How Do You Know Where You Are Without Looking? A Map in the Head

Edward Tolman argued in the 1940s that rats learning a maze were not simply acquiring a chain of turns but building something like a map, and the evidence was that they could use it flexibly. Rats trained to run a particular route to food, when that route was blocked and a fan of new paths was offered, chose the one heading towards where the food had been rather than the one nearest their old route. Rats allowed to explore a maze with no reward at all learned it anyway, showing that as soon as reward was introduced, which meant they had been acquiring knowledge without any reinforcement. Both results were awkward for the behaviourist position that learning consists of reinforced responses.

The capability shows itself in behaviours that a memorised route cannot explain:

• •Taking a novel shortcut between two familiar places never directly connected
• •Detouring correctly when a habitual route is blocked
• •Pointing towards an unseen landmark from an unfamiliar position
• •Estimating distances between places by a route never travelled
• •Learning a layout without any reward for doing so
• •Recognising that two routes lead to the same place

The neural basis was found in stages over four decades. John O'Keefe reported in 1971 that certain cells in the hippocampus of a rat fire when the animal is in one particular location and are silent elsewhere, so the population encodes position directly. May-Britt and Edvard Moser reported in 2005 that cells in a neighbouring region fire at multiple locations arranged in a regular triangular grid across the whole environment, which provides a coordinate system rather than a set of landmarks. Further types have been identified that encode heading direction, proximity to boundaries, and distance travelled. The three researchers shared the Nobel Prize in 2014, and the work established that the representation Tolman inferred from behaviour has a specific physical form.

Satellite navigation appears to make internal spatial knowledge unnecessary, and research on whether it degrades has produced findings worth taking seriously without overstating them. People navigating by turn-by-turn instructions learn the layout of an area measurably less well than people navigating by map or by exploration, which is consistent and unsurprising, since following instructions requires attending to the next turn rather than to where things are. Brain imaging shows the regions involved in spatial representation being less engaged during instruction-following. Studies of London taxi drivers, who memorise an enormous street layout, found structural differences in the hippocampus associated with the training. Whether habitual device use causes lasting change rather than simply reflecting what is practised remains genuinely unsettled.

Human internal representations of space are systematically distorted rather than accurate, and the distortions are consistent enough to be catalogued. Distances are judged as longer when the route contains more turns or more landmarks. Angles between streets are remembered as closer to right angles than they are. Places within a recognised region are judged closer together than equally distant places across a boundary, so people misjudge which of two cities lies further north when a national border intervenes. Routes are remembered as more symmetrical than they are. Familiarity increases detail without increasing accuracy. The representation is therefore better described as a collection of related fragments than as a map, which is why the metaphor misleads if pushed.

Rats choosing a novel path towards a remembered goal, and learning mazes without reward, indicated an internal spatial representation rather than a chain of memorised turns. Cells firing at one location, and others firing on a triangular grid across a whole environment, were found decades later and won a Nobel Prize in 2014. Human versions are systematically distorted, with turns lengthening remembered distance and borders exaggerating separation.