London taxi drivers could hold the key to better AI route mapping in the future, according to a study published in the journal PNAS. A team of researchers at the University of York, in collaboration with University College London and the Champalimaud Foundation, has uncovered that they don’t follow the same method as computer satnav, which calculates every possible route. Taxi drivers plan each route by prioritising the most challenging areas first and filling in the rest of the route around these tricky points.
London taxi drivers are famous for knowing more than 26,000 streets across the city. To understand how these drivers plan routes, the team measured the time it took them to plan different journeys to various destinations in the capital city.
The results show that taxi drivers use their cognitive resources much more efficiently than current AI systems. Learning about expert human planners can help this technology develop in several ways.
“London is incredibly complex, so planning a journey in a car ‘off the top of your head’ and at speed is a remarkable achievement. If taxi drivers were planning routes sequentially, as most people do, street-by-street, we would expect their response times to change significantly depending on how far they are along the route,” said Dr Pablo Fernandez Velasco, British Academy Postdoctoral Fellow at the University of York. “Instead, they look at the entire network of streets, prioritising the most important junctions on the route first, using theoretical metrics to determine what is important. This is a highly efficient way of planning, and it is the first time that we are able to study it in action.”
“The development of future AI navigation technologies could benefit from the flexible planning strategies of humans, particularly when there are a lot of environmental features and dynamics that have to be taken into account,” added Dan McNamee from the Champalimaud Foundation. “Another way to enhance these technologies would be to integrate the information about human experts into AI algorithms designed to collaborate with humans. This is a very important point because if we want to optimize how an AI algorithm interacts with a human, the algorithm has to ‘know’ how the human thinks.”
Velasco P, Griesbauer E, Brunec I terms al. Expert navigators deploy rational complexity–based decision precaching for large-scale real-world planning, Proc. Natl. Acad. Sci. U.S.A. 122 (4) e2407814122, https://doi.org/10.1073/pnas.2407814122 (2025).