Are we losing our minds to AI? | BBC News — Transcript

Full transcript

1. 0:08Hello and welcome to AI Decoded with me,
2. 0:11Mark Chislac, the BBC's AI
3. 0:12correspondent. This week, we're asking a
4. 0:15question that goes right to the heart of
5. 0:17what makes us human. AI can write
6. 0:19essays, generate software code,
7. 0:21summarize research papers, even help
8. 0:23plan journeys and holidays. Every single
9. 0:25week, tech companies claim another task
10. 0:28that once required expertise and years
11. 0:31of experience can now be completed with
12. 0:33a carefully worded prompt. But as AI
13. 0:36becomes a bigger part of our lives, are
14. 0:39we becoming less skilled? Are we
15. 0:41witnessing something researchers are
16. 0:43calling cognitive offloading?
17. 0:45Outsourcing our thinking to machines to
18. 0:48explore whether AI is enhancing human
19. 0:50intelligence or quietly eroding it.
20. 0:52We're joined by two people with slightly
21. 0:55different perspectives on this
22. 0:58transformative technology. Oasis Gelini
23. 1:00is one of Britain's leading pediatric
24. 1:03neurosurgeons based at Great Orman
25. 1:05Street Hospital. And from Helsinki, Dr.
26. 1:07Tapani Rintteralia, who conducts
27. 1:09research into how digital technologies
28. 1:11transform human work, expertise, and
29. 1:14decision-making. And of course, we're
30. 1:16joined by AI decoded co-host Dr.
31. 1:19Stephanie Hair, author of technology is
32. 1:21not neutral. Oasis, I'm going to start
33. 1:23with you if I may. Medical technology is
34. 1:26always evolving from stethoscopes all
35. 1:29the way through to things like MRI
36. 1:30scanners. Now, this program's recently
37. 1:33followed some of your work to separate
38. 1:35conjoined twins, joined at the head. The
39. 1:37surgery was successful, and that
40. 1:40wouldn't have been possible without some
41. 1:42advanced technology, but also a massive
42. 1:45amount of human skill. So what do you
43. 1:48think of the concept of AI assistance?
44. 1:52>> I think it's hugely hugely important AI
45. 1:55assistance. I know that's a strong
46. 1:57stance. So allow me to explain.
47. 2:00>> We we are focusing perhaps a bit too
48. 2:03much on the harm that AI can can do or I
49. 2:07mean clearly there's the potential for
50. 2:09enormous harm but there's a lot of focus
51. 2:11on the harm it it can do to us humans in
52. 2:14the future.
53. 2:14>> [snorts]
54. 2:15>> there's not enough attention given to
55. 2:17the potential harm that the human mind,
56. 2:20our way of thinking can do and does to
57. 2:22humans.
58. 2:23>> Okay.
59. 2:24>> And it's important to draw that balance
60. 2:25because AI when used properly and we
61. 2:28will discuss what that means. AI when
62. 2:30used properly has the potential to
63. 2:33improve our way of thinking
64. 2:35exponentially exponentially. But the
65. 2:37trick is getting that balance right
66. 2:38where we use it at the right level in
67. 2:41the right way to answer the questions
68. 2:44appropriate questions but not rely on it
69. 2:46too much.
70. 2:47>> Okay, Tony I want to bring you in here.
71. 2:49The word deskkilling sounds dramatic.
72. 2:53What exactly do researchers mean when
73. 2:55they talk about AIdriven deskkilling?
74. 2:58>> Yes. So there are a couple of different
75. 2:59meanings actually for that. So typically
76. 3:02I think the way we think about it
77. 3:04especially with AI we see it as an
78. 3:06unintended side effect of technology in
79. 3:09a way I also I call it skill erosion. So
80. 3:12that could be something that happen.
81. 3:13You're an expert. You rely too much on
82. 3:15automation AI and then you lose your
83. 3:18skills over time. But then you could
84. 3:20also refer to not being an expert yet,
85. 3:23but you are actually training to be one.
86. 3:25And in the training nowadays, you often
87. 3:28have many automated tools and then you
88. 3:31never really grapple the real kind of
89. 3:34nitty-gritty part of the work because
90. 3:36you have such a sop sophisticated tools
91. 3:38and then you never really become that
92. 3:40skilled. And then there is also a third
93. 3:42meaning which is more of a from the work
94. 3:44organizing perspective
95. 3:46uh deskkilling has been uh defined as
96. 3:50simplifying the work task. So using
97. 3:52automation uh to simplify the work task
98. 3:55so that less skilled workers can do
99. 3:58that. So then you can actually hire less
100. 4:00skilled workers who uh are typically
101. 4:03cheaper labor.
102. 4:04>> A way I want to bring it into your
103. 4:06sphere of of expertise into medicine. If
104. 4:09a young doctor learns how to practice
105. 4:11medicine with AI support from day one,
106. 4:15could they develop differently from
107. 4:18earlier generations of clinicians?
108. 4:20>> For me, Mark, there is no question of
109. 4:21that. It is um you know the analogy that
110. 4:25I would use is I'll go as far back as
111. 4:27the discovery of fire. When humans
112. 4:30discovered fire and the use of fire, a
113. 4:32lot changed not just with our physiology
114. 4:34but with our way of life. We learned how
115. 4:36to use it for the most part. creatively
116. 4:39and then the humans chained and adapted
117. 4:42to this technology. AI is going to do a
118. 4:44similar thing for humans. Humans will
119. 4:46adapt and adjust and change
120. 4:48physiologically, anatomically because AI
121. 4:50exists in this world. The question
122. 4:52really is how do we how do we how do we
123. 4:56monitor that change and how do we lead
124. 4:58that change? And that's for me the the
125. 5:01question is you know AI can go in many
126. 5:03different directions. It is a tool which
127. 5:05is available to us at the moment. But
128. 5:07what we do have a clear say on is the
129. 5:11direction we want to end up in. What
130. 5:13would we like to become in 20 50 years
131. 5:16time one we have some control over that
132. 5:18for sure. And the pace with which it
133. 5:21happens we have control over that too.
134. 5:23So the doctors of the future, the people
135. 5:25that are really going to crack the the
136. 5:28enigmas and get the inventions and do
137. 5:31the things that we really would like
138. 5:32them to do are the people that learn to
139. 5:34use the AI as an augmentative tool. We
140. 5:38we've spoken about this before. I think
141. 5:40the word artificial intelligence is the
142. 5:42wrong characterization. There's nothing
143. 5:44artificial about artificial
144. 5:46intelligence.
145. 5:47>> I would feel much more comfortable if we
146. 5:50change the lingo, if it's possible at
147. 5:52all. I don't know. But if we change the
148. 5:53linger,
149. 5:54>> that ship has already sailed. I think
150. 5:56>> it back and call it augmentative
151. 5:57intelligence because that is what it is.
152. 5:59It's here to stay whether we like it or
153. 6:01not.
154. 6:02>> But if we can use it, augment it, use it
155. 6:04to augment our intelligence in the right
156. 6:06way,
157. 6:07>> that's when we make the most of it.
158. 6:09>> So what do you perceive as the risks in
159. 6:12the future if doctors become really
160. 6:14really good at interpreting things like
161. 6:16AI recommendations but less confident at
162. 6:20challenging those recommendations? Okay.
163. 6:22So,
164. 6:24the first thing we need to remember is
165. 6:26that Moses did not bring AI down from
166. 6:29the mountain. That's not where AI came
167. 6:31from. Okay. So, where did AI? [laughter]
168. 6:34>> I don't remember that anywhere in my
169. 6:35Bible study at school. I went to went to
170. 6:37a Catholic school and I was educated by
171. 6:39nuns and I definitely don't remember
172. 6:41anything about large language models in
173. 6:44the the first bit of the Bible.
174. 6:45>> The Ten Commandments, the last time I
175. 6:47read them again, does not talk about AI.
176. 6:49It does not come from there.
177. 6:51>> That shall not prompt was definitely not
178. 6:53[laughter] in there.
179. 6:54>> Indeed. And so, you know, I and that's
180. 6:56really important to just keep in mind
181. 6:58and you know, I'm I'm making a little
182. 6:59light of it. But the point is a lot of
183. 7:03what AI is based on. Not all of it, but
184. 7:05a lot of what AI is based on is human
185. 7:07generated or human interpreted
186. 7:09knowledge. What we how we see the world,
187. 7:11what our reality is. And that is what a
188. 7:14lot of our language models are based on
189. 7:15at the moment and our AI platforms. Now,
190. 7:17why is that important? Because there are
191. 7:20shortcomings in the human way of
192. 7:21thinking. Our way of thinking, we deal
193. 7:25with data. Our way of thinking is
194. 7:27replete with biases, is replete with
195. 7:30prejudices and our strategic ambiguity.
196. 7:32Right? And it is important to bear in
197. 7:35mind that the AI agents that are being
198. 7:38developed at the moment and will carry
199. 7:40on do have those shortcomings as well.
200. 7:43So we we can't take them as gospel. we
201. 7:46they are good assistants but they can
202. 7:48make all the errors that humans make and
203. 7:50it's really important to be cognizant of
204. 7:52that. So for the doctors of the future
205. 7:54if they can learn to work with AI in
206. 7:56that fashion for certain task they far
207. 7:58exceed human capabilities for sure and
208. 8:00they've proven that time and time again.
209. 8:02So great, it's a really good tool for
210. 8:05this particular task. But if you then
211. 8:07say turn around and say, okay, how would
212. 8:08you separate a set of twins? Can I just
213. 8:10rely on my AI? That's not going to work.
214. 8:13But where it will help is when we did
215. 8:15our most recent separation of twins.
216. 8:17There was one particular task I needed
217. 8:18to understand if I use distractors
218. 8:21between these two bones, how much
219. 8:23distraction can I achieve over a period
220. 8:25of a few weeks. Well, we have an AI
221. 8:27model that we've built to give us that
222. 8:29answer. We use that model. It was quite
223. 8:31accurate. So to for that specific bit of
224. 8:34information AI was wonderful but the
225. 8:36whole package how you put it together is
226. 8:38still the domain of the doctor.
227. 8:40>> You're an expert in your field though
228. 8:42and you have taken you know the you have
229. 8:45exams at a at regular at regular
230. 8:48intervals and more
231. 8:51you are you are constantly tested and
232. 8:53constantly learning what you what you
233. 8:56actually do. When we think about these
234. 8:57technologies and their applications
235. 8:59outside of fields where people are
236. 9:02experts at an elite level,
237. 9:05>> are there different sorts of risks?
238. 9:08>> Mark, the the reason I'm called an
239. 9:10expert and the reason I am doing the
240. 9:13work I'm doing is because I work at the
241. 9:16edge of my abilities pretty much every
242. 9:18week.
243. 9:19>> You're in the swimming pool and your
244. 9:20feet are only just touching the bottom
245. 9:23of the pool.
246. 9:24>> Uh, yes. Yes. That's your comfort zone.
247. 9:27>> So when we're doing these surgeries, a
248. 9:29lot of these surgeries have not been
249. 9:30done before. We have to push the
250. 9:31boundaries. We have to discover new ways
251. 9:33of doing things. We have to invent new
252. 9:35platforms. So this is all iterative
253. 9:38which adds every time we do this adds to
254. 9:40our knowledge base that little bit and
255. 9:42working at the edge of your capabilities
256. 9:44is what makes you go forward. Now if you
257. 9:47for whatever reason take that away from
258. 9:49the doctors of the future, they will not
259. 9:51progress as we do. AI is not going to
260. 9:54come in and create creativity for them.
261. 9:57That is very much a human thing. We have
262. 9:59to lead that and then use AI as an
263. 10:01augmentative tool.
264. 10:02>> Stephanie,
265. 10:03>> my understanding is that robot surgery
266. 10:05is really really good for a lot of
267. 10:07different types of surgeries and many
268. 10:09surgeons have been worried that they
269. 10:11could be replaced and yet we don't
270. 10:13replace them. We keep human surgeons
271. 10:17at their ability where they, you know,
272. 10:19they could scrub up and do an operation
273. 10:20today if they absolutely had to in an
274. 10:22emergency situation. You don't want
275. 10:24those skills and experience and
276. 10:26confidence to atrophy. So, I think about
277. 10:29that a lot for all of us. Most of us
278. 10:30aren't under the pressure of being a
279. 10:32surgeon, but all of us have to be able
280. 10:35to constantly have that awareness of who
281. 10:37am I with my augmented self, all of my
282. 10:40tools, all of my technologies, of which
283. 10:41AI is probably the most powerful at the
284. 10:44moment. And then who am I, you know,
285. 10:47just on my own sheer talent, experience,
286. 10:51and how do you train? How are you
287. 10:53mindfully training yourself? That's what
288. 10:56I that's what I worry about.
289. 10:57>> I want to bring into pan here. I want to
290. 10:59bring in Pani because Tony what do you
291. 11:00think when when AI and human judgment
292. 11:03sort of disagree
293. 11:04who do you think ultimately should have
294. 11:06the final say?
295. 11:07>> Oh well depends on context obviously but
296. 11:10mainly human like in most cases
297. 11:12definitely human. It is a very powerful
298. 11:14augmentative tool especially for those
299. 11:16who are already you know experts who who
300. 11:19can who know how to use it very
301. 11:20mindfully and and and responsibly. Um
302. 11:24but there's a certain tendency of course
303. 11:26humans like there's a whole body of
304. 11:27research on automation complacency
305. 11:29uh when we have automated systems like
306. 11:31let's say in aviation you're a pilot and
307. 11:34those systems work so well we get kind
308. 11:36of get used to those system working
309. 11:38really well and when you know sometimes
310. 11:40something happens technology fails or
311. 11:42for some reason you can't use it maybe
312. 11:44it's a type of a situation that the the
313. 11:46technology hasn't been you know designed
314. 11:48to deal with uh and then that that's
315. 11:50where the concern is that have you been
316. 11:52practicing that muscle uh you know
317. 11:55whether it's cognitive muscle or or
318. 11:57physical activity. Have you been
319. 11:59actually training that are you ready to
320. 12:01react to that situation?
321. 12:03>> I can see a way chomping at the bit
322. 12:04there to get to respond.
323. 12:06>> Yeah, absolutely. Because I'd like to
324. 12:07follow it up with a counter question for
325. 12:09you Mark and and you Stephanie, when are
326. 12:12you going to be ready to god forbid you
327. 12:15need medical intervention, medical
328. 12:17advice? When are you going to be ready
329. 12:19to go to an AI agent and say, "I need my
330. 12:22gallbladder taken out. I've heard that
331. 12:24you are very, very good and Google's
332. 12:25done a great job with you, son, and so
333. 12:27forth. I trust myself to you, Mr. AI
334. 12:30agent. When are you going to be ready
335. 12:31for that?"
336. 12:32>> Never.
337. 12:32>> I can say, yeah, I can say very
338. 12:34confidently, [laughter]
339. 12:35absolutely never.
340. 12:36>> So, so here's the thing. When we look at
341. 12:37human interactions, I mean, certainly
342. 12:39applies in medicine, but I'd go as far
343. 12:40as all human interaction, it's
344. 12:42fundamentally based on trust or lack
345. 12:44thereof. Yeah. But let's talk about
346. 12:46trust. And for me as a children's brain
347. 12:49surgeon, day in day out, week in week
348. 12:52out, I have parents that come to our
349. 12:54hospital, parents that have sick
350. 12:57children, and parents that entrust you
351. 13:00with their child, potentially their most
352. 13:02beloved, their most prized
353. 13:06possession. And it's it's why do they do
354. 13:09that? Yes, they have read about you and
355. 13:11so on and so forth, but the ultimate
356. 13:13thing is trust. They look at you, they
357. 13:15talk to you, they place their trust in
358. 13:17you. So the responsibility lies with you
359. 13:20in my mind. You know, it's you can you
360. 13:22can never go back to those parents and
361. 13:24say, "Oh, by the way, I did a really
362. 13:25good operation. The AI agent didn't
363. 13:27really play ball today, so sorry, didn't
364. 13:29turn out quite well, but the operation
365. 13:31was good." That doesn't I mean, that's
366. 13:32just a nonsensical response, right? So,
367. 13:35you carry the ultimate responsibility.
368. 13:37You've been trusted with something and
369. 13:40medicine in particular, but as all human
370. 13:42intervention, it's based on trust and
371. 13:44that trust is fundamentally between
372. 13:45humans and humans.
373. 13:47>> AI can augment that. It can augment you
374. 13:49in your work,
375. 13:50>> but it cannot take over that role.
376. 13:52That's critical.
377. 13:53>> Would you get in a self-driving car?
378. 13:55>> Yeah, I've thought about that often
379. 13:57after at the end of a late shift when I
380. 13:59don't feel like driving home. Um the
381. 14:02short answer is yes because for me
382. 14:04that's a much more repetitive task and
383. 14:07the stats will show that in time they
384. 14:10are safer. They already show that but
385. 14:11you know so I'd be more comfortable with
386. 14:13a self-driving car when it comes to
387. 14:15surgery that level of care that level of
388. 14:17detail. I don't think so.
389. 14:19>> Do you think someone could appear more
390. 14:21highly skilled than they are because
391. 14:23they're working alongside AI?
392. 14:26>> Yes, it happens all the time. You look
393. 14:28around and people come to you with these
394. 14:29beautiful presentations and your
395. 14:32students often and then you ask them the
396. 14:33first question and they fall apart.
397. 14:35>> So those people they actually understand
398. 14:36a lot less than they think they
399. 14:38understand or at least they understand a
400. 14:39lot less than they are purporting to
401. 14:40understand.
402. 14:41>> What they present is just fantastic. But
403. 14:43when you have to dive into what's behind
404. 14:45the presentation there not the substance
405. 14:48is not quite there.
406. 14:49>> Okay.
407. 14:50>> You got to use it or you lose it right
408. 14:53with your brain
409. 14:54>> is just a muscle. As Tani just said it
410. 14:56is a muscle. You use it or you lose it.
411. 14:58It's as simple as that.
412. 14:59>> Yeah.
413. 14:59>> And so the choice is yours.
414. 15:01>> Tobani, do you think that listening to
415. 15:03Oasis, do you think that medicine
416. 15:04represents the clearest example of risks
417. 15:08around areas of of deskkilling?
418. 15:11>> Oh yeah. It's it's a very clear and and
419. 15:13really good example uh because that's
420. 15:15you know where we don't want things to
421. 15:17go wrong and and that's where that
422. 15:18interpersonal trust what he was talking
423. 15:20about is really important. uh but
424. 15:22obviously there are because of the the
425. 15:24kind of general purpose technology
426. 15:26nature of of the modern AI systems is
427. 15:28that you know they can do so many things
428. 15:30in different contexts. Though obviously
429. 15:33that concern is kind of in in in many
430. 15:36areas nowadays and something I wanted to
431. 15:38also point out I I think the acceptance
432. 15:42of the risk if you know if you hand over
433. 15:44something to a machine I think that has
434. 15:47a lot to do with the cultural and social
435. 15:49context where we live in but then you
436. 15:51know if once AI systems are so good that
437. 15:53they can be shown statistically to lead
438. 15:56into fewer errors than with human
439. 15:57surgeons that might change and it's I
440. 16:00guess the general
441. 16:02social, cultural, technological context
442. 16:03in that how much have we actually
443. 16:06immersed our lives with technologies and
444. 16:08automation that might change over time.
445. 16:10So it'll be an interesting thing to to
446. 16:13see and to think about is that you know
447. 16:15what are we willing to to hand over over
448. 16:18to to to machines when you know when
449. 16:21they start performing really well when
450. 16:23uh when they let's say they are better
451. 16:25or more trustworthy kindergarten
452. 16:27teachers than than humans you know than
453. 16:29fible humans. So uh but yeah I mean
454. 16:32right now uh we are still fortunately at
455. 16:35not not at that stage.
456. 16:37>> Is this about though things like trust
457. 16:39are also about you know what level of
458. 16:41accuracy do you need in a certain task
459. 16:43or in a certain job something that
460. 16:45you're commissioning. Um liability is
461. 16:48another thing. So maybe we're willing to
462. 16:50tolerate more risk for a PowerPoint
463. 16:52presentation if we're a management
464. 16:53consultant than we are from our doctor
465. 16:56where we want you know zero risk. Thank
466. 16:58you very much. Right. So there's that
467. 17:01question and I also wonder if there's a
468. 17:03question that would be useful for us to
469. 17:04think about in terms of when we
470. 17:06introduce AI into building our skills
471. 17:09and knowledge. Do we need children
472. 17:12working with AI right now or do we need
473. 17:14children reading books given that we
474. 17:15know literacy rates are really
475. 17:17struggling? Do we want children doing
476. 17:18mathematics the old-fashioned way first
477. 17:21and then they graduate into using AI at
478. 17:24a later date? did over at university or
479. 17:26for for junior doctors starting out. Is
480. 17:29there something valuable about learning
481. 17:31things the old way, the graft way, the
482. 17:33painstaking staking way? So, you develop
483. 17:35the feel to sense checks that you know
484. 17:38when AI is wrong. I don't know how you
485. 17:40develop that ability if you've never
486. 17:42worked without it.
487. 17:43>> Well, wait. I mean, as a neurosurgeon,
488. 17:46>> what does neuroscience tell us about how
489. 17:49the brain develops and if AI has any
490. 17:51effect on that development? this
491. 17:53discussion, this whole global interest
492. 17:55in intelligence is really telling. I
493. 17:57lecture quite a bit um medical students,
494. 18:00doctors and and uh what have you. You go
495. 18:02into a lecture theater and you say,
496. 18:04"Okay, how many people here have heard
497. 18:06about artificial intelligence?" Pretty
498. 18:08much everybody raises their hand up. I
499. 18:09mean there may be the odd bloodite who's
500. 18:12who's not, but pretty much everybody
501. 18:13100%. Then my next question is how many
502. 18:16people here have heard about organic
503. 18:18intelligence? And there is at most a
504. 18:20handful of hands that go up. And I find
505. 18:23that utterly fascinating. AI has been
506. 18:25around in the public domain for perhaps
507. 18:27the past 15 20 years. The human brain
508. 18:30organic intelligence has been in
509. 18:32development and us.
510. 18:36>> Intelligence 1.0 if you like.
511. 18:38>> Intelligence indeed. And the and it's
512. 18:40really imperative that we spend a lot
513. 18:42more time understanding intelligence per
514. 18:45se organic intelligence because a lot of
515. 18:48the questions that we're asking about AI
516. 18:50where it would be good where it you know
517. 18:52we need to watch it come from the
518. 18:54lessons we learned from organic
519. 18:55intelligence. So for example the human
520. 18:57brain what does the human brain do? What
521. 19:00you know we've got 86 billion neurons we
522. 19:02only need a handful to really breathe
523. 19:04and eat and walk around. What does the
524. 19:06rest 86 billion neurons do? Put in a
525. 19:09sentence, the human brain is a tool of
526. 19:11prediction. Period. It's a prediction
527. 19:12engine. That's what your brain does, my
528. 19:14brain, everybody's brain all the time.
529. 19:16Now, people who are that little bit
530. 19:19better at predicting the future are the
531. 19:21people we call lucky. There's nothing
532. 19:23godsend about being lucky. It's just
533. 19:25their algorithms that little bit better
534. 19:27refined. They can see things perhaps one
535. 19:29or two steps clearer than the rest of
536. 19:30us. The BBC will do the world a huge
537. 19:34service if we can have a program called
538. 19:37OI decoded, organic intelligence
539. 19:39decoded. That's my proposal.
540. 19:41>> That's getting commissioned next week.
541. 19:42Okay.
542. 19:43>> I'm taking it upstairs. We're taking
543. 19:44that one upstairs. Okay. I've got an
544. 19:45audience question now, and this is for
545. 19:47one for you, Stephanie. It's from John
546. 19:49Hin in Columbus, Ohio, and he says, "Is
547. 19:53AI using the Gillette razor marketing
548. 19:56strategy where they essentially give
549. 19:58away the razor for free, but make up
550. 20:02both cost and profit on selling the
551. 20:04blades?"
552. 20:05>> Yes. Um, a great question from someone
553. 20:08from the great state of Ohio. So, I'm
554. 20:09not surprised. I think that's probably
555. 20:11pretty accurate. Um, it's also like, you
556. 20:14know, the drug easing model. you know,
557. 20:15your first hits free and you get them
558. 20:17hooked and then they have to come back.
559. 20:19So the goal is you're looking for
560. 20:21stickiness when you are building a
561. 20:23technology or a tool. You want to give
562. 20:25it away to people. Let them play with
563. 20:27it. Let it let them see how much it
564. 20:28makes their lives better. And then if
565. 20:30you threaten to take it away, what
566. 20:32you're hoping is they will howl and
567. 20:34protest and then you say, "That's fine
568. 20:35because we have this lovely subscription
569. 20:37model."
570. 20:38>> That's where and then you know they're
571. 20:40using it for the rest of your lives.
572. 20:41What you want is that to be embedded in
573. 20:43their lives and then they can't live
574. 20:45without it. If you start to get
575. 20:47dependent on AI and you can't do your
576. 20:50work without it, you can't do school
577. 20:52without it. You don't know how to
578. 20:54navigate your emotions without it
579. 20:56because it's your therapist, it's your
580. 20:57girlfriend, it's your companion. That
581. 21:00starts to get quite dangerous, that kind
582. 21:02of dependency. Oasis, do you think that
583. 21:05AI could actually enhance critical
584. 21:08thinking rather than replace it?
585. 21:11>> I do. I do. No, I think I I you go back
586. 21:14to this. I think AI has the potential,
587. 21:17enormous potential in many different
588. 21:19ways, whichever ways you let it go. You
589. 21:21know, it has the potential to do us
590. 21:22enormous harm. There is no question of
591. 21:24that. No question. And I completely
592. 21:26understand the people who say, you know,
593. 21:28shut down all AI. We don't want an AI.
594. 21:30Now I understand their sentiment but we
595. 21:32need to be pragmatic. AI is here. It's
596. 21:34going to stay whether we like it or not.
597. 21:36Our best chance is to understand it and
598. 21:39to work iteratively with it and try and
599. 21:42stay at pace or ahead of it. That's the
600. 21:44optimal. The tasks that AI is really
601. 21:47good at. We should be using it for those
602. 21:49tasks. The repetitive tasks to augment
603. 21:51our way of thinking. It is our plan and
604. 21:54AI feeds into it. That is what we should
605. 21:57be doing. The risk that we run, as we've
606. 21:59just heard from Stephanie, is if we if
607. 22:02we outsource our creativity, which then
608. 22:05becomes a very very sort of fundamental
609. 22:08human trait, something that we learn
610. 22:10from a very young age. If we outsource
611. 22:12our creativity, that is where Stephanie
612. 22:15used the word dangerous. I'd use the
613. 22:16word sad. That is where we start to lose
614. 22:18our humanity. That is what makes us
615. 22:20human. We are special. All of us here,
616. 22:22all eight billion of us are special in
617. 22:24our own right. And we need to maintain
618. 22:26that. That is imperative.
619. 22:29>> Tali, where have you seen AI genuinely
620. 22:34improve human expertise?
621. 22:38>> That's a great question, Mark. Um, I'd
622. 22:40say personally, uh, in areas, you know,
623. 22:43there are some areas that are not your
624. 22:46core expertise. Like I'm I'm an
625. 22:47information systems researcher, but I'm
626. 22:50not an expert in tax law. And if I have
627. 22:51to understand something about tax or tax
628. 22:53deductions, um it's actually very
629. 22:56difficult to interpret the the tax
630. 22:59regulation that's available on tax
631. 23:00office website. Uh but I've noticed like
632. 23:03AI can be helpful in that. I can
633. 23:05actually upskill myself in tax by
634. 23:07talking to AI that makes it more
635. 23:09understandable and then I can go back
636. 23:11and check like okay does it actually
637. 23:12check out what the AI is saying. So in
638. 23:15that sense yes.
639. 23:16>> Yeah. I I mean I I I I I I wonder if if
640. 23:19people are using AI every day, if people
641. 23:22in the audience are using AI every day,
642. 23:24what practical habits do you think uh
643. 23:29could help preserve critical thinking?
644. 23:31Stephanie, what do you think about that?
645. 23:33>> I mean, I know some very heavy AI users,
646. 23:36you know, to the point where it's
647. 23:38because we forget it depends also on
648. 23:39what's your entry point. So for me for
649. 23:42instance when I use AI I try to only do
650. 23:44this on my computer and I have really
651. 23:46strict rules for myself but that's
652. 23:49because I've been studying this for a
653. 23:50really long time and I know and even
654. 23:52then you know I'm valuable I also get
655. 23:53tired and lazy. I try not to go to it
656. 23:56first but there are people who might be
657. 23:58going to it through their Amazon Echo.
658. 24:00So they might have a few of those around
659. 24:01their house and they're just it's just
660. 24:03there. It's in the the ether of their
661. 24:05home life and they and their children
662. 24:07are talking with it. Everybody's talking
663. 24:09with it and you just learn to kind of,
664. 24:11you know, when you're lazy, you just ask
665. 24:14the AI, what's the weather, what should
666. 24:16I cook, what should I wear to this
667. 24:18party, um, what's the fastest way for me
668. 24:20to get across London, you know, in 30
669. 24:22minutes, etc. And you stop actually just
670. 24:25realizing that you could have done that
671. 24:26heavy lifting yourself, you know. So,
672. 24:29I'm just saying there's that versus if
673. 24:31you're doing it on your phone, going in
674. 24:33through your computer, if you have sort
675. 24:34of rules for yourself about it, and I
676. 24:37understand it. There's no judgment here.
677. 24:38All of this was pushed out on us really
678. 24:40fast. Everybody's experimenting with it.
679. 24:42Nobody had best practice. But I think
680. 24:44now that we're seeing some of these
681. 24:45studies coming in like Tony and others
682. 24:48who are saying be really careful how
683. 24:49you're using it. Maybe we don't want
684. 24:51this in schools right away or maybe not
685. 24:52for six-year-olds. Maybe it's better for
686. 24:54when you're 16 and you're a bit older or
687. 24:56even at university. Yes, we want to
688. 24:58prepare you to go into work, but we
689. 25:00really need you doing your core skills
690. 25:02of literacy, numeracy, your creat your
691. 25:04creativity skills, constructive
692. 25:06thinking, critical thinking. How are we
693. 25:09going to really embed those first? And
694. 25:11that goes for us adults, too.
695. 25:13>> A better understanding of organic
696. 25:15intelligence then. Well, that is all we
697. 25:18have time for. Uh, a big thank you to
698. 25:21all of our guests. If you have any
699. 25:23thoughts, questions, or feedback on
700. 25:25today's program, do email us at
701. 25:27aidcodedbc.co.uk.
702. 25:29The QR code is on screen right now. Uh,
703. 25:33and that takes you straight to the AI
704. 25:34Decoded YouTube playlist. And every
705. 25:37episode is, of course, on the BBC i
706. 25:40Player. We'll see you next time.
707. 25:43[music]