Parsing the Knowledge of London cabbies

[Music] Welcome to Science Sessions, the podcast of the proceedings of the National Academy of Sciences, where we connect you with Academy members, researchers, and policymakers. Join us as we explore the stories behind the science. I'm Paul Gabrelson. Who would you talk to if you wanted to learn how humans plan complex routes through, say, a large city? In a recent PNAS study, Hugo Spears of University College London, Pablo Fernandez Velasco of the University of York, Eva Maria Grease-Bauer of University College London and colleagues asked London taxi drivers who must pass a stringent test certifying their knowledge of the city to talk through how they'd plan a route from one point to another. Their responses revealed elements of how they thought through the task and what characteristics of the street network proved most challenging to navigate. Hugo, tell me about the background of the study. What did we already know about how humans plan routes? There'd been a range of studies done looking at planning and humans for how they might choose routes through space, but it was always a very small space. Spaces of people would learn in a short afternoon, particularly minutes, rather than an hour. But the real world is much more complex and so with this study we wanted to look at how do humans plan in a very large city, which is something humans do very well. Pablo, why did you choose to study London taxi drivers? There is a lot of previous studies of London taxi drivers because they mastered the knowledge of London. You have to pass these very, very difficult examines which you have to be able to find routes from any two points in a very big area. It's quite remarkable the exam they do. If they give a wrong turn, they're dismissed, and they typically fail this exam like 12 to 16 or 20 times before they're awarded the badge. They have to know about 26,000 street names to be able to do that exam properly. So London is a really interesting city to navigate because it's got this combination of some gritty bits to it, but a lot of it goes in twists and turns. It has structure to it, but it's got a variety of streets and there's got lots of iconic landmarks, but the most important thing about London goes by to 1843, or so when there was a great exhibition in Hyde Park, the world came to see it, but very few people got there directly. Taxi drivers were awful, and the government were embarrassed, and they instituted an exam from that point onwards. Without that exam we wouldn't have the research paper we have. Tell us about the experimental setup. What did you ask the taxi drivers to do? This is one of those experiments that takes a certain skill at talking to London taxi drivers. So that magic was undertaken by Eva Maria Gracebarra. So she was really good at getting uncomfortable, and she worked very hard at working with taxi drivers to come up. What would be a painful journey or what would be super easy? And so a lot of credit goes to her. We went to court the taxi drivers and they're having a lunch and we played an audio recording of some location in London that was their origin point, and then we said you've got to tell us all the streets between that point and another destination point. So the taxi drivers scratched their heads, you know, I'm an Arab, think about the route they would take. Sometimes they were quite quick, sometimes they were slow. Slow would be maybe 40 seconds or so. Quick might be 10 seconds. We were interested in how long that time was, we call that pre-planning. And then they had to tell us the names of all the streets that they would go through, and what direction they would proceed into them. So for example, it might be something like south to Peter Street, turning left onto Peter Street, and turning right onto Wardau Street, left onto Sharsbury Avenue, right onto Charn Cross Road, and so on. And here's what an actual London taxi driver detailing a route in central London sounds like. The driver, whose voice has been altered to protect privacy, is reciting a route between Burley Street near the Strand, and Great Marlboro Street in Soho. Right to Burley Street, left to Stokes Street, right southampton Street, left in Rietta Street, right into. I mean, the street goes right into Bedford Street, four Garick Street, four Crabble Street, right into Charn Cross Road, and Charn Cross Road, I would then go into. left into Sharsbury Avenue, right towards Wardau Street, left into Nome Street, four Dintelay Marlboro Street, set down on the right. Occasionally, they pause. It might be 20 seconds, in some cases, to really think what is the best next step I need to take, and they finally, say, set down on the left when they get there. Once you have the audio file, you look at the poses, right, how long they take in between the streets, also how long they take before they start rattling through all of the street names, all the pre-planning. And we take the length of the pose in between street names as a marker of cognitive load, and then we are able to simulate what is the structure that they're using for their planning task. Eva Maria, what was it like to talk to the taxi drivers in this study? I really enjoyed it. It seems like they are a very special group of people, very approachable, very friendly. They try to help as much as possible. The first step was the approach, getting a nice conversation going, finding a little bit of a bond, talking about the experience, and also how we want to do the study, what it will mean for them as taxi drivers, what they can find out about themselves, basically. What did you learn about how these drivers plan routes? You could see that maybe the taxi drivers go one by one from the beginning to end, and then they tell you, but what we found is that they actually do a different thing. They prioritize certain parts of the route. What they prioritize is the junctions that are high in complexity. That's really at the heart of what they're doing. Collaboration with Daniel with Nanny. The key addition was the idea that we could take specific metrics, mathematical metrics of the street structure, and see if those predict the patterns. There are cases where you just need to go straight, or it could be you have to pass through bits of London that have got lots of options, and they're connected to other options. That you can capture by metrics like how complex is the transition structure in the road network. You can also look at how the structure of the streets is organized with successive steps you can take to your destination. This study found evidence that exactly these measures and not other measures do describe how quickly those taxi drivers can plan their routes. But I think the key as well is it's the early planning. So you could just discard that bit with Armingen-Arre, but it did matter. It was important. The taxi drivers thinking longer, planning more. They were faster, more efficient for particular streets and junctions. There is actually a change depending on how long the streets are in London. So they take longer to think about longer streets. And this does seem to correspond to some of the descriptions they give about how to do their planning of routes, where they might look at it from an area and then a first-person perspective. A bit of a simulation. It was amazing to see how taxi driver actually plan, because whilst I was talking to them, I could see how in front of their eyes they seem to go through the routes. They were actually driving down the streets. They were actually doing the turns in their minds. Your brain would go through the same things if you are thinking in a familiar space. Taxi drivers would tell me that London for them is just like my living room to me. I know where to walk across my living room, how to get round the chairs. For them, it's thousands of streets just feel like they're living room. What our data is suggests is that it's the structure, the complexity of the route will cause more time delay to plan. We have evidence to suggest that your brain would likely do the same, would struggle with certain options more than others. What we're showing here is it's a particular type of complexity. It's not just generally like if there's loads of options, it's hard. Of course that's true. But London is a street network, it's just full of options, so it's about these metrics. How does this study add to what we know about route planning? It points to ways in which human route planning is efficient. It gives quite a lot of detail about how humans are able to do something that's quite exceptional. Planning routes in incredibly complex spaces like London. Here we saw one way in which this planning can be a structure. So as to tackle these very, very challenging problems with limited cognitive resources. What's nice for me about this study is that it's brought together a neuroscientist, which is me, computational modeling expert, Daniel McNamey, who's got a full background in applying analysis to graph networks. And then Pablo, who's coming from a philosophical background, a cognitive science. And then the three disciplines that come together to kind of look at this problem, think about it and make an interesting discovery that requires all these different ways of looking at the data. We now have insights to how humans plan. You've di-bidened stuff to Google maps often for most people and it tells you where to go. But I think there's an interest in thinking about how can we skill people up in the population to navigate or independently. It is better if you can keep your brain active and engaged and think about the environment around you for brain health. Here's where I think about the fact that these taxi drivers are doing this navigation by eating a banana. To get up in the morning, have breakfast, eat a banana, whatever it might be, and solve all these insanely complicated routing problems, whereas the data centers that run the AI, they're any more than a banana. They're run one and they're non-organic. How can you make these things super efficient? What are the caveats or limitations of the study? This is a very specific culture of navigation, and these are expert navigators that over a long time have developed this culture is adapted to London, to an urban environment. So I think it's very important to keep in mind that there are many expert navigators in very different cultures, by the many in the university cultures, they're expert navigators. And there is the possibility that they will have a different approach that is adapted to those environments. The planning here is not dynamic, they do plan at a particular time of a particular day of the week. So maybe Monday, 10 AM or whatever it was. So in real situations, the taxi drivers could be planning in very specific times, and they have to adapt to changing traffic conditions. So that's something also to be explored compared to real world planning. [Music]