Fraser, Nick and Peter wonder if they have had enough of experts.

For more information on Aleph Insights visit our website https://alephinsights.com or to get in touch about our podcast email [email protected]

Hello and welcome to the Cognitive Engineering Podcast produced by Tell Me Studios for Aleph Insights. In this series of podcasts we take a look at interesting topics and discuss what we think they tell us about analysis and decision making. I'm Fraser McGruer and I'm here with Nick Hare and Peter Coghill of Aleph Insights and this week we're discussing experts and whether we should trust them. So British politician and prominent Brexit campaigner Michael Gove recently said people in this country have had enough of experts and it's arguable that he was right since soon after he said that the UK voted to leave the EU with the majority of UK voters ignoring the expert advice to stay in the EU. So tell me Nick what's going on here and why in this case did these people ignore expert opinion which I think was mainly economists but I'm sure there are others as well. So why did these people ignore expert opinion and were they rational in doing so?

I suppose the question we ought to ask first is what is an expert and we get a lot of insight from a recent project called the Good Judgment Project which I've been involved in which was sponsored by the US government and run by Professor Philip Tetlock who has been doing research into expertise particularly about political and economic matters for really the last sort of 30 or so years and what that project set out to study was the extent to which people knowing about things, people being experts made them more likely to get their forecasts right and in this context when we mean a forecast we mean someone putting the right probability on a particular outcome. So this project essentially asked vast numbers of people, thousands of volunteers questions like will Bashar al-Assad still be in power in Syria in a year's time or you know will the Scots vote for independence and those sorts of things. So very measurable outcomes? Well-defined outcomes and then looked at the probabilities that people put on those outcomes and then sought really to find out what kinds of factors influence people doing well. So doing well in this context is you know if you're for every 100 things that you put an 80% probability on, 80 of those ought to happen. If it's 99 of those things or if it's you know only 10 of them you're getting something wrong, your probability is not well calibrated and what Tetlock found was a very robust finding which was that what you know, so what you know when you go into that project. So your level of expertise? If you like your level of depth of knowledge of something is not a good predictor of your ability to make good forecasts about that thing or indeed about anything else and that what determines good forecasting performance is actually the cognitive skills of the forecaster. So it's the methods and techniques that they use, the fact that they take account of base rates which are essentially what's the sort of long-run average frequency of things like this happening and their ability to take what's known as the outside view which is really to abandon the narrative-based approach which is sort of oh you know this guy wants this to happen and then he's going to do this and somebody else is going to do that

and take more of a statistical approach. So okay so for example one thing so let's say if to use the case that you talked about will Assad be in power in one year from now in two years from now etc. So you could go to a bunch of Middle East experts or Syrian experts and ask that question or you could go to someone who knows nothing about the Middle East, knows nothing about Middle East politics but has a you know who for whatever decisions they make let's say that they're for want of a better word they're an engineer for example and they're used to applying processes and methodological rigor to their decision-making process and if you present that person with certain facts or certain bits of information then this study showed that that second person is more likely of coming out with a better response. Is that what you're saying?

That's right so it's not that they're coming it's not a total ignorance it's not that they don't need facts right they need to be able to get hold of facts and that will point to something I think which we'll almost certainly discuss but about the types of expertise and demand for different types of expertise but the point is yes I mean that actually a Syria expert who didn't have good forecasting methods would do worse than someone who came at it totally cold but who used robust forecasting methods and in this case that would involve for example saying well what it you know what other situations have been like this in the past who have there been any other kind of embattled leaders in civil wars and what tends to happen to them and you know what reasons have we to believe

that this situation is different. Okay so that could reframe our definition of an expert because it could an expert could just be someone with excellent forecasting skills rather than knowledge

is that one of the things we're potentially saying Peter? Yeah I think so I mean I think just to summarize something that Nick suggested that in extremes there are two types of experts you can split the spectrum up so on one end you have one class of experts you have the methodologist the person who would probably say I don't know anything about anything but I know how to use the data I've got available to form an opinion about what might happen on the other end you have the the the and they're sometimes called foxes so they're referred to as foxes by Tetlock and and previous writers then on the other end you have the the the knowledge-driven people who may know lots and lots of specific things about a topic and they will rely on these knowledge and be able to spin together a logically consistent narrative about what might happen so that and they're sometimes called hedgehogs they're often said that they know one thing very well rather than and they will and they're called hedgehogs and they're called hedgehogs why do hedgehogs know how to do I think it's because I think it's in reference to the way the hedgehogs will sort of curl themselves up in a ball and be stubborn about something I think

that's what it I'd say yeah it's a reference to us to a story by or a kind of parable by Isaiah

Berlin but that's the idea sorry I interrupted your flow there sorry I got I got diverted by hedgehogs what the how we define experts at the moment I think is is is is a little dangerous I think the the the the people who are able to spin together a nice sounding logically consistent complex story are quite appealing I mean they know stories are appealing things they they they they appeal to people and they're often easy messages to pick up and and to understand and pass on whereas looking at lots of data and saying this is this is objectively the best thing to do within these within this within this out within this goal that's quite maybe a bit boring maybe a bit geeky and much more difficult message to explain especially if you're having to it's predicated on complicated statistical methods sorry I want to stop you there sorry

it sounds like all you're saying it's not all you're saying but between the two of you what you're saying is that this is a qualitative approach versus a quantitative approach and the quantitative approach is superior which is kind of what you always say with stuff and but I think what's interesting is is as much as you know one can keep saying that and the statistics back it up I think what's interesting is and although we I know we didn't start this this I mean this what we're talking about is whether we should trust experts or not so we've been talking about well what is an expert but what is interesting is that people don't you know whether it doesn't matter what kind of expert it is because it could in this case it could have been someone who knows a lot about the effects of an economist who knows a lot about the European and the potential effects or it could be a statistic statistic statistician who can both those different kinds of experts would have a similar opinion let's say and yet still people

don't listen to them so well I mean no let's qualify that right so people listen to their doctors people listen to their piano tuners you know that no people don't try and do that themselves people don't you know people listen to their builders and their painters and decorators people listen to their plumbers their boiler but people listen to their car mechanics you know no one I mean very few people try to fix their own cars people don't pay much attention to people to people who are supposedly experts about big complex systems like the economy and like the you know the political environment and I would say probably with with some justification because you know they don't actually have a very good track record there are there are reasons to distrust them not just you know by looking at Tetlock's work specifically by looking at that study but but also the fact that they don't generally people you know pundits on TV people who set themselves up as experts commentators do not you know routinely publish their track record now you know if someone comes to try and fix your boiler and they they accidentally blow your house up they're going to end up you know probably not getting very much work and you know in this day and age getting a bad review on you know on the internet so so it I think Gove is touching on something which is you know quite justifiable which is that experts about the big issues do not have a very good record actually let's come to that in a

second so Peter you wanted to say something there when we were talking about track records exactly

so I think a technocratic solution might be replicating some of the functionality that the Good Judgment Project used when they were running this large-scale experiment where they tracked and recorded people's performance in forecasting be quite easy to imagine a similar system used for experts of all kinds answering questions of all different types classified in different buckets and you could you could based on people's previous performance decide how much you were going to listen to them on a particular topic so Nick might be an expert in Syrian politics but he might be atrocious at Russian economy I just want to point out that I'm atrocious at both of those things yeah so hypothetically speaking but so I would so I could look at look up his sort of portfolio his CV of things that he's been right and wrong about and I would definitely go to him about Syria but not about Russia and it would be quite easy quite easy to run if it was run in some sort of market then it could be self-funding etc etc but I think what's interesting that Tetlock found was that he invited lots of these so-called pundit-like experts to be involved in this experiment and I think almost universally nobody wanted to be involved because they sort of recognize that the ground that they stand on is quite tenuous and that to be actually scored on how accurate they are would probably be detrimental to their career because they've made a living out of making statements making judgments making predictions but never actually being held accountable or score sort of retrospectively scored yeah I build a mini career on that for two

and a half years as a country risk analyst and thank God our you know no none of us were interested in having our results sort of checked because God knows what it might throw up I think well luckily probably no one actually read or acted on what you wrote thank you thank you it's well actually to be honest that was part of it was a disconnect maybe this is something we're talking about here there was very much when one of the reasons I left being an analyst and I know I've mentioned this to you before because I had no idea what was happening to my analysis you know were multi-million billion pound decisions being made on it or was no one reading it at all we had no idea

but that's that's an interesting thing and that might be a whole other topic for a whole other podcast but the idea of feedback so you you you'd be nice as an analyst to be incentivized by what feedback you by what what your analysis was used for would incentivize you to make better analysis and I think that's a common problem is the lack of feedback but I just want to something else I wanted to mention is that there's lots of work by people like Gary Klein and Daniel Kahneman looking at how decisions are made and what rationality means and how much you should rely on your gut versus how much you should not rely on your gut and ignore your gut and rely on the data available and I think it's been shown Gary Klein particularly is a big fan of the the sort of learnt and developed heuristics the kind of instinctive decision making that many people use such as firefighters as an example he he often cites where if you're in if you're very regularly in situations with lots and lots of things going on let's say you're a firefighter with very sort of visceral feedback about your decisions being right or wrong you can you can develop these these decision making heuristics these sort of shortcuts that means you don't have to use your rational brain you can your your other parts of your brain can tell you what to do without actually without you actually consciously knowing what to do so in situations where you get regularly you're making lots of decisions of a similar type and you get very sort of obvious feedback about whether it was right or wrong you can develop a sort of subconscious decision making mechanism that can be that in some cases could be relied on quite rationally and actually this is

an issue which has a parallel in the development of machine learning and the kinds of things that we know that machine learning at least at the current levels of technology technology are good at so the there was a a learning architecture developed I think funded by Google but called DeepQ which learned to play computer games and the computer games that it did well on were the ones where there was an immediate type of feedback so where you know you shot an alien and then got points for it and the one and the things it did badly on were the things where the the the distance between action and reward was was was distance so ones that it required advanced planning where you had to you know pick up a key go through a door you know climb up a ladder and then pick something up and that's when you got points you know so be that the proximity of your decision to the feedback you get for that decision is absolutely crucial in determining the extent to which someone is able to acquire expertise okay let's wrap up there

um so if we go back to our original question um we're discussing experts and whether we should trust them so what's our conclusion what's our wrap this up i i would say trust experts if they

have um got a demonstrable track record if there is no track record um no reason to trust them

and um you know we need no no no no no no what because i think this is this is what we talked i think this is one of the problems this is one of the issues is that it's all very well saying um you know find the track record one what about if there is no track record but actually probably more importantly most people don't care and people tend to be quite apathetic or lazy about um they just want to know what the right decision is or they'll go i don't think people go into it that that as much as you're saying whether i'm saying questions i'm saying what they ought to do i care

what they do do if people want to go around ignoring everything that's fine um you know they'll may end up making bad decisions but it but they can't but you know if somebody has got so that sorry as well as you know looking have they got a track record um the other key thing is look at their methods so you know are they are their methods the kinds of methods that we should expect to produce true beliefs so peter what would you suggest people should do in terms of

listening to experts or not well to mirror a point nick made um don't get swept up by a compelling story um find analysts who have considered multiple options but um define what your success criteria are so as close as possible so it's a measurable point um and track and maintain performance your own performance and performance of others and i mean it's

it's not easy it's not like there's an easy checklist of how you should trust someone um you have to put a bit of work in yourself okay we'll wrap it up there um it's not easy you have to do a bit of work yourself but i just one final thing yeah is the the skepticism that gove expressed yeah is justified but the response is not to throw out all kinds of expertise the response is to be more sophisticated in our approach to to evaluating it okay so to slightly

modify uh shakespeare don't kill all the economists right don't kill all the don't kill all the experts i don't even did shakespeare say something well he said kill all the lawyers oh i see yeah

i think he said that so uh i suspect one of the characters in one of his plays said that uh i

suspect that you have quoted you've cited that more accurately than i just did okay so uh that's um experts and should we should we trust them so um thank you very much um i'm fraser mcgrew i've been here with nick hare and peter coghill of aleph insights as part of our regular cognitive engineering podcast thank you for joining us and until next time bye