AI Is Cognitive GPS: What London Taxi Drivers Tell Us About Our Brains

Neuroscientist Eleanor Maguire and her team made a striking discovery in 2000 while studying London taxi drivers.

Licensed London taxi drivers have to memorise the layout of 25,000 streets and the locations of thousands of places of interest. They spend 3–4 years preparing for "The Knowledge"; an exam that tests their ability to navigate London without GPS. When Maguire's team scanned their brains, they found something remarkable: the posterior hippocampi of taxi drivers were 3.5% larger than those of control subjects Maguire et al., PNAS, 2000. The larger the hippocampus, the more years the driver had spent navigating. The correlation was clear: the brain was physically changing in response to the work.

A follow-up study deepened the finding: when taxi drivers left the profession, their hippocampi shrank Woollett et al., 2008. The brain had been built by years of navigation — and unmade when the navigation stopped.

Then came the killer detail. A 2020 study of regular drivers found that people with greater lifetime GPS use performed worse on self-guided spatial-memory tasks. In a smaller follow-up sample tracked over three years, increased GPS reliance was associated with a steeper decline in hippocampal-dependent spatial memory Dahmani & Bohbot, Scientific Reports, 2020. The authors cautioned that the longitudinal sample was small—but the pattern was difficult to ignore.

Two people navigating the same city. Different technology. Different brains.

Wayfinding vs. Mapping

The difference between GPS users and non-GPS taxi drivers isn't just about efficiency. It's about the kind of knowledge each builds about the city.

The non-GPS drivers didn't just know their routes — they built rich, internalized maps. They understood relationships between neighborhoods. They knew shortcuts that GPS couldn't suggest because they weren't on any map. They had a model of the territory.

The GPS drivers knew where to turn. But they didn't know the city.

This distinction — wayfinding versus mapping — is the key to understanding what's happening with AI.

Wayfinding is about getting to the destination efficiently. You follow instructions. You don't need to remember the path because the instructions will always be there.

Mapping is about building an internalized model of the territory. You remember relationships between places. You learn shortcuts. You develop intuition about where things are and why.

GPS optimizes for wayfinding. Non-GPS navigation builds mapping. And the brain physically changes depending on which you do.

Most people think about AI as a speed tool. The taxi driver studies suggest it's really a depth tool: each technology shifts you along a spectrum from deep, internalized knowledge to efficient, surface-level wayfinding.

AI Is Cognitive GPS

Now imagine this for writing, coding, and reasoning.

In 2025, researchers at MIT’s Media Lab reported that people who wrote essays with ChatGPT struggled to recall their own writing minutes later. In one session, 83.3% of participants in the AI-assisted group could not accurately quote a single sentence from their essays—compared with substantially better recall among people who used a search engine or wrote without tools. Kosmyna et al., 2025.

AI is to thinking what GPS is to navigation: it gets you to the answer faster, but the mental work you do along the way is different. Or absent.

When you ask an AI to write an email, you get the email in seconds. But you didn't work through the choices: which detail to emphasize, which tone to strike, which sentence to cut. When an AI assistant generates code for you, you get working code fast. But you didn't struggle through the algorithm, test edge cases, or learn why one approach is better than another. Each choice you outsource is a turn you don't remember taking.

The Friction We're Removing

Why does the difference matter? Because the cognitive benefit of a task isn't just the answer — it's the work of getting there.

When AI removes the friction from a task, it changes what kind of memory gets encoded. You still get the output. But you lose the mental work that turns that output into understanding.

This is what's happening with AI-assisted writing: you get the email, the essay, the report. But the choices you would have made — the small decisions that build your sense of voice and structure — are made by the AI. The result isn't laziness. It's a different kind of learning, one that's shallower.

The Compounding Effect

Every generation of AI users will likely have a different cognitive architecture than the generation before. We're watching it happen right now. The first generation of "AI natives" — people who grew up with AI as infrastructure rather than novelty — will build different maps than we did. We just don't know what those maps will look like yet.

The question about AI isn't "does it make us smarter?" It's "what kind of cognitive map are we building, and does it matter?"

The taxi drivers who stopped driving didn't just forget the city — their brains physically changed. Their hippocampi shrank. The memory of the territory left their heads, and the head forgot what it used to hold.

AI may work the same way. Not by making us less intelligent, but by quietly changing which forms of intelligence we practice, preserve, and eventually lose.

The danger isn't that AI will erase the map. It's that, one shortcut at a time, we may stop noticing which parts of the map we no longer know how to draw.