WEBVTT - The Story: Embracing AI and All Life’s Uncertainties w/ David Spiegelhalter 0:00:14.800 --> 0:00:18.520 Welcome to tech Stuff. This is the story. Each week 0:00:18.560 --> 0:00:21.599 on Wednesdays, we bring you an in depth conversation with 0:00:21.720 --> 0:00:23.880 someone who has a front row seat to the most 0:00:23.880 --> 0:00:31.720 fascinating things happening in tech today. A conversation with David Spiegelholter, 0:00:32.360 --> 0:00:35.760 a professor of statistics at Cambridge an author of the 0:00:35.880 --> 0:00:40.560 Art of Uncertainty, How to Navigate Chance, Ignorance, Risk and Luck. 0:00:41.400 --> 0:00:46.600 Spiegelhlter has devoted his life to understanding uncertainty. After all, 0:00:46.720 --> 0:00:49.320 it's one of the most uncomfortable aspects of being human, 0:00:49.960 --> 0:00:54.120 particularly when it comes to our health. Since the nineteen seventies, 0:00:54.200 --> 0:00:57.760 Spiegelhulter has worked on algorithms to assist clinicians and patients 0:00:58.120 --> 0:01:01.760 make better decisions about what treatment options to take for cancer, 0:01:02.640 --> 0:01:05.840 and he has a deeply personal understanding of the topic. 0:01:06.640 --> 0:01:09.240 In nineteen ninety seven, he lost his son Danny to 0:01:09.280 --> 0:01:12.680 cancer at the age of just five, and the epigraph 0:01:12.800 --> 0:01:15.959 to David's book is a quote from the Bible. The 0:01:16.080 --> 0:01:18.720 race is not to the swift, nor the battle to 0:01:18.760 --> 0:01:22.120 the strong, nor bread to the wise, nor riches to 0:01:22.200 --> 0:01:25.960 men of understanding, nor favor to men of skill. But 0:01:26.160 --> 0:01:30.360 time and chance happened to them all of course, with 0:01:30.440 --> 0:01:33.280 the explosion of AI, we now have better and better 0:01:33.360 --> 0:01:36.840 tools to help us understand the world and make informed decisions. 0:01:37.360 --> 0:01:41.200 And in fact, Spiegelhalter was an early pioneer of the technology. 0:01:41.840 --> 0:01:45.679 So that's why I decided to start our conversation. We 0:01:45.800 --> 0:01:50.120 live in this extraordinary moment where technology seems to be 0:01:50.160 --> 0:01:53.120 giving us more of a view around the corner into 0:01:53.160 --> 0:01:56.960 the future than it ever has. How has the march 0:01:56.960 --> 0:02:00.640 of technology impacted your work and your understanding these topics 0:02:00.720 --> 0:02:01.360 over your career? 0:02:02.080 --> 0:02:04.080 Oh a huge amount. I mean, we needn't get into 0:02:04.080 --> 0:02:07.400 the whole Asian methodology, but that's what I was interested in, 0:02:07.440 --> 0:02:09.040 and we couldn't do it because you just couldn't do 0:02:09.080 --> 0:02:12.600 the calculations. You couldn't do the maths. But instead of 0:02:12.639 --> 0:02:15.960 trying to do the maths, you just used brute computing 0:02:16.080 --> 0:02:20.160 force to simulate millions of different possibilities and look at 0:02:20.200 --> 0:02:24.600 their distribution and the algorithms we knew would converge to 0:02:24.680 --> 0:02:27.440 the correct answer if they ran long enough. You had 0:02:27.440 --> 0:02:30.639 to wait till nineteen ninety or so. Just before that 0:02:30.800 --> 0:02:35.440 technology that ability to do such huge simulation exercises was 0:02:35.480 --> 0:02:38.920 on everyone's desktop, and then there was an explosion and 0:02:38.960 --> 0:02:41.360 a complete change in the way statistics was done. Up 0:02:41.360 --> 0:02:44.320 to then people did clever maths and then programmed that 0:02:44.480 --> 0:02:47.200 in and it changed into no, we don't have to 0:02:47.200 --> 0:02:50.079 do any maths. We just have to program in the problem, 0:02:50.160 --> 0:02:53.160 the model that we're trying to solve, present the data 0:02:53.240 --> 0:02:56.960 to it, and send it off. And wait. But what 0:02:57.080 --> 0:02:59.560 I think you might be starting to allude to, which 0:02:59.560 --> 0:03:02.680 I'm sure get onto, is the role of AI. Our 0:03:02.760 --> 0:03:05.200 AI is already changing my I used a lot in 0:03:05.240 --> 0:03:10.040 writing the book, both in the researching and summarizing of 0:03:10.120 --> 0:03:12.760 literature and of course in the coding all the time. 0:03:13.160 --> 0:03:15.640 You know, I always rewrote everything, but I used it 0:03:15.680 --> 0:03:17.640 a lot. I use it all the time in my 0:03:17.800 --> 0:03:21.040 daily word, daily life. But actually, will it be able 0:03:21.080 --> 0:03:25.240 to make predictions? And I am rather skeptical about claims 0:03:25.720 --> 0:03:27.560 both that are sort of you know what, you might 0:03:27.560 --> 0:03:30.160 call it a global level, or a social level, or 0:03:30.240 --> 0:03:33.520 even at a personal level, about our health, about the 0:03:33.560 --> 0:03:35.559 ability of AI to make predictions. 0:03:36.200 --> 0:03:39.080 Your book has the epigraph from the Bible. How did 0:03:39.120 --> 0:03:40.200 you come up with that? Oh? 0:03:40.280 --> 0:03:44.040 By using AI to ask for quotes that use chants 0:03:44.080 --> 0:03:44.640 and things like that. 0:03:44.760 --> 0:03:45.320 Is that true? 0:03:45.440 --> 0:03:47.480 Yeah, I'd actually for that one, I knew that one, 0:03:47.520 --> 0:03:49.240 but otherwise so yeah, I use AI. 0:03:49.320 --> 0:03:51.200 Well why that one? Why was that the first one 0:03:51.240 --> 0:03:52.040 that you used? Oh? 0:03:52.120 --> 0:03:55.600 I think because the whole book, and especially if we're 0:03:55.640 --> 0:03:58.480 the first chapter, which is about my grandfather, was intended 0:03:58.520 --> 0:04:01.040 to give the idea of the utter lack of control 0:04:01.400 --> 0:04:04.440 we have in our lives, and we have an illusion 0:04:04.520 --> 0:04:08.040 of control, which I think actually is not helpful. I 0:04:08.080 --> 0:04:10.880 think to realize that how little control we do have 0:04:10.960 --> 0:04:12.960 in our lives, how much of what happens to us 0:04:13.400 --> 0:04:16.080 is what, for want of a better word, we might 0:04:16.120 --> 0:04:18.720 call chance. In other words, events that will happen to 0:04:18.800 --> 0:04:21.440 us that were unpredictable and that you know, and that 0:04:21.520 --> 0:04:24.479 we had no control over. I think that's rather important 0:04:24.480 --> 0:04:27.760 to realize that. And because the word that appears in 0:04:27.800 --> 0:04:30.520 the book more than almost any other is humility. There's 0:04:30.520 --> 0:04:33.080 almost no mention of the word rationality in the book. 0:04:33.120 --> 0:04:35.960 This is not a book about being rational, It's a 0:04:36.000 --> 0:04:37.080 book about being humble. 0:04:37.480 --> 0:04:39.400 He mentioned your grandfather, and in the book you talk 0:04:39.440 --> 0:04:42.560 about how he survived various battles in World War One. 0:04:43.160 --> 0:04:45.520 You'll talk about your mother being captured by parents in 0:04:45.520 --> 0:04:49.800 the South China. See exactly and they're making it to 0:04:49.839 --> 0:04:52.000 the UK, where you should meet your father. 0:04:52.080 --> 0:04:54.080 Yeah, who then nearly died in the Second World War 0:04:54.080 --> 0:04:54.440 as well. 0:04:54.600 --> 0:04:54.920 He did. 0:04:55.040 --> 0:04:57.120 Yeah, he got TB. I mean it was an illness 0:04:57.160 --> 0:04:59.480 and he was in the hospital for weeks, and he 0:04:59.600 --> 0:05:02.360 was a vat cuated from Jerusalem as he was there 0:05:02.360 --> 0:05:04.520 when he heard the Saint David's Hotel being blown up. 0:05:04.920 --> 0:05:06.920 So then you know, there so much could have happened 0:05:06.920 --> 0:05:09.279 to both of them. Then, as I mentioned in my book, 0:05:09.320 --> 0:05:13.320 the biggest chance event of all is my conception. It's 0:05:13.360 --> 0:05:17.600 not just me, everyone's conceptions. An extraordinarily unlikely of it 0:05:17.720 --> 0:05:21.039 so easily could not have happened. So me realizing that 0:05:21.120 --> 0:05:24.640 and researching the situation of my conception, it really made me. 0:05:24.960 --> 0:05:26.960 You know, it changed my whole attitude to life. In fact, 0:05:27.160 --> 0:05:29.560 it really did. It made me think, God, I'm here 0:05:29.680 --> 0:05:33.599 just by total chants in what I call these micro 0:05:33.760 --> 0:05:38.279 contingencies that just accumulated, and here I am. And so 0:05:38.400 --> 0:05:41.800 the idea somehow that I'm you know, on the earth 0:05:41.800 --> 0:05:44.279 for a purpose, or I'm you know, in any way 0:05:44.360 --> 0:05:48.280 special I find is now for me a complete illusion. 0:05:48.760 --> 0:05:53.360 When did you start getting interested in this relationship between poverty, statistics, 0:05:53.400 --> 0:05:54.160 and medicine. 0:05:54.279 --> 0:05:58.040 Yeah, I was interested in the mathematical aspects of statistics particularly, 0:05:58.040 --> 0:05:59.960 But then there was a job going and the funny 0:06:00.080 --> 0:06:03.280 the job going was in nineteen seventy eight, and it 0:06:03.320 --> 0:06:07.479 was to work on what was then called computer aided diagnosis. 0:06:07.839 --> 0:06:11.200 Well now we've called it AI. So nineteen seventy eight 0:06:12.000 --> 0:06:15.320 was using some basic statistical algorithms what is now called 0:06:15.400 --> 0:06:19.280 naive base simple algorithms. It's still around and used as 0:06:19.279 --> 0:06:23.039 a very basic machine learning algorithm, for example in a 0:06:23.080 --> 0:06:25.440 spam filtering or whatever. And we were doing that in 0:06:25.480 --> 0:06:28.239 the late seventies. So it's one of the first jobs 0:06:28.279 --> 0:06:32.880 to work on algorithms in medicine for both diagnosis and prognosis. 0:06:32.920 --> 0:06:35.400 We were working on the likelihood of someone with head 0:06:35.440 --> 0:06:38.880 injury surviving and so on. And because the computers you 0:06:38.880 --> 0:06:40.960 could even carry them around a thing. They were sat 0:06:41.000 --> 0:06:43.920 in the corner with a huge, great machine with a keyboards. 0:06:44.000 --> 0:06:47.600 It's incredibly clumsy to use, but we were doing that. 0:06:47.640 --> 0:06:49.520 And then in the early eighties I was working on 0:06:49.560 --> 0:06:53.760 more algorithms. Then we got into developments in AI and 0:06:53.800 --> 0:06:56.080 so on. So you know, this stuff is not new. 0:06:56.279 --> 0:07:00.480 It's been around for ages, what was predict addicted. It's 0:07:00.480 --> 0:07:04.560 still going because that's an algorithm for predicting the survival 0:07:04.680 --> 0:07:07.600 of women with breast cancer and men with prostate cancer, 0:07:07.640 --> 0:07:13.080 still available, hugely, widely used. It's a very good statistical algum, 0:07:13.240 --> 0:07:18.520 a regression algorithm. Of course, in practice, any actual clinician 0:07:18.640 --> 0:07:22.280 making a decision with the patient would use much more 0:07:22.280 --> 0:07:25.200 personal information that they might have the patient, because you know, 0:07:25.480 --> 0:07:28.480 for example, physical status doesn't go into the algorithm, and 0:07:28.560 --> 0:07:31.640 yet that might be you know, someone's basic underlying health 0:07:31.760 --> 0:07:34.640 might be incredibly important. So that's when we wrote the 0:07:34.640 --> 0:07:37.840 interface for a predict we try to emphasize not say 0:07:37.880 --> 0:07:41.000 this is the risk of this patient. It's just what 0:07:41.040 --> 0:07:43.160 would expect to one hundred happen to one hundred people 0:07:43.400 --> 0:07:46.600 who ticked the boxes that she did or he did. 0:07:46.960 --> 0:07:51.440 But actually people have tried different, more sophisticated machine learning 0:07:51.800 --> 0:07:54.320 methods and they don't make much difference. And so it's 0:07:54.360 --> 0:07:55.880 about as good as you can do. I think with 0:07:56.000 --> 0:07:58.920 the data that is available, you could always do better 0:07:58.960 --> 0:08:02.600 by collecting more data to and having a bigger, better database. 0:08:02.960 --> 0:08:05.960 It's going to be marginal marginal benefits just from using 0:08:06.000 --> 0:08:08.880 AI with the same data so the real you know, 0:08:08.960 --> 0:08:11.400 benefit in the future, of course, is just by having 0:08:11.480 --> 0:08:12.200 better data. 0:08:12.640 --> 0:08:14.400 So where did you say a few moments ago that 0:08:14.440 --> 0:08:16.920 you doubted that AI would be a useful. 0:08:16.720 --> 0:08:18.720 Oh well, I mean it's going to be marginal, marginal 0:08:18.760 --> 0:08:21.640 benefits just from using AI with the same data. 0:08:22.080 --> 0:08:24.520 What about other things like drug discovery. 0:08:24.080 --> 0:08:27.000 Or oh yeah all that, Yeah that's very important. No, no, 0:08:27.040 --> 0:08:29.960 then it's going to be great, huge, huge benefits there. 0:08:30.120 --> 0:08:32.120 Now I'm talking about predicting what's going to happen to 0:08:32.200 --> 0:08:34.600 me in the future, because I've got prospect cancer so 0:08:34.800 --> 0:08:37.719 I am quite interested in this. And of course when 0:08:37.720 --> 0:08:39.920 I got it, I looked at all the algorithms and 0:08:39.960 --> 0:08:42.840 which were sort of helpful, but they're very broad. All 0:08:42.880 --> 0:08:45.800 they do really is tell you what we would expect 0:08:45.840 --> 0:08:47.880 to happen to It is one hundred people who ticked 0:08:47.920 --> 0:08:52.240 the boxes you've ticked, and of course everyone is so different, 0:08:52.480 --> 0:08:55.920 Everyone's so different, so I know that that will give 0:08:55.960 --> 0:08:58.400 me a broad figure, but it's only a ballpark figure. 0:08:58.400 --> 0:09:01.640 It's still very useful until of my tenure survival, but 0:09:01.920 --> 0:09:06.160 one that I know could be changed by just having 0:09:06.160 --> 0:09:06.840 more information. 0:09:07.600 --> 0:09:10.440 So you spent your career trying to help doctors and 0:09:10.520 --> 0:09:14.000 patients make better decisions about what to do when a 0:09:14.040 --> 0:09:17.520 patient gets sick. But you've also lived this with your 0:09:17.520 --> 0:09:21.760 son Danny, this experience of how to make medical decisions 0:09:22.240 --> 0:09:24.560 in the face of horribly serious illness. 0:09:25.280 --> 0:09:27.599 Oh, that's interesting because I do think about that. We 0:09:28.120 --> 0:09:30.920 did some decisions and I'm not sure, you know, maybe 0:09:31.200 --> 0:09:32.839 we always say, well, maybe if we had taken him 0:09:32.840 --> 0:09:35.320 to these, to Canada or something, and that we could 0:09:35.320 --> 0:09:38.760 have got a different therapy. In a way, that's one 0:09:38.760 --> 0:09:42.200 thing I prefer not to dwell on too much, because 0:09:43.679 --> 0:09:46.640 you know, you don't know. But it has made me 0:09:46.800 --> 0:09:50.960 very aware of the importance of making informed medical decisions, 0:09:51.360 --> 0:09:53.440 and that's what I with my team we worked on 0:09:53.480 --> 0:09:56.720 those on providing decision aids, not in any way to 0:09:56.840 --> 0:10:00.200 encourage people to make any particular decision, but just as 0:10:00.240 --> 0:10:02.840 I mentioned in the book, I don't believe that decision 0:10:02.920 --> 0:10:05.360 theory and all the advances that have been done in 0:10:05.400 --> 0:10:07.440 that can never tell you what to do, because it 0:10:07.520 --> 0:10:10.640 assumes that you know all the possible outcomes, you know 0:10:10.720 --> 0:10:14.000 all the possible options, you know all the probabilities and 0:10:14.080 --> 0:10:18.720 the values, and of course this is totally infeasible apart 0:10:18.720 --> 0:10:21.600 from the simplest sort of gambling type examples. You never 0:10:21.679 --> 0:10:23.520 know any of these things. You never know how you're 0:10:23.520 --> 0:10:25.680 going to feel in the future if something happens, and 0:10:25.720 --> 0:10:28.960 so on. So it's impossible to be rational in those 0:10:29.000 --> 0:10:32.760 situations and use decision theory. But I think it's really 0:10:32.800 --> 0:10:36.600 helpful to try to examine the problem to face up 0:10:36.640 --> 0:10:38.840 to a decision has to be made. One of the 0:10:38.840 --> 0:10:42.640 biggest problems about decisions is that people don't actually make decisions. 0:10:43.120 --> 0:10:45.720 I don't you just go. You just find yourself going 0:10:45.760 --> 0:10:48.640 down a path, and you never just stop and say, 0:10:48.640 --> 0:10:51.440 this is a decision point. There is a branch in 0:10:51.480 --> 0:10:53.360 the road. We have to choose which way to go, 0:10:53.640 --> 0:10:57.080 and sometimes you can recognize those points, but they're few 0:10:57.080 --> 0:11:00.600 and far between, and so would I love to encourage 0:11:00.600 --> 0:11:04.280 people to actually have much more of those decision points. 0:11:04.520 --> 0:11:05.920 This is the time we have to make a decision. 0:11:05.920 --> 0:11:09.200 These are the possibilities, the benefits and harms of the 0:11:09.240 --> 0:11:11.520 options that face you. We're not going to tell you 0:11:11.559 --> 0:11:13.600 what to do. We might be able to put some 0:11:13.720 --> 0:11:17.480 rough probabilities. For example, for women in breast cancer, we've 0:11:17.520 --> 0:11:20.320 got such a lot of data we can actually produce 0:11:20.400 --> 0:11:23.000 reasonably good ten year survival rates and what the benefit 0:11:23.040 --> 0:11:26.559 would be if you had chemotherapy. So, for example, in Cambridge, 0:11:27.080 --> 0:11:31.040 unless you your absolute survival benefit is going to go 0:11:31.160 --> 0:11:34.480 up by three percent with chemotherapy, they don't recommend chema 0:11:34.559 --> 0:11:38.440 therapy because that means essentially that of all the people 0:11:38.480 --> 0:11:42.000 that give chemotherapy, out of thirty people, only one will 0:11:42.040 --> 0:11:45.840 benefit after ten years. Only you get one extra ten 0:11:45.920 --> 0:11:49.240 year survivor for thirty people being given to chemotherapy, which 0:11:49.280 --> 0:11:53.120 can have a really awful effect on people. And by 0:11:53.160 --> 0:11:55.360 producing those numbers you can get a feeling that well, 0:11:55.480 --> 0:11:58.040 you've got to have a reasonable benefit in order to 0:11:58.440 --> 0:12:01.800 take the hit of the tree. So in those situations, 0:12:01.840 --> 0:12:04.440 I think it's really good. You can't do it this 0:12:04.600 --> 0:12:08.120 exactly to some idea of what the benefits might be. 0:12:08.559 --> 0:12:11.200 But on the whole, you know, it's difficult to do 0:12:11.280 --> 0:12:15.240 that in situations where you havn't got all that data 0:12:15.280 --> 0:12:18.240 and all that analysis, all that tech behind. 0:12:17.880 --> 0:12:27.480 It coming up. Is it possible to predict murder? Stay 0:12:27.480 --> 0:12:39.760 with us? Well, it just is remarkable to think that 0:12:40.000 --> 0:12:44.440 kind of Hiding underneath all of these numbers and statistics 0:12:44.480 --> 0:12:48.360 and maths are so many life and death decisions. And 0:12:48.400 --> 0:12:49.960 the other thing I wanted to ask you about was 0:12:50.280 --> 0:12:54.360 your work on one of the most prolific serial killer 0:12:54.400 --> 0:12:56.880 cases of all time, the Herald Shipment case. Just for 0:12:56.920 --> 0:13:00.960 a US audience, can you explain that case and what 0:13:00.960 --> 0:13:01.800 your work on it was. 0:13:02.760 --> 0:13:05.880 Harold Shepman was a family doctor who, over a twenty 0:13:05.960 --> 0:13:08.920 year period, murdered at least two hundred and fifty of 0:13:08.920 --> 0:13:13.199 his patients and possibly up to four hundred without being caught, 0:13:13.280 --> 0:13:17.760 of course, until he finally, rather stupidly forged, rather badly, 0:13:17.880 --> 0:13:20.040 did a rather bad forgery of a will in order 0:13:20.080 --> 0:13:24.000 to inherit some money. Absolute madness. And it was this 0:13:24.040 --> 0:13:26.719 a woman whose daughter was a solicitor and looked at 0:13:26.720 --> 0:13:30.520 this willness, it just didn't believe it. And so suspicions rose, 0:13:30.600 --> 0:13:35.760 and finally he was arrested and they exhumed the last 0:13:35.840 --> 0:13:40.079 fifteen patients that had died, and they all had incredibly 0:13:40.160 --> 0:13:44.360 high levels of diamorphine heroin essentially in their bloodstream. I mean, 0:13:44.360 --> 0:13:46.080 he only got away with it because there were never 0:13:46.120 --> 0:13:50.840 any post mortems. There's many old people and they liked him. 0:13:50.920 --> 0:13:53.200 He was a very trusted family doctor for many people. 0:13:54.080 --> 0:13:57.040 He did a lot of home visits, and that's of 0:13:57.040 --> 0:14:01.000 course when he murdered people. So when when someone went 0:14:01.040 --> 0:14:03.720 back and looked at all the certificates of the time 0:14:03.760 --> 0:14:07.400 of death, for most people, people died all times of 0:14:07.400 --> 0:14:09.400 the day and night, the sort of uniform distribution of 0:14:09.480 --> 0:14:12.000 the twenty four hours, Harold Shipman's deaths had a great, 0:14:12.200 --> 0:14:16.280 huge spike between around one to three in the afternoon 0:14:17.040 --> 0:14:18.360 when he did his home visits. 0:14:18.559 --> 0:14:20.640 And what was your involvement personally with the case. 0:14:21.400 --> 0:14:24.560 There was a public inquiry because the families quite really 0:14:24.560 --> 0:14:26.200 on other people ask how do you get away with 0:14:26.240 --> 0:14:29.600 it for so long? It's an absolute scandal, And the 0:14:29.800 --> 0:14:33.880 public inquiry, I think very sensibly brought in quite a 0:14:33.880 --> 0:14:37.600 substantial team of statisticians to look at the data, which 0:14:37.920 --> 0:14:41.000 like the time of death data, but also the deaths 0:14:41.040 --> 0:14:44.000 of all his patients when they had occurred, how it 0:14:44.040 --> 0:14:47.120 compared with other doctors, how many would have been expected, 0:14:47.360 --> 0:14:51.720 compared with how many ratually observed. And we used sort 0:14:51.760 --> 0:14:56.000 of fairly standard industrial quality control methods to work out 0:14:56.040 --> 0:14:59.760 when you could have spotted with considerable confidence when something 0:14:59.800 --> 0:15:03.000 al was going on. So it's like industrial quality control 0:15:03.040 --> 0:15:05.280 methods spot when a production line is going out a 0:15:05.280 --> 0:15:08.280 bit out of kilter. They would have been developed over decades, 0:15:08.480 --> 0:15:11.040 and we used those for his death rates and worked 0:15:11.080 --> 0:15:14.040 out he could have been caught after about forty deaths, 0:15:14.440 --> 0:15:17.840 or he could have been identified as being odd. In 0:15:17.880 --> 0:15:21.640 other words, someone could have done an investigation. Now, Shipman, 0:15:21.720 --> 0:15:24.840 when the algorithm that we developed was applied to a 0:15:24.960 --> 0:15:29.240 thousand other gps without their knowledge, there were six who 0:15:29.800 --> 0:15:31.080 as bad as Shipman. 0:15:31.040 --> 0:15:32.520 I used many deaths on their watch. 0:15:32.720 --> 0:15:37.840 Exactly why do you think that was? They were really 0:15:37.880 --> 0:15:42.440 good gps who were working in retirement communities and who 0:15:42.480 --> 0:15:45.000 were enabling their patients to die at home rather than 0:15:45.000 --> 0:15:48.360 going to hospital by being really good caring gps, and 0:15:48.400 --> 0:15:52.480 so they were signing a lot of death certificates. And 0:15:52.680 --> 0:15:54.640 these were really good people, but they had very high 0:15:54.640 --> 0:15:57.240 death rates. So I used this as an example all 0:15:57.280 --> 0:16:01.240 the time about how statistics is about correlation not causation. 0:16:01.760 --> 0:16:04.200 You know, if someone has got a high death rate, 0:16:04.720 --> 0:16:08.280 it's an indication that someone perhaps should look at the data, 0:16:08.440 --> 0:16:10.680 but you cannot conclude the cause for that. 0:16:10.960 --> 0:16:13.080 Well, just in the last few weeks, the UK has 0:16:13.120 --> 0:16:17.800 announced an algorithm to predict the likelihood of committing murder. 0:16:18.200 --> 0:16:21.400 Well, those algorithms have been around for ages, but the 0:16:21.560 --> 0:16:24.440 chance of predicting someone, all you'll do is find a 0:16:24.520 --> 0:16:26.440 small change in odds. You're never going to be able 0:16:26.440 --> 0:16:30.160 to predict an event like that. At an individual level. 0:16:30.200 --> 0:16:32.480 You'll be able to predict someone's at someone it's slightly 0:16:32.560 --> 0:16:36.600 increased risk. But there's so much puff befind these algorithms 0:16:36.640 --> 0:16:39.280 to make it. You know, they get headlines, but actually 0:16:39.360 --> 0:16:42.640 I'm deeply skeptical about their actual ability and certainly to 0:16:42.800 --> 0:16:44.600 predict events like murders. 0:16:45.080 --> 0:16:47.120 You spend quite a bit of time working on public 0:16:47.160 --> 0:16:50.560 inquiries and informing a public about various issues, and I 0:16:50.600 --> 0:16:52.440 think you have one of the most interesting, if not 0:16:52.520 --> 0:16:56.400 the most interesting title in British academia Professor for the 0:16:56.440 --> 0:16:59.240 Public Understanding of Risk. Can you talk a little bit 0:16:59.280 --> 0:17:00.200 about what that means? 0:17:00.800 --> 0:17:04.040 Yeah, I think I am the one and only Professor 0:17:04.040 --> 0:17:07.000 for the Public Understanding of Risk because after I retired 0:17:07.040 --> 0:17:11.080 they renamed it when the next person got the funding. 0:17:11.160 --> 0:17:14.800 So this was a fascinating I'd been an academic and 0:17:15.160 --> 0:17:18.160 was doing okay, it'd got a good reputation, but fancied 0:17:18.160 --> 0:17:21.800 a change in direction from the normal business of writing 0:17:21.920 --> 0:17:26.639 papers and all that stuff, and then a philanthropist David Harding, 0:17:26.680 --> 0:17:29.959 a hedge fund manager, wanted to endow a chair in 0:17:30.080 --> 0:17:33.560 Cambridge that was to do with the improving the way 0:17:33.600 --> 0:17:37.359 that statistics and risk was discussed in society because he 0:17:37.400 --> 0:17:39.240 got so fed up with all the stories in the 0:17:39.280 --> 0:17:42.760 news and all the misunderstandings, and so he paid for 0:17:42.840 --> 0:17:45.080 this chair. And if you gave three and a half 0:17:45.160 --> 0:17:47.240 million pounds the University of Cambridge, you could have a 0:17:47.320 --> 0:17:49.919 chair of absolutely anything. 0:17:50.040 --> 0:17:51.520 And he had the good grace not to name it 0:17:51.560 --> 0:17:52.720 after himself. 0:17:53.040 --> 0:17:55.640 Exactly why it was the Winton. It was the Winton 0:17:55.720 --> 0:17:58.199 Professor for the Public Understanding of risk, so which was 0:17:58.359 --> 0:18:01.080 just fine because he was very good. I always give 0:18:01.359 --> 0:18:03.960 my career advice for young people now is to say, 0:18:04.040 --> 0:18:06.680 find a billionaire and get him to give you lots 0:18:06.680 --> 0:18:08.960 of money to do what you feel like, because he 0:18:09.080 --> 0:18:11.680 just gave the money and then completely hands off. 0:18:11.760 --> 0:18:17.199 In their capacity. What was the biggest misunderstanding you encountered 0:18:17.240 --> 0:18:18.720 about how the public understand risk? 0:18:19.720 --> 0:18:23.560 Oh goodness, that's so difficult. I mean you could the 0:18:23.600 --> 0:18:27.440 absolutely standard one, of course, which the media don't help 0:18:27.480 --> 0:18:30.679 with is the difference between absolute and relative risk. So 0:18:31.000 --> 0:18:33.320 you know, the media stories are always full of oh, well, 0:18:33.320 --> 0:18:36.160 if you eat meat, it's going to increase your risk 0:18:36.200 --> 0:18:39.239 of bowel cancer by twenty percent or so on. And 0:18:39.280 --> 0:18:41.240 that's a relative risk. And I think it's actually true 0:18:41.240 --> 0:18:44.440 that actually red meat is some process meat in particular, 0:18:44.520 --> 0:18:47.639 is associated with an increased risk of bow cancer. And 0:18:47.680 --> 0:18:51.800 that's what gets in the headlines increase risk. I've got lovely, 0:18:52.080 --> 0:18:54.679 you know, headlines of the killer bacon soandwich and this 0:18:54.720 --> 0:18:57.600 sort of thing. But when you actually translated, and I 0:18:57.680 --> 0:19:00.320 talk about this all the time to schools audiences, when 0:19:00.359 --> 0:19:02.560 they hear a story like this, they want to know, well, 0:19:02.800 --> 0:19:04.520 you know, is this the big number? Do we care 0:19:04.560 --> 0:19:06.520 about this? And to know that, you have to know 0:19:06.600 --> 0:19:09.880 twenty percent of what, in other words, the baseline risk 0:19:10.080 --> 0:19:12.399 of which there is a twenty percent increase. Now, the 0:19:12.400 --> 0:19:16.520 baseline risk of getting boo cancer, sadly is about six percent, 0:19:16.560 --> 0:19:19.240 about one in sixteen will get it during our lifetime, sadly, 0:19:19.640 --> 0:19:24.120 and a twenty percent increase over those six percentage points 0:19:24.880 --> 0:19:28.080 is about seven percentage points. So that's out of one 0:19:28.160 --> 0:19:31.280 hundred people eating a bacon sandwich every single day of 0:19:31.320 --> 0:19:35.360 their lives, one extra will get bow cancer because of that. 0:19:35.960 --> 0:19:39.640 And that's a complete different way of reframing the story 0:19:40.119 --> 0:19:43.639 to make it look frankly fairly reassuring, especially if you 0:19:43.720 --> 0:19:47.960 like bacon savwiches. So it's a great example of this 0:19:48.400 --> 0:19:52.199 difference between relative risks and absolute risk because percentage, the 0:19:52.240 --> 0:19:55.000 word percentage is used for both. It's just in one 0:19:55.040 --> 0:19:57.520 you're talking about a percentage increase and the other talking 0:19:57.560 --> 0:19:58.720 about percentage points. 0:20:01.119 --> 0:20:04.159 When we come back, we break down the probability that 0:20:04.240 --> 0:20:17.560 AI could lead to human extinction. Stay with us. When 0:20:17.640 --> 0:20:21.200 the consumer internet boomed in the late nineties early two 0:20:21.240 --> 0:20:24.879 thousands and Google came along, you could either click Google 0:20:24.960 --> 0:20:29.080 Search or I'm feeling lucky and I'm feeling Lucky would 0:20:29.119 --> 0:20:32.560 bypass the search results and take you directly to a website. 0:20:32.640 --> 0:20:34.679 And this is basically a way for Google to flex 0:20:34.720 --> 0:20:36.720 and say like, this is how incredibly good we are 0:20:36.760 --> 0:20:39.320 in search. And they've since abandoned the I'm feeling a 0:20:39.359 --> 0:20:42.879 Lucky button, but actually, in parallel, the whole Internet in 0:20:42.920 --> 0:20:45.280 the last two or three years has become an I'm 0:20:45.280 --> 0:20:47.800 feeling Lucky engine in the sense that you now get 0:20:48.240 --> 0:20:51.440 a generative AI response rather than a selection of links 0:20:51.480 --> 0:20:53.760 to follow, or at least you get both. How do 0:20:53.880 --> 0:20:59.920 you see that incredible cultural shift of sort of outsourcing information, 0:21:00.080 --> 0:21:03.439 summarization and predictions to large language models. 0:21:04.200 --> 0:21:07.320 I think it's great. I'm a real fan of the 0:21:07.359 --> 0:21:09.800 AI summary. So as long as you know, just like 0:21:10.000 --> 0:21:13.359 using any large language model, you have to grasp the 0:21:13.359 --> 0:21:15.679 fact that it doesn't know anything at all. You know, 0:21:15.720 --> 0:21:18.640 it is. All it does is string words together and 0:21:18.880 --> 0:21:21.840 comes up with something that sounds plausible. Now maybe in fact, 0:21:22.240 --> 0:21:24.760 as we all know, it comes up if it talks 0:21:24.800 --> 0:21:27.240 about facts, it can be deeply wrong and say all 0:21:27.240 --> 0:21:29.760 sorts of things that are just incorrect. So it has 0:21:29.840 --> 0:21:32.680 to be taken with a huge pinch of salt when 0:21:32.720 --> 0:21:36.840 it's saying anything factual. When it's summarizing an argument, or 0:21:36.880 --> 0:21:39.960 perhaps you know, with a discussion on a topic, I 0:21:39.960 --> 0:21:41.960 think it can be enormously helpful. I mean, if you 0:21:42.119 --> 0:21:44.960