Transcription
[Music] Hi everybody. I'm Nicolola Tangan, the CEO of the Norwegian Sovereign Wealth Fund. And today I'm in really good company with Sir David Spiegel. Well, I would say he's the world's best statistician and for sure the best communicator of risk that I've ever seen. Written lots of fantastic books and is particularly known for making statistics, which is uh difficult, uh, really accessible for most people. So, big thanks for joining us.
So David, no, great, great pleasure to be here. Having worked with risks, what, what's the main thing that you have learned about human psychology?
Well, first of all, that I'm not a psychologist. So, you know, I am a statistician, but I, I have done my best to learn from psychologists that I have worked with and I suppose just from observing how people react to un-risk and uncertainty. I tend to think of the broader idea of uncertainty rather than just risk. Everything to do with not knowing about what might happen in the future, or even not knowing what's going on at the moment, or what's happened in the past. Um, all these things we are uncertain about. Some of them, you know, usually things do have an upside and a downside, and so they could be considered risks depending on how they occur. But I think, you know, what I've learned is that uh people, you know, have to live with uncertainty. When you ask people, they say, "Oh, I don't like uncertainty." And then when you ask, but some, but some people like it, right? Well, some people like it. Some people are a bit more bold than others. But the point is that when you actually then go a bit further and say, well, do you want to know what you're going to get for Christmas? Do you want to know how a match is going to end? You know, if you've recorded it or something, do you want to know, you know, do you jump, just jump to the end of your uh, you know, series on TV to see the last episode and see what happens? Or also the thing I ask is, do you want to know when you're going to die, if I could tell you? And no, they all say no. Or some would qu, some would like to know when they're going to die. Some are so in a way uncertainty averse uh that they really like to have everything planned. But most people realize that you got to live with uncertainty and think of a life without uncertainty, without some risk. Think how awful that would be.
What are the people who hate risks the most?
Oh, there are some, there are some people I think who are very cautious, who uh would like to have everything planned out, uh, who want to feel that they can control all contingencies and have mapped out the possibilities. And of course, this is impossible. Um, and this is, you know, when I, when I talk to audiences, when I say, the first thing is not, not only that we have to face uncertainty, but we have to face actually deeper uncertainty that we can't even list the possibilities of what might happen to us in the future. We have to deal with that, you know, you know, cloud of unknowing. Um, and that's the part of human life. And uh, I'm interested in strategies that people use personally to to deal with that.
How is it linked with the Big Five? So, my impression is that introverts like risks less, uh, Americans like it more, uh, old people perhaps less than young people. Just how, how does that?
Yeah, I haven't actually looked at that. I mean, people will have looked at that because people have studied risk and proneness from risk aversion. What they found is that there's not a single characteristic. I think this is why it's not part of the Big Five, a single risk characteristic in people's personalities. Some people, I, I've known people who are incredibly sort of, what I would think, consider reckless physically with what they did with their bodies. They were extremely cautious with money. Uh, other people, uh, may be, you know, uh, very bold, um, socially, and take all sorts of uh, risks, uh, in in changing jobs, in changing friends, going into new environments, um, but again, may be very cautious about their physical health. And so, I, I, I, there is not a single risk scale where you can put everybody on. It's much more multi-dimensional than that.
Why do people fear the unknown so much?
Oh, I mean, I suppose if I was an evolutionary biologist, I might say, I mean, the, we, we fear the unknown risks more than, you know, more, more than the known, right? Which is why COVID was so scary because we hadn't seen it before. Exactly. Exactly. And that's been known, you know, for ages that unknown, unknown risk. Ellburg paradoxes that people invest in, you know, looking in the 1950s that if you couldn't actually say how big the risk was, people were much more averse to being exposed to it. And that's been known for ages. A well-known that idea of risk aversion to a well-known, um, to uncertainty about the risk has has been, you know, 70 years ago, I think Daniel Ellburg did his original, um, you know, study of that. And, um, and I think quite reasonably because, um, if you don't know what the possibilities are and roughly how likely they are, then all sorts of other, perhaps rather deeper attitudes to caution and precaution come in, and people might start being really hedging themselves against, and quite reasonably, against major losses. And we can see that going on, you know, in all, all parts of our life.
Um, what kind of habits can help us interpret risk in a more rational way?
Well, first of all, I don't like the word rational. So, I, I, in my book, I hardly use the word rational at all because I think this, um, claim that's been, oh yeah, we can look at risks rationally and deconstruct them thing. I actually, in real life, I think that's pretty nonsensical because, and I've taught this stuff for decades, and I've taught decision theory. I know how you should be doing it, and I know you can't do it in practice because it, for the, for the theory to work, you have to be able to list all the possibilities. You have to list all the options. You have to look at the probabilities of the of the possible outcomes and their, and their value to you, and then according to, you know, economic principles, you should maximize your expected return and things like that. Well, it just doesn't work. It, it fails at the first step that you can't even list everything that's going, going to happen, let alone, except in really simple sit, circumstances, um, put, um, numbers on everything. So, I, I, I, I think r, nobody can be rational. It, it just doesn't exist. I think. So, I really don't like thinking in those.
And this is some objective?
No, I think one can try to be in a sense reasonable in a much broader sense of, uh, you know, I suppose I, how I would say is, think is, first of all, you have to use as much imagination as you can. The things might happen that you didn't think of, but really, it, it, you're really, um, opening yourself up to problems if you haven't at least made a big effort to envisage the possible futures, to consider quite extreme scenarios. And that requires diversity of inputs. I'm hopeless at this. I have no imagination at all. Absolutely disastrous. So, but I know that if I were, if I were, you know, God forbid, in an important, you know, making important decisions for society, I'd want advisors with a, with a real range of different inputs. The, um, the example I like is is Barack Obama when he was faced with the decision about whether to send in the seals when it was suspected that Osama bin Laden was in a compound in Ababad. And he had a, a very diverse team of advisors who didn't speak to each other. And some were, you know, they may have been set up as kind of red teams, they, they were really pessimistic, 30 to 40% chance he was going to be there. Others were real gung-ho, 80 to 90% chance he's there. And uh, he had to put all that information together. But I think that's what uh, someone who actually has to take the rap, the real decision maker should be open to a diversity of opinion when there is no clear correct answer. So, I think the first thing is we have to acknowledge there's no correct way of doing this. We have to have a diversity of opinion and we have to work in in a a combination of an analytic approach, which I love. I'm, you know, based in maths, I, I've done maths and stats. I love de, trying to deconstruct uncertainty, looking at the sources, analyzing data, building statistical models, and it's great, but it's never enough. It never tells you what to do. You also have to have judgment.
A bit of an increasingly numbers are emotional. They're weaponized in debates and so on. Just how do you counteract that?
Oh, I, I think it's, you know, it's well known numbers, you know, it's a, it's a complete myth that they're cold, hard facts. Even before they're weaponized, we know that someone has made a decision to collect that particular data. There's lots of judgments that's gone into every analysis, every sort of measurement. They are, there's always judgment behind every statistical analysis, and it's good as that's made very explicit. Um, and so other judgments can be added to it. People may disagree about the fundamental tenants, the fundamental assumptions that are always underlying any statistical analysis. So, it's a mixture. It's not thinking fast and it's not thinking slow. It's has to be a combination of the two.
Now, you started out in medical statistics. Just how did you, how did that come about?
Oh. Oh, very naturally. It's just because there's a good job going and, um, and, and curiously, the job, my first, you know, a real job after, um, after university, I taught in America for a year and then came back, and it was, um, it was 1978, and it was working on artificial intelligence in medicine. 19 late 1970s. It was a booming idea. It was then largely called computer-aided diagnosis and computer-aided prognosis, and it was building statistical models to enable, um, you know, diagnosis, but it included computer interviewing of people with stomach complaints and things like that. It was really advanced, and the tech was terrible, but the ideas were absolutely modern, and, and, you know, the problems, the issues about how to integrate this in with medical practice were in there. So, I was working on uncertainty in AI for much of the 1980s, and we thought we'd solved it. Ah, how wrong could we be? But, um, because it's still a massive topic. Obviously, the machine learning techniques that people use now have developed, you know, beyond all imagination, and they are really are incredible. However, they are really struggling with uncertainty still.
What were, what was some of the strangest stats that you've seen in in the medical sector?
Oh, I don't know. Some of the mind-boggling, the mind-boggling ones. Yeah. I, I suppose it's the, the ones I've been involved in actually, four major public inquiries into health scandals in the UK. That's kind of where, um, I developed quite, you know, a slightly higher public profile, I think. So, ones where, you know, over 30 babies died with heart surgery at a center, more than you would expect to have died. And then, of course, the second one was Harold Shipman, the mass murderer, um, who murdered, you know, at least 250 and possibly 400 of his patients over a 20-year period. And, uh, we were brought in, he had been caught by then, but we were brought in to say, uh, could he have been detected earlier or not? And, uh, could he? Yes. Yeah. We can clear he could have been, he could have been, um, detected after a few years if somebody had been looking at the data, because he had so many excess deaths, but nobody was looking at the data. So nobody could be in.
Are people now looking at data properly across, across hospitals, across?
It's got, it's got better, but it's still, it's still slow. I'm in char, I'm on a, a group in the NHS that's only now building a, a really rigorous statistical monitoring system for adverse events in maternity units, and there's been endless maternity scandals in the UK. And, uh, finally, we've got a system based on, it's essentially a statistical process control system, but that we applied to Shipman. And then people took what we'd done, which was adapting industrial quality control to medical outcomes, and then applied it to, for example, intensive children's intensive care in the UK has got a monitoring system based, you know, almost precisely on the work we did for Shipman for early detection of problems.
Now, you, you are a leading kind of public communicator of, of, uh, statistics. Why is it important that people have a grasp of this field? Oh, what are the big wrong decisions people are making?
So, we only have to look at some of the, I, without mentioning any names, uh, well, do a few. Okay. We only have to look at what's happening in America at the moment to see what happens when high-level public discourse is not based on evidence. Uh, it's not based on numbers. It's, it's just based on on people saying what they feel like saying, regardless of of the evidence behind it.
Give some examples. What are, what are the most horrifying ones in your mind?
Oh, well, I, I, I think the, you know, what RFK is saying about vaccines at the moment, you know, because he's got a built-in bias. I mean, vaccines are not perfect. They're not perfectly safe, and they're not perfectly effective. So, I'd be the first one to say the term vaccines are safe and effective is is actually misleading. However, they are of enormous value. And, um, and he's got his particular, I think, biases there. And, uh, and of course, I'm not going to talk about Trump and his way he was setting charts and things like that. So, you know, it seemed to be, you know, how he originally did that seemed to be based on what a, you know, 20-year-old intern might do on a, you know, on a in a spreadsheet. And so, I, it, it just upsets me when I see, um, you know, the enormous. Oh, apart from, of course, sacking the head of the Bureau of Labor Statistics when he doesn't like the numbers. So, all these things deeply upsetting to a nerdy statistician who I don't, I, I don't want to tell anyone what to do. I don't want to tell anyone what the right policy is. All I want to do is say, please respect the evidence that we've got. Just respect it. And it, it doesn't tell you what to do or whatever, but just try to respect it. And, and it's not just, of course, politicians, it's social media, it's conspiracy theories everywhere. The, um, lack of concern or the, I think deliberate lack of understanding of what good evidence is and how a piece should be used is deeply upsetting to a to a nerd.
Why is it happening?
Oh, God, this is beyond my look. I'm a statistician. I'm not a great big sociologist. Okay.
But give us some more kind of examples of where you think the world is going totally bananas and moving away from from facts.
Yeah. Yeah. I, I, I, I think of obviously there's enormous blame on social media, on the algorithms, on the recommendation algorithms that that mean that something, oh wow, that's looks impressive, and it's almost certainly wrong. And so often they are based on numbers. You know, people love numbers, and they kind of think, as we said before, they kind of think they're cold, hard facts. But no, and often the numbers are actually not completely wrong. It's just that they're grossly misinterpreted and exaggerated and one-sided, cherry-picked. And, uh, you know, a, a wrong number that's that's blown up and people making some bold claim, uh, is around the world, you know, is just triggered by the algorithms everywhere. And it looks good. It looks impressive. And trying to backtrack on that is really, really, really is really difficult. And that's why, um, one of the things, you know, I'm on the board for the UK Statistics Authority, and one of the things I try to hammer all the time in the communication of official statistics, boring old official statistics, is, um, to try to preempt the misunderstandings that people will make. You know, because once everything's out there, it's really difficult to counter the misinformation, misclaims which might be made accidentally or deliberately. Um, you can't stop every people saying everything. You can't stop misclaims. Um, but you can, if you can preempt them, if you can understand by knowing your audiences, by listening to people's concerns, even the people you don't like, um, to, to know what might be said, you can get in there and actually say, this, this data means, you know, I think we can interpret it to mean at least this, but it does not mean this. And, uh, that, I think, is going to become, is becoming a more common trend in the communication of official statistics to say what things don't mean.
What was the most important thing we learned from COVID?
Oh, the importance of data. The importance of data in it.
Which data in particular?
Oh, everything. Good. We wrote a whole book, you know, with every chapter on a different data source. There was so much. Which I mean, I was working, you know, round the clock, really, analyzing data and communicating about it because I didn't have an official role, which was great, which meant I could get out there with the media and trying to explain things. And I, again, never said what should be done. It was only trying to explain the numbers. And you got had everything, you had the vaccines, you had the rollout, you had, um, the testing, you had, of course, the disease, the infections. But what a, what a fantastic time for a statistician. I mean, it must be, must be paradise for you.
It was, well, I don't, paradise isn't quite the right word, but it was, it was very exciting, very challenging, and incredibly rewarding. Um, and important. I mean, statisticians have hardly been that important before. Exactly. And I've had, I keep on, I now even now get people coming up and say, "Oh, thank you so much for the work you're doing during COVID." And the media had to learn. I, you know, that it wasn't just me or other statisticians who were out there talking about the numbers. And yet, we always at the beginning, at least, asked, well, you know, who's to blame and what's going to happen or what should be done? And we'd have to say, I'm not going to say no, no, that's not my job. You'll have to ask somebody else. All we're doing is explaining the numbers. And after a while, the media learned that their, this is what the audience actually wanted, an unbiased, ungendered discussion of the numbers, and they loved it. And so, you know, as I, as I was really popular, to be honest.
Well, you, you in your book talk about the five FCON rules. You know, you tell people what you know, you tell people what you don't know. Tell us about these five things.
Oh, yeah. Well, this is from, this is for communicating evidence in a crisis. This is all derived from John Krebs when he was head of the Food Standards Agency in the UK when he faced crisis after crisis. He had foot and mouth, he had mad cow disease, he had everything one after the other, total disasters. And he developed sort of playbook that he then afterwards he wrote and he said, this is what I did when I was talking five points. And I, you got to have everyone should have these written tattooed on them, I think. First, what you know. So, you know, be really clear about what we know. And then you say what you don't know. You say where the areas of uncertainty are. You admit them straight away. Second, not first, but second. And then you say, um, you say what we're doing about it. You know, we are learning more, we're doing experiments, we're finding out, we're collecting data, we are learning, we are learning. Then you tell people what they can do in the meantime. Time that you may want to be cautious. You may not want to eat beef. You may not want to do everything except you. So, you give people advice, self-efficacy in the meantime. But the final one is the most important. You say, "We will come back to you, and our advice will change as we learn more." So, you emphasize the provisionality of what you're saying. Now, this is, this is both deeply trustworthy because it's true. Um, it's also, as far as I can see, absolutely impossible for politicians to do. They just, it's just not in their vocabulary at all. This idea of provisionality.
Why is it so difficult?
Oh, well, they think they have to be absolutely confident about it. They say, "Oh, if we're not certain about everything, nobody will believe us. They'll just listen to somebody else. We have to be absolutely certain about everything." And I think they believe it. And our research with psychologists has, and, and not just our work, but other research we've done, randomized trials for different ways of messaging, strongly suggests this is a complete myth. That if you actually do, you're in a position of authority, you do actually admit some uncertainty, that there are pros and cons, etcetera, etcetera, um, that you are trusted more. And what's more important, you're trusted more by the people who were initially skeptical. Absolutely. By the very people you're trying to reach, trust you more because you're finally listening to their concerns. You're finally acknowledging that, um, there are issues out there that perhaps vaccines aren't completely safe and effective. So, what that means is the common political way of communicating, which is I think put into practice by communication departments in in government, which hammer through the message, bam, bam, bam, are actively decreasing trust in the group they're trying to reach. Those who are most skeptical, people who believe them already, there's no point. So, they're making it worse by their attitude. And I, I, since we did these trials and actually saw data on thousands of people showing that trust was improved in the most skeptical way if you gave a balanced, trustworthy message including uncertainty, I, it totally changed my mind. It absolutely convinced me about this. I also believe it's correct on an ethical point of view because it is correct, but it's also purely from a practical point of view, it should be more effective.
No, totally agree with you. We, we, uh, I've done a podcast with Rachel Botsman, who is a specialist on trust. And indeed, this is very, very important. But it's a bit, um, in the public sector generally, they never apologize either, right? It's kind of tied into the same thing, I think.
But well, what did we not learn from COVID?
Oh, a lot of it was, uh, not being flexible enough. Um, getting tied into, you know, we were told to, you know, wash our hands and wipe surfaces and things like that. And within about a month, we knew this was pretty useless scientifically. And yet nobody ever said, "Oh, what you can, this is actually, this is not the point. The point is fresh air that's more important than ventilation." Nobody said that for a year. And, and so I think what we didn't learn was that you need to be agile and flexible and take people with you by acknowledging the uncertainties and that you change course. So, the, the guy we worked with who was, uh, you know, most impressive, I thought, um, was Jonathan Van Tam, the Deputy Chief Medical Officer. And we worked together when the UK, you know, did admit that the AstraZeneca vaccine was causing these very nasty, um, well, which was actually, I think, pretty well first detected in Norway, causing these nasty blood clots, particularly in young people. And, uh, we worked on the communication of that. And where we showed that the, the benefits of the vaccine went down massive when you got younger, but the risks went up. And so there comes a point, you just shouldn't give the vaccine stratified by age. And he then said to the public, he explained all this, used our graphics, went through the numbers, treating the audience with respect, admitting the evidence had changed, and then said, "We're changing policy." And everyone said, "Well, fine." And, and he said, "Oh, we're adjusting our course. It's not a U-turn. It's adjusting our course." And, um, and there was no pushback from the media. There's no accusations. People really accepted it because he, he actually showed the evidence to the public as he was explaining it to them, and he could understand it. A politician would have been hopeless at it because he wouldn't have understood what was going on. He wouldn't be able to explain it. And, uh, so that, to me, showed that, um, you know, if you can get good scientists actually doing the communication, and they're good, and they're reliable and trustworthy, this can have an enormous impact on the public and on, well, public trust in authority, I think.
Uh, the world is totally overflowing with, um, with data. How do you, um, distinguish kind of the signal from the noise, so to say?
Well, sorry. That's my entire career. You're asking me to explain what being a statistician means. Um, that's a statistician's job, you know, trying to split the signal from the noise. And of course, you can't ever do it. And, you know, what's a signal, what's noise is never absolutely a black and white thing at all. But, um, it's by trying to understand, and this is a standard statistical, you know, thing that would have been said for the last century, the sources of variation. Just like in, you know, pre-war, you know, the 19, all the statistics developed in, um, in Rothamsted breeding stations for, for plants. And, uh, it was understanding the sources of variation, what led to the variation between the crop yields, what factors led to it. And of course, there is, in a way, unavoidable variation which tends to be called noise or random error. And then there's unpredictable variation which is due to factors that you might be able to control. And that's what statistics has worked on for about the last hundred years. And, um, and it's, it's not been bad. It's got some pretty good techniques for, you know, largely regression methods and so on. Um, when, and so it's actually done quite well. But you notice that that is different, really, from a, a strict machine learning black box approach, which just throws the data in and tries to extract a prediction, you know, a classification, possibly with some uncertainty, or a prediction of what, what's going to happen, where, um, there's no, you know, real understanding there of where it came from. People, again, you can't, people will have real difficulty for explaining why a piece of AI came up with its conclusion. Again, people are really working on that now, just like they are with uncertainty. But, um, you know, if you have rather slightly more basic statistical methods, uh, that you do, I, which I, that's why I support them, uh, you can, uh, then generate a much clearer explanation of why you came to that conclusion and what are the important factors.
And in which area are you the most impressed by the improved predictions?
Oh, well, I, I, I mean, AI has, is in terms of what you might call rather tightly controlled, um, areas, um, has done brilliantly. I mean, the people at Google DeepMind who started, you know, started on games like chess and Go and things like that, and then moved into, what obviously protein folding, um, and then, and also sort of medical diagnostic from images, in terms of breast cancer and eye problems. They've worked with the Moorfields Hospital. It is tremendously impressive in that way, in that they can take a really quite a big area, but these are all tightly constrained problems. Yeah. You know, there's a block of data, you've got an image, uh, you've got, you know, a set of data, and then you produce an outcome, and it's just brilliant at that. Where I'm much more skeptical is about, I think, you know, very, uh, unfounded claims that, oh, well, we can just put your medical record into AI and it'll tell you X, Y, and Z. And so, I'm much, people make have made a jump from these kind of quite tightly constrained problems, which are, which are just brilliant, into much more general problems. And it's hardly surprising they do that when we look at large language models, how effective they are at coming back with a, what is quite often a reasonable response to very generic issues, because they've been able to mine, you know, vast amounts of stuff on the web. Now, how good that would be about, and, of course, it can come up with medical lists of medical diagnoses and things like that. Yeah, it's just, it's fine. It can make suggestions, and it's can be extremely effective on that. But another, if you take another, I mean, you've been around for a while, I mean, just, uh, during, during that period, the accuracy of weather forecasts, for instance, is just mind-bogglingly different, right?
Oh, weather forecasts are fascinating. What's going on? Because there's a real compet, there's a real competition going on, I think, fairly friendly competition about the totally different philosophies for weather forecasting. Because traditionally, they, it's been kind of applied mathematicians and physicists, you know, building huge model, weather models based on Navier-Stokes equations and third-order differential equations, and they build massive models of the, of the atmosphere and then make a prediction, you know, six, 10, a week, 10 days ahead. But they don't use any data apart from the initial conditions, really. And then they vary the initial conditions, and that produces in an ensemble model, and they produce a probability of what's going to happen. The, the alternative approach, which people like DeepMind have taken, is to throw out all our knowledge of physics, the atmosphere, everything that's been learned in the last 300 years, and just get, and just throw it all out and take a, the massive amount of data and to do a pattern recognition, essentially, and make a prediction, which has no ability to explain why it's come up with a conclusion, particularly a pure black box. And they're doing really well.
Which one is going to win eventually?
Well, I, I, I'm, I'm really pleased that the UK Met Office has got both teams working and collaborating, because it, it, I think it's going to be complimentary in the end. But I'm kind of, because my background in statistics rather than applied maths, I'm kind of secretly on the side of the black box machine learning people, because I just love the thought of just putting the data in and out it comes with a, with a prediction, uh, without any ability to say why. It's just saying, well, in the past, when this pattern was there, this is what happened. Well, actually, maybe that's as good as you can do. So, I, I think, um, it's, it's a fascinating competition going on at the moment, which I'm watching with glee.
Would you have liked to be a weather forecaster?
Oh, well, I kind of, I kind of think if I had to study something, meteorology would have been, I mean, one. But what I'm interested in is the, I don't care about the meteorology. That's why I quite like the data-based approach. What I'm interested in is the first main Nature paper from DeepMind on this didn't have uncertainties in it. And I think it, it's really important to have uncertainties. When I, um, you know, I, all the time I look at probabilities of rain and things like that. I, you, I use those uncertainties. If I don't have them, I, I really, um, would feel a bit lost. And so, um, what I'm.
So, you, so when you look at the forecast, you don't look at, is it going to rain or be sunny? You, you look at, what's the probability of rain?
Okay. And what probability you need to bring your umbrella?
Well, exactly. I don't know. It depends how that's person very personal about how it depends what I want to do. If I want to have a picnic or not. So, uh, I know I need the probabilities. And so, um, this is, you know, this goes back to the 1950s with Glenn Brier developing a scoring rule for probabilistic precipitation forecasts. And it's, it's terribly exciting, I think, and the, the skill of these probabilistic forecasts is growing all the time, and it will continue to grow.
Do you buy, do you buy lottery tickets?
Uh, I bought one a few years ago when the, the expected return was higher than the ticket price. There had been so many rollovers. It's still almost un, completely impossible to win. But I thought, I've got to have a go at a gamble where the expected return is higher than the stake.
Where, in which cases do you look at probabilities where other people would look at yes or no?
Oh, um, I got prostate cancer. And so, uh, I, I really interested in in forecasting effect of people with cancer. And we've been involved in algorithms for, um, for breast cancer and prostate cancer, building software to demonstrate those to people. And they're all in terms of probabilities, you know, you know, roughish, but not bad in terms of 10-year survival, uh, and how those will change depending on the different treatments you've got. And I, I think it's absolutely essential when somebody says, "Oh, my doctor told me I had six months to live." I think, what? First, I never believe they said that anymore. Maybe there's some doctors who'd be so stupid as to say that, but I can't believe that you have to show, we don't know how long anyone's going to live. We can put some broad bracket on it because we, it's a survival curve. First of all, I want to say I'm really sorry to hear that you've got cancer. But given that you are a mathematician, what are your stats?
Oh, mine. Oh, yeah. Um, is that's the problem. I've got locally advanced prostate cancer. So, I, I've got some minor metastases, um, oligometastatic, it's called. I got a few of them. And so, I, I, and I, but the drugs now, the new drugs, the new hormone drugs, I'm on abiraterone. And, um, they just are so effective. Uh, now it's, I've got my PSA is essentially non-measurable. But the problem is, we try to get the use the data from survival in clinical trials. I can't. It's really depressing when I go back on the trials for abiraterone. They're terrible. You get sort of immediate survival of 18 months or something. I think, what? No, I'm going to live longer than that. I'm sure because, you know, in the trials, they were trying this on on very sick people. Um, and things have improved so much. And the, actually, the, um, I'm, I'm a, the pe, I'm at an earlier stage than the people who got the chance. So, there's no good data really on what are my survival prospects. It's very difficult. I, I, it's a shame, and I wish there were, um, you know, better, you know, databases. I really wish I could just tap in and find out, well, out of a hundred people who are most similar to me, what happened to them. But for a start, we don't know. We don't know. It's only been given to people like me for a few years. So, we got no long-term follow-up. We don't know. I mean, I, my, you know, some people got it earlier. My oncologist said, "Oh, I've had someone on this for 16 years." And said, "So, that's the problem with something that's fairly rapidly changing. When you're asking for a long-term prediction, you can't, you can't say." So, uh, well, all I can say, all I can say is, uh, fingers crossed. And, uh, yeah. Yeah. I mean, it's, it does seem a bit, you know, a bit pathetic for a statistician just to say, well, I, you know, hope I'm lucky. But I hope I'm lucky.
Why are we so bad at predicting elections?
Oh, elections. Oh, well, for a start, because when you go out and ask somebody, the, all the election things are just ask people, "What would you vote if you had to vote now?" I mean, that's the question. You're not even asking them what they're going to vote in the election. You're asking them what, what if you had to vote today, what would you vote? And, um, and that's a very biased measure of what you're trying to estimate of, of, you know, what that person will vote in three weeks' time, two weeks' time, one week, or even, you know, three months' time. People change, people are not necessarily honest. Um, and, uh, so people may vote, they may not vote at all. So, I, I think that the basic data source is always going to be biased. Um, and so it's not like predicting weather. You know, weather changes, but in a sense, it's not changing because of what people feel. Um, it's not like changing. So, I think they're never going to be great because they're trying to predict, estimate something that you cannot predict. You can't observe. No.
You've been bringing, um, statistics into some new areas. So, for instance, um, anti-doping, you know, you mean the World Anti-Doping Agency. What kind of things were you doing there?
Oh, then I, I'm not sure what's happened then. They had the idea of an athlete's passport. Um, because they wanted to actually a passport which kept a complete record of their drug testing history. Um, and it was to try to allow to a certain extent for the fact that, um, you know, there is individual variation between how people do respond, um, to drugs. And so, when you take a measurement from somebody, um, uh, they were particularly interested in, um, people getting blood transfusions, you know, just before an, um, athletic event. And so, you would be wanting to look at someone's red cell level or something like that. Was it remarkably high? Well, you only people vary anyway. So, to know whether it'd been pumped up high, you have to know something about their past history. So, essentially, you have to have a model for, um, you know, for someone's variability and their, their natural baseline level for their hemoglobin level. And so, uh, it was quite complex. No, it was fascinating statistically, and, um, and I, I haven't, I actually, I would be interested to know whether it's, what, what the current situation is about that. But it was, in a sense, trying to make the drug testing regime a bit fairer in order so it could adapt to the individual biology of the athletes.
From, uh, drug testing to the financial industry. Is there anything in the financial industry that puzzles you?
No, I've kept well away from that. I mean, I, for a long time.
Why, why are you keeping well away from?
I'm not interested. For a long time, of course, because I was, um, I was funded, uh, from the charitable arm of Winton Capital Management hedge fund. David Harding supported, and he was great. He gave us money and let us get on with what we wanted to. And he was, and I think, I hope he felt it was a good investment. But I, I, I, I had nothing to do with the hedge fund, um, business, and I've, I've just never had any interest in money. I hope it's just, it just puts me off. It's, I just, why didn't, why didn't, uh, Winton, uh, you know, who's established a very, uh, successful firm, why didn't he ask me, hey, David, come and check out all things and make sure? No, that was not part of the deal at all. No, he had plenty of really good people. He's a mathematician, right? Yeah. Yeah. Yeah. So, he, he had the whole place stuffed full of, of, of mathematicians, really competent people, much more competent mathematicians than me, for a start. And so, um, he, for a start, I couldn't have contributed anyway, and it would have, it would have not, it would have really inappropriate, I think, for for that to happen, and, and I just wasn't interested. I couldn't care less.
Do you think people are worried enough about the really, really bad outcomes? So, for instance, um, bioterrorism, uh, you know, the comeback of smallpox, that kind of thing, nuclear war, you know, climate change, you say.
Um, well, nuclear war is relatively, uh, less dangerous than bioterrorism. Yeah. Well, it could be on a larger scale, but, yeah, it depends on the scale, but it's, um, nuclear war would not be great. And so, um, I, I'm that's tricky, I think, because in the end, we're all going to die, you know, and, you know, within a very finite, finite period. I, from a personal point of view, I would understand because I think I do that that people do not want to spend their time obsessing against about all the terrible things that could happen in the world. It seems to me, um, that this is not a, um, not beneficial to your mental health, shall we say, to be really obsessed. I, I kind of hope there are people studying it, you know, more professionally and who are trying to counter it with appropriate regulation, appropriate policing, and so on. Um, but I personally do not want to spend my time waking up in the morning worried about smallpox and bioterrorism. Um, if people are so, in other words, I, I think I can understand because I don't do it why this is not full of top of the agenda in people's concerns.
Catastrophic existential risks.
Um, I, for a start, I, I, but I'm hopelessly optimistic. I think actually they tend to be overrated. Um, and so, but that's maybe because I, my particular personality is far too optimistic.
Was Norway lucky to find the oil?
Oh, that's, I don't know. I think Norway, I, I don't know about how finding it, but it was extremely sensible in how it dealt with it once it had found it, compared with the UK, um, who, so, and, you know, which is why you, why you've got your job at the moment, to some extent. And so, um, I think that Norway dealt with this in an absolutely brilliant way of seeing it as a, as a national resource to be, in a sense of, you know, um, you know, to be nurtured for the whole entire community, rather than just in the UK to, to sell off the rights, uh, in order to raise, raise some money at the in short-term view.
How do you look at climate risk?
It's difficult because I, so, so you have people talking about it like, uh, some places in America, you sit there in the middle of ash, rain, um, forest fires, and you're kind of in the middle of it. You're sitting in the middle of it, and you're saying, "Climate," you know, "there's no problem with the climate." I know. I know. No, it's, you know, it's just happening, and that's it. Um, you could just tell by the events which are just going to carry on. There'll be, you know, ups and downs, and but there are going to be more extreme, extreme events. I'm interested in, uh, in, um, attribution. What, what it's brought to the fore is attribution studies, which is not so much about climate risk, but about looking backwards and say, to what extent was this caused by man-made climate change? Which is a really big growing area in research. And it's going to be a big growing area, uh, financially, when people start suing, uh, fossil fuel companies for, um, for events that happen. And so, the, uh, you know, and, but it all requires models. You have to have a model of, uh, what we would, you know, how the climate has developed, um, with man-made, you know, man-made interventions, and how we think it would have developed had we not been throwing all this muck into the air, um, since the 1700s. And, uh, and you see, see how likely these events were under these two different scenarios. And the relative risk can, is, our meteorological office do have got an attribution center, and they just, they will give you a relative risk. Technically, that can be converted into a probability of causation to say, the probability that this hot weather event, that this tornado was caused by man-made climate change was X percent. And, uh, officially, once that gets above 50% by the balance of probabilities on a civil court case, um, you could say that man-made, man-made climate change was responsible. Now, trying to then attribute it to particular companies, I think, is rather more difficult. But it, it's brought from a technical point of view, it's brought this fascinating idea of attribution. And of course, for that, you have to have climate models, and the climate models are used to make projections as to what's going to happen. A lot of uncertainty, and they, the models take a lot of time in of dealing with that. They also have what I think is really good, independent teams coming up with different climate models, which then they pull, and then they make it even more uncertain. They broaden the things out. And, um, you know, we're never going to know what's going to happen. And, uh, one, one should not state too confidently about what's going to happen. But we know bad.
Things are going to happen. Um, and what's to do about it? Again, not my job. Not my job, I'm afraid.
Talking about big things, uh, in, uh, 2021, you said that, uh, AI poses an extreme risk, but that it perhaps is overrated. What do you think now?
Oh, I think it poses an extreme risk and I think it's probably overrated. [Laughter]
What uh, could you explain? Yeah, I mean it's amazing how how strong it is and it turns what you mean about extreme risk. It's certainly going to take some jobs and certainly we're all going to have to adapt our jobs, our work to it. It's also incredibly useful and valuable. I use of course use it every day and um that I but in terms of sort of risks well people when they talk about this they're talking about sort of existential risks about you know um the the say self-aware AI that's going to start uh having essentially a will of its own and deciding that the these people get right get in the way of what it wants to optimize and um I I think of course that is a possible scenario. And and I think it's quite reasonable then that people respond by wanting additional overview of and guard rails on AI.
So it's like a lot of these things people say it's a bit like COVID before um, you know, COVID. The modelers in the UK said, "Oh, there could be half a million deaths," and that still gets quoted. Oh, they said there were going to be half a million deaths. No, they said there would be half a million deaths if nobody did anything about it. If we all just sat there and let it wash over us, there would be half a million deaths. And there would have been. But they're now being accused of saying, "Oh, they said there would be half a million deaths," and there weren't. So the point is that one could talk about there being risk. But I think it's limited the value of that if we're all going to assume we're just going to sit here and allow ourselves to be taken over by killer robots. So I think that what it does do is of course call for far greater scrutiny of um what's being done um in the o in openness about the guardrails put in um and so on.
I mean, but are you seeing, but are you seeing the necessary guardrails and so on being put in place? I I don't know enough about it. Again, it's not really my area.
But when you look at AI, what are the type of things you look at in order to to gauge the risk? Oh, again, I don't know enough about. I don't know enough about these sort of existential risks and how those could occur, the super intelligence and things like that. I really don't know enough about it and I wouldn't want to claim to.
And how do you use? You said you use it all the time. What kind of What kind of things do you use it for? I use it in writing my book. I use it for researching and I use it for coding and I use it for personal things like trying to work out where I'm going to go on holiday. So, so I'll I'll use it all for all sorts of stuff.
Where does it where does it tell you to go on holiday then? Oh, well. I I I I don't just say, "Where should I go on holiday?" It's not quite that broad, but I no, I use it all the time and it's and of course we we're almost forced to use it now because it comes up number one thing when we do a Google search. So um I I I I think it's incredibly valuable. It's unbelievable what it can do. But I the guardrails of course are already in there in many ways in terms of, you know, violent speech and racism and all sorts of stuff that it can't do that are built in. Um, and I want those. The crucial thing I feel is that these should of course should be open and public and uh and they should be there should be a regulator to make sure they're being adhered to.
So, but lots of people are saying this. I mean, there's a massive AI safety is such a massive because all the tech people are going on about it as well. So, so I I'll leave them to it and hope that there's some some decent people involved in it. But it is a crucial area, not my job.
Uh now if you do look at your job and and kind of your legacy, what what is the one kind of idea or principle that you hope you will be remembered for? Oh, I don't know. I've moved around rather a lot. Um, yeah. Oh, one thing. Oh, I think I I think it's the stuff I've been doing later. I think it's on trustworthy communication of evidence. Um, in a way, it's not my. I've done quite a lot of technical stuff. That's where I get all my citations from, etc. and I get loads of those which is lovely. But in the end, uh what I'm just obsessed by is the need for trustworthy communication about evidence. Uh, as we said, not that it tells you what to do, but unless I mean, we're s sort of doomed if we're if we don't use evidence in in an appropriate way and don't communicate it properly, we are just left up to thinking fast. In Danny Kahneman's term, we're just left up to gut reactions and emotional feelings in order to for everything. And I think, wow, that is just disastrous.
And where is the next step of this trustworthy communication? Where is it? Where is that kind of part of the science going? Oh, uh, yeah. Well, people are concerned about it. They're obviously concerned about, you know, the quality of what's published in the scientific literature, which is a massive problem because there's so much junk out there, um, un from paper mills and so on. Um I think that's well well I can talk about in the UK with actually this is now you've got a very high level. There's a new there's a code of practice of statistics in the UK which is a pretty dull document, but it's incredibly important. There's a new version coming out which is really putting down, you know, the way in which all official statistics and even non-official statistics should be communicated to the public and based on these ideas of preempting misunderstandings, of not being misleading, of being open about limitations and so on. And that's all down there and people have to adhere. Government departments have to adhere to it. They are actually bound by it. And so it's I think quite setting quite a good because in the UK we've got an office for statistics regulation, which I think is might be fairly unique around the world. An actual body that is there as a as an inspector as a regulator for statistics. And uh I I'm a, you know, huge believer in that, obviously. And I would love to see that model developed elsewhere that you have got a body who can really tell people off when statistics are being misused.
My god, they'd have their work cut out in the US at the moment, wouldn't they? Oh, that's for sure.
What's the big what's the big unanswered question that you still want to tackle? Oh, well, I'd like quite like to know understand consciousness and whether there is such a thing as free will, but that's again going somewhat outside my um in my my professional expertise, shall we say. Um, but it is in my book. I do discuss it because I think it comes quite important when you start talking about whether to what extent is the world genuinely stochastic and random or whether in to what extent is it actually deterministic but staggeringly complex in a way that renders it unpredictable. Um, I think is an interesting issue.
Um, oh, oh, interesting question. I suppose, you know, broadly it comes down a bit to what we're discussing before about whether COVID was actually quite encouraging because in a crisis situation, um, good communication did rise to the surface. There were obviously people on each side arguing. I was right in the middle getting attacked by everybody. So, I thought, yep, doing the right thing. And it wasn't just me, but there was an enhanced respect for the mainstream media and largely people um, you know, were interested and uh, you know, there was a a whole body everybody was discussing the data. It was a very active community and I think it was it was as we discussed a very exciting positive time. Now, once a crisis has gone, everyone goes back to their normal normal stuff and loses interest in all of this. And um, I suppose my qu my question I'm interested is whether that kind of interest and attention and wanting trustworthy information can be retained in societies that are becoming increasingly popular. You know, follow populist politicians who object to authority, who are distrustful of what they call elites and experts and so on. And as that happens, well, first of all, I suppose the big thing is, can that be countered by having trustworthy experts, you know, people who do know something out there?
Yeah, I think it's so interesting. According to Bill Gates, it would take roughly a billion dollars a year to make sure the world is really ready for the next pandemic. And we, the world is not spending that money. Yeah. And I, the next pandemic will be different. Um, I and and it's people certainly wouldn't respond. I don't think in the next pandemic people would accept lockdowns. I just don't think they'll be politically acceptable. So I think uh so we all so so we'll all be Swedes in the next pandemic. I think well, the as Swedes as they as they said, the guy, the minister said, well, we practice social distancing anyway in Sweden. So I thought that's great. So um, yeah, I think we would be, I think we would be more Swedes in the future. Um, and so but the crucial thing about that is that what it says is that, you know, you said you can spend a billion dollars, but you can't change how you do. What are you going to do about people? I mean, how people react is absolutely crucial and you don't necessarily change the way they react by just spending money on on on things. So I I think that the, particularly in pandemics, the the role of human reaction to the situation is which is the most important and the least predictable aspect um and uh and actually very quite difficult to research, very difficult I think to learn from the past pandemic about exactly what worked and what didn't because things were so different across um every every bit of society with within and between societies. So, um, I'm not quite convinced that just throwing money at something is going to necessarily uh, you know, protect us against it.
What do you read outside statistics? Oh, I read um some crime novels, but I'm interested in. I really like history and biography um especially military history. I'm obsessed with the Second World War. So, I I spend my time visiting um, you know, war sites around Europe and and further and and in India. So that's actually what I'm interested in.
Why are you why are you so interested in that? I've been asking myself that, why am I so interested? I think it's growing up in this generation and, you know, born in 1953 in the UK. The war was around us all the time in the films and the culture and the experience of of the adults around us and things like that. We grew up with it as small kids, absolutely obsessed with it. And um, and I never really have quite lost my my interest. I I think partly because every time I read anything about it, I just sort of thank uh heavens for my, well, what's known as constitutional luck. The fact that I was born when I was born into a society I was born into, which was staggering constitutive luck.
Now, if you were to apply your if you were to put on your professor in statistics hat and give advice to young people using some numerical math or whatever, what what is your advice to young people? How should they think about their life in statistical terms? Yeah. Well, I do I mean, some of them even read my books, which is I don't I had a 17-year-old come. I nearly burst into tears because he came up and said, "Oh, I really like your book." And I said, "Well, it's not actually aimed at teaching you." I really I was so moved that he liked it because I actually think there is some stuff in there about um, you know, fa we have to face up to uncertainty. We we don't know what's going to happen. I didn't know when I was 18. So I had no idea what was going to happen in my life. I was, as I said, I had enormous constitutive luck where I was at that time. Lots of opportunities, you very secure situation, everything paid, healthcare, university, everything all paid for. Um, I was in a very really privileged situation. Um, and so but there's still of course massive unpredictability. But I didn't mind. I took I felt that I'd got a good upbringing which gave me resilience. So the crucial thing and of course this is as relevant to corporations as it is to human beings, I think, is resilience. It just it's the number one priority because that's how we deal with the deep uncertainty in all situations. The fact that we don't even know, we can't even list what might happen, particularly some way in the future. So all what we have to do is cultivate resilience, which is an ability to deal with, if not and learn from, benefit from anything that can happen, whether it's good or bad. And um, for some reason, I think I developed quite a lot of it. I'm not quite sure, maybe it's a nice secure um upbringing I had um, and the the good fortune I've had, but for some reason I think I got it. I don't know. And I think know again, I would say this for young people, that you have to go out and take risks. Don't be reckless. I say this to I do I I say I give talks in schools and I say, you got to take risks now. Don't be reckless. You know, look, cover yourself from the damn major downsides. You know, just be careful, but take risks. So go out, you know, camping on your own in the middle of a moor, but don't be stupid. Let people know where you're going and make sure you got the proper kit in Norway because everyone knows about that stuff, but not in the UK necessarily. So, go and take, go and have those adventures, but don't be stupid. And so, and it's through having, you know, those sorts of adventures and taking some risks and being in situations where you're not quite sure what's going on that you develop resilience. So, I just say to young people, go out there and have adventures, but don't be stupid.
David, I had a 90% probability on this uh podcast being very good, but it's even better than I expected. I just absolutely love talking to you. So, big thanks for everything you do to, you know, society and increasing the knowledge of stats and just for being such a wonderful communicator. Big thank you.
Well, thanks so much. This was somewhat outside my comfort zone, a lot of this, but so anyway, so I suppose I had to take the risks of of being there. The biggest risk I've done for quite a long time is being on this podcast, I tell you that.
Great. Big thanks. Thank you. Bye-bye.