feel
  740. 27:10the pressure, and then nothing works out
  741. 27:13because automation is a difficult task.
  742. 27:16It needs to be integrated into the
  743. 27:18company's work. Many are saying now that
  744. 27:20AI has a jagged frontier of capabilities
  745. 27:23. It is strong in some tasks, but
  746. 27:25unexpectedly fails in others. I like
  747. 27:27this image, but in essence, it is a
  748. 27:30polite way of saying: "We overestimated
  749. 27:32it."
  750. 27:33AI turned out to be much more complex,
  751. 27:35and we thought we could cut staff thanks
  752. 27:37to AI, but that is not the case.
  753. 27:40And if you think about it seriously, it
  754. 27:42turns out that I will have about the
  755. 27:44same number of employees left.
  756. 27:46AI is added, and tasks are mixed between
  757. 27:48them.
  758. 27:49>> Yes, there is only more work. Let's move
  759. 27:52on to education.
  760. 27:53You work at a university.
  761. 27:55Will higher education still be worth the
  762. 27:58money in 10 years?
  763. 27:59>> I think so.
  764. 28:02But with a few caveats. Firstly, I have
  765. 28:05never believed that everyone should go
  766. 28:08to university.
  767. 28:10And now, for economic and social reasons
  768. 28:13, I am even more convinced of this.
  769. 28:16People have different abilities and
  770. 28:18interests.
  771. 28:20You need to learn your whole life, but a
  772. 28:22broad university education does not suit
  773. 28:24everyone.
  774. 28:26And we don't need everyone in society to
  775. 28:28look at the world in the same way.
  776. 28:30Secondly, a university diploma is, in
  777. 28:33many ways, also a signal to others. By
  778. 28:37enrolling at Harvard, MIT, or Stanford,
  779. 28:39you are buying the name, among other
  780. 28:41things.
  781. 28:42That is good for you, but not
  782. 28:45necessarily for society.
  783. 28:47Third, all universities now realize they
  784. 28:50need to change their business model and
  785. 28:53the way they teach.
  786. 28:55They are searching for which skills to
  787. 28:58focus on. From my rather superficial
  788. 29:00point of view, universities realize this
  789. 29:03. But, you know, the same applies to
  790. 29:06schools.
  791. 29:07And they don't seem to understand it
  792. 29:09there. Schools are lagging even further
  793. 29:11behind.
  794. 29:13In the US, I see that many students
  795. 29:15arrive at university poorly prepared.
  796. 29:18The situation will worsen, and it will
  797. 29:20become even harder for universities.
  798. 29:24I wrote an economics textbook for
  799. 29:25freshmen and I see how often they lack
  800. 29:27basic knowledge, and it’s getting worse
  801. 29:30over the years.
  802. 29:32This started even before the arrival of
  803. 29:34artificial intelligence.
  804. 29:35Universities are having to do more and
  805. 29:38more of what should have been taught in
  806. 29:40school because schools are failing. We
  807. 29:43are still at the very beginning;
  808. 29:45everything could still change.
  809. 29:47But the available data shows that
  810. 29:49artificial intelligence has complicated
  811. 29:51learning in middle and high school.
  812. 29:53This means this problem will only
  813. 29:56intensify.
  814. 29:57>> Yes, if AI solves all the problems, it
  815. 29:59will only be harder at university. So
  816. 30:01what should be done? Which skills should
  817. 30:02be developed? And do they require even
  818. 30:04more academic knowledge? I am thinking,
  819. 30:06for example, about taste and strategic
  820. 30:08thinking.
  821. 30:09>> Absolutely. I do not believe that true
  822. 30:12AGI will appear any day now.
  823. 30:15Therefore, human creativity is needed
  824. 30:17more than ever, and creativity without a
  825. 30:19solid foundation is simply impossible.
  826. 30:24If you want to know my opinion—though
  827. 30:26I’m not the best person for such advice
  828. 30:29, I’m not a mathematician or a computer
  829. 30:32scientist—go study mathematics.
  830. 30:34Don't be discouraged by the fact that AI
  831. 30:37is already solving complex mathematical
  832. 30:39problems. We need human knowledge, a
  833. 30:42deep understanding of mathematics. This
  834. 30:45will be very important.
  835. 30:47Study physics, economics, engineering.
  836. 30:51All of this will be useful.
  837. 30:53But also learn flexibility. We will
  838. 30:56really need it. Even while staying in
  839. 30:59the same profession, we will often
  840. 31:01change our work following technologies
  841. 31:03and societal changes. This is truly
  842. 31:05important.
  843. 31:08But in the next 10 years, in my opinion
  844. 31:10, human ingenuity and creativity will
  845. 31:12play a key role.
  846. 31:14>> Yes, and it’s time to stop considering a
  847. 31:15non-linear career something strange.
  848. 31:17That resonates with me. I have changed
  849. 31:19my field of activity more than once
  850. 31:20myself. I think this will soon become
  851. 31:21the absolute norm.
  852. 31:22>> And two more principles that, in my
  853. 31:24opinion, should be instilled starting
  854. 31:26from middle school.
  855. 31:28Of course, learning to work with AI, but
  856. 31:31placing even greater emphasis on civic
  857. 31:34responsibility.
  858. 31:37In our country, and indeed almost all
  859. 31:39over the world, we have lost a real
  860. 31:41sense of community, care for the common
  861. 31:44good, and willingness to work for a
  862. 31:46common cause.
  863. 31:49In the era of AI, this is especially
  864. 31:52important, and it all starts with
  865. 31:54understanding what is happening and
  866. 31:57participating in society.
  867. 31:59>> In fact, we need to start as early as
  868. 32:01elementary school.
  869. 32:02>> Yes, of course, we can start in
  870. 32:04elementary school as well.
  871. 32:06Perhaps we have encouraged selfishness
  872. 32:09too much, pushing people to retreat into
  873. 32:11their own cocoons, on social media and
  874. 32:14beyond. We have devalued the important
  875. 32:16aspects of community life.
  876. 32:19This is one of the main themes of my new
  877. 32:21book about what happened to liberal
  878. 32:23democracy.
  879. 32:25At the center of a liberal democratic
  880. 32:28society, there must be community.
  881. 32:31I am neither a pessimist nor an optimist
  882. 32:34, but I am definitely not a pessimist. I
  883. 32:37try to remain hopeful.
  884. 32:40And two things give me that hope.
  885. 32:43First, people are capable of adapting to
  886. 32:46anything.
  887. 32:48Second, even in such a divided world,
  888. 32:51our ideas of justice are very similar.
  889. 32:55There are shared moral values on both
  890. 32:57sides of the political spectrum. We just
  891. 33:00need to articulate them more clearly and
  892. 33:01stop seeing the other side as the enemy
  893. 33:03.
  894. 33:05Even with the deepest disagreements, it
  895. 33:07is important to remember that we are all
  896. 33:09in the same boat.
  897. 33:11Oh, right, we discuss all this, and then
  898. 33:13Dario says, "There is a 25%chance that
  899. 33:15things will go very, very wrong with
  900. 33:17artificial intelligence."
  901. 33:19What do you think when you hear that?
  902. 33:21Especially from a person who will soon
  903. 33:23take his company public and prove that
  904. 33:25his technology will truly change the
  905. 33:26world.
  906. 33:27>> If people in leading AI labs are worried
  907. 33:30about the consequences of this
  908. 33:33technology, of course, they should talk
  909. 33:36about it.
  910. 33:38I welcome and respect their courage.
  911. 33:41But I am much more concerned by the loss
  912. 33:44of societal control over AI than by the
  913. 33:47existential threat to humanity that
  914. 33:50Dario and Elon Musk talk about. I mean
  915. 33:53that ordinary people—workers, students,
  916. 33:56parents—are losing the ability to make
  917. 33:59decisions for themselves and manage
  918. 34:02their own lives.
  919. 34:05Not necessarily because AI has become
  920. 34:08superintelligent and turned into an
  921. 34:11autonomous entity, but partly because of
  922. 34:14how companies and leading labs develop,
  923. 34:17use, and market artificial intelligence
  924. 34:20.
  925. 34:22This loss of control is what worries me
  926. 34:25deeply. We must talk about the
  927. 34:26existential threat to humanity, but we
  928. 34:29cannot ignore other risks in the process
  929. 34:31.
  930. 34:31You rated artificial intelligence on a
  931. 34:34scale from -10 to +10, and you gave it a
  932. 34:36-6. Has it gotten worse since then? I
  933. 34:38believe you said that in 2025.
  934. 34:41>> No, I don't think so.
  935. 34:43In 2024 and 2025, I thought that AI
  936. 34:46could cause great harm, but I didn't yet
  937. 34:50understand what obstacles would arise in
  938. 34:53its development path. Progress has been
  939. 34:56rapid, so the risks have increased as
  940. 34:59well.
  941. 35:00But now it is also clearer how much good
  942. 35:02can be done with AI.
  943. 35:05We just need to change direction. We
  944. 35:07need to guide the development of AI
  945. 35:09ourselves.
  946. 35:11And its fate cannot be decided by a
  947. 35:13handful of people, even if they are very
  948. 35:15smart and largely have the best of
  949. 35:16intentions. This is too important to
  950. 35:19leave to just a few people.
  951. 35:21>> And what can each of us do to achieve
  952. 35:23this?
  953. 35:24I think we need to start with
  954. 35:27information and with how artificial
  955. 35:29intelligence is being discussed.
  956. 35:33I’m about to say something risky about
  957. 35:35those who largely decide what we even
  958. 35:37learn about the world.
  959. 35:40But, in my view, the media has failed us
  960. 35:43the most.
  961. 35:45The media itself.
  962. 35:47They have done a very poor job of giving
  963. 35:49people the information needed for a
  964. 35:51meaningful conversation about artificial
  965. 35:53intelligence.
  966. 35:54For 10 years, there were two extremes.
  967. 35:57It's changing a little now, but very
  968. 36:01slowly and unevenly. Either Elon Musk
  969. 36:04and his ilk could do no wrong, and any
  970. 36:06criticism only elevated them.
  971. 36:09And everything related to artificial
  972. 36:11intelligence was delightful, amazing,
  973. 36:13and wonderful.
  974. 36:15Or artificial intelligence was portrayed
  975. 36:18as an enemy, either as a parrot just
  976. 36:21repeating words or as a killer robot.
  977. 36:24There was no room for a balanced
  978. 36:26conversation so people could understand
  979. 36:29, get involved, and influence our future
  980. 36:32as consumers and citizens of a
  981. 36:34democratic society.
  982. 36:36The media didn't create that space, and
  983. 36:38this applies to almost every news
  984. 36:40organization.
  985. 36:41>> Let's finish with a few practical
  986. 36:43questions. I want the audience to be
  987. 36:45left with a clear plan of action. What
  988. 36:47do you do with your savings when you
  989. 36:48hear this news? Or has nothing changed
  990. 36:50for you?
  991. 36:50>> Almost nothing. I keep a balanced
  992. 36:53portfolio. Beating the market is very
  993. 36:55difficult.
  994. 36:56To do that, you either need to be an
  995. 36:59expert in investing, spend a lot of time
  996. 37:01, or have insider information. I don’t
  997. 37:04have that much time. I don’t have
  998. 37:06insider information either, and I
  999. 37:09wouldn't want it, so I stick to a
  1000. 37:11balanced portfolio.
  1001. 37:13Let’s fast-forward to 2035.
  1002. 37:15If everything goes well, what will a
  1003. 37:17typical workday look like for a
  1004. 37:19specialist who spends 8 hours at a
  1005. 37:20computer today?
  1006. 37:21>> Great question.
  1007. 37:24I don't know.
  1008. 37:27I can easily imagine a future where
  1009. 37:29people work a little less and remain
  1010. 37:32very productive, but I can also imagine
  1011. 37:35a scary future where many won’t have
  1012. 37:38meaningful work.
  1013. 37:41And I can imagine people still going to
  1014. 37:44the office or combining office and
  1015. 37:46remote work, but working differently
  1016. 37:48because artificial intelligence truly
  1017. 37:51augments their capabilities, with human
  1018. 37:53ideas and talent at the center.
  1019. 37:57Even leaving artificial intelligence
  1020. 37:59aside, we live in extraordinarily
  1021. 38:01turbulent times.
  1022. 38:04Climate change, an aging population,
  1023. 38:06geopolitical shifts, and the huge middle
  1024. 38:09class emerging in developing countries.
  1025. 38:13All of this requires new goods, services
  1026. 38:15, tasks, and new approaches to markets
  1027. 38:18and company organization.
  1028. 38:20And who will handle that? Not artificial
  1029. 38:23intelligence, but people.
  1030. 38:26And it would be wonderful if we helped
  1031. 38:28these people think creatively.
  1032. 38:31What would you advise those who see the
  1033. 38:32massive wealth being created by
  1034. 38:34artificial intelligence?
  1035. 38:36At least judging by company valuations,
  1036. 38:37while real economic returns are yet to
  1037. 38:39be seen. How can one practically
  1038. 38:41participate in creating all of this?
  1039. 38:44>> Honestly, despite all the investment in
  1040. 38:46leading AI labs, I still don't
  1041. 38:48understand how they plan to make big
  1042. 38:50money on foundation models.
  1043. 38:52>> So, we are in for some failed IPOs.
  1044. 38:54>> Not immediately.
  1045. 38:55For now, there is enough enthusiasm, but
  1046. 38:58I really don't see how to make trillions
  1047. 39:00of dollars from base models.
  1048. 39:03Their development or distribution may
  1049. 39:06face serious obstacles.
  1050. 39:08And even if that doesn't happen, the
  1051. 39:10next generation of open models will be
  1052. 39:13very good, so the potential to make
  1053. 39:15money from the models themselves will be
  1054. 39:18limited.
  1055. 39:19The money, in my opinion, is in quality
  1056. 39:21data and applications.
  1057. 39:23That is where you can gain an advantage
  1058. 39:25.
  1059. 39:26If your company or industry has a lot of
  1060. 39:29data, work with it, use it, improve its
  1061. 39:32quality, find ways to collect new data,
  1062. 39:35and pay the employees who create it.
  1063. 39:39This is both fair and good for business
  1064. 39:42. Think about applications. That is what
  1065. 39:45will provide the advantage.
  1066. 39:47If an application is truly original and
  1067. 39:49solves a problem in a specific niche, it
  1068. 39:51will be much harder to replicate it
  1069. 39:53using open models.
  1070. 39:55That is where I see the opportunities.
  1071. 39:57And for workers, it is important to stay
  1072. 39:59flexible, understand AI, but not lose
  1073. 40:01their creative thinking and fundamental
  1074. 40:04knowledge.
  1075. 40:05>> It is also important for employees to
  1076. 40:06understand that they are the source of
  1077. 40:07this data, having worked with clients
  1078. 40:09and products for years.
  1079. 40:10>> By the way, this is also important for
  1080. 40:12those building applications. A lot of
  1081. 40:14practical knowledge exists only within
  1082. 40:16the employees. It is not written down
  1083. 40:18anywhere.
  1084. 40:19You can't just sit at a computer in an
  1085. 40:20office, take a powerful AI, and invent
  1086. 40:22an application. You need to go on-site,
  1087. 40:24understand the work, and talk to people
  1088. 40:27.
  1089. 40:27>> I hope this conversation gives ideas to
  1090. 40:29the entrepreneurs watching us who want
  1091. 40:31to create useful AI applications. And
  1092. 40:33one last question.
  1093. 40:35You said that students joining AI
  1094. 40:37companies need a reliable moral compass
  1095. 40:39. I think everyone needs one,
  1096. 40:41entrepreneurs and every member of
  1097. 40:42society alike. What do you mean by that
  1098. 40:44? Could you please explain that in more
  1099. 40:46detail?
  1100. 40:47>> Let me give an example from another
  1101. 40:49field. Gene editing.
  1102. 40:52It is also an amazing technology that
  1103. 40:55has seen major breakthroughs.
  1104. 40:58But most people going into medicine or
  1105. 41:01related fields understand that there is
  1106. 41:04a lot of ethical ambiguity here.
  1107. 41:07And many regulatory bodies, from the
  1108. 41:10National Institutes of Health to the
  1109. 41:12U.S. Surgeon General and the American
  1110. 41:15Medical Association, have issued ethical
  1111. 41:18guidelines for gene editing.
  1112. 41:21For example, germline gene editing,
  1113. 41:23where changes are passed on to
  1114. 41:25descendants, is currently not permitted
  1115. 41:27.
  1116. 41:27It is too dangerous. We don't understand
  1117. 41:30the consequences well enough. There are
  1118. 41:32too many risks.
  1119. 41:34We need such moral guidelines in
  1120. 41:36artificial intelligence as well, but we
  1121. 41:38don't have them.
  1122. 41:40People say, "I am doing this for the
  1123. 41:42greater good," and I believe them.
  1124. 41:45Almost everyone I know in this field
  1125. 41:47sincerely wants to benefit not just
  1126. 41:49themselves, but society as a whole. But
  1127. 41:51there are no clear moral guardrails.
  1128. 41:54That is what I am talking about.
  1129. 41:55>> Another interesting topic. Perhaps we
  1130. 41:57will continue it separately.
  1131. 41:58>> I hope there will be another opportunity
  1132. 42:00to continue. Thank you so much for an
  1133. 42:01amazing conversation.

About this transcript

This page contains the full transcript of Кого ИИ сделает богаче, а кого оставит без работы | Нобелевский лауреат по экономике by Marina Mogilko, generated from the public captions YouTube serves with the video. The transcript has 6,456 words across 1,133 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.