Why Taxi Drivers Rarely Die of Alzheimer's — And What Your GPS Might Be Costing You

When was the last time you drove somewhere without punching in an address? I turn on navigation to reach a grocery store four blocks away. That reflex might be leaving a mark on your brain decades down the line — or so goes an argument that keeps resurfacing, kicked off by a study with a hook too good to ignore: taxi drivers don’t seem to die of Alzheimer’s.

Let me be upfront. This isn’t a fresh controversy blowing up on Hacker News this week. It’s more of a perennial — a debate that gets exhumed every year or two, usually when someone rediscovers the London cabbie research. So instead of tracking the discourse, let’s look at what the studies actually say, and where they fall apart.

Two Jobs Out of 443

The study everyone cites came out of Harvard Medical School, published in the BMJ’s 2024 Christmas issue. That issue has a tradition: rigorous research on subjects that are a little bit fun. This one fit perfectly.

The researchers combed through roughly 9 million US death certificates spanning 1992 to 2022. They sorted the dead into 443 occupations and compared Alzheimer’s death rates. The overall average was about 3.9%. Taxi drivers came in at 1.03%. Ambulance drivers at 0.74%. Roughly a quarter of the baseline.

The control group is where it gets interesting. Bus drivers and airline pilots also spend their working lives behind a wheel or a yoke. Their Alzheimer’s death rates: 3.11% and 4.57% — statistically unremarkable. Which points to a hypothesis: it isn’t the driving. It’s the real-time route computation. Buses run fixed loops. Flight paths are filed before takeoff. Taxis and ambulances get a new destination every time and have to rebuild the optimal path from scratch, in their heads, under time pressure.

London Cabbies Really Do Have Bigger Hippocampi

The hypothesis leans on a much older and much more famous piece of research. In 2000, Eleanor Maguire’s team at University College London ran MRI scans on London black cab drivers.

To get a London taxi license, you have to pass an exam called the Knowledge. Its reputation is deserved. Candidates memorize roughly 25,000 streets and 20,000 landmarks within a six-mile radius of Charing Cross. Average preparation time is three to four years. Fewer than half of those who start ever finish.

Maguire’s team found that these drivers had a significantly enlarged posterior hippocampus compared to controls. The hippocampus handles memory and spatial cognition — and it’s the first structure Alzheimer’s dismantles. More striking: the longer someone had driven, the larger the region. That tilted the interpretation away from self-selection. It wasn’t that big-hippocampus people become cabbies; the job appeared to grow the tissue. Follow-up work tracked trainees before and after the exam and confirmed it — only those who passed showed the growth.

It became the textbook demonstration that adult brains restructure themselves through use. Every neuroplasticity survey course includes it.

Now the Part Where We Slow Down

Time to pump the brakes. The BMJ authors explicitly warned in their own paper against reading causation into it. The holes are worth walking through.

First, this is death certificate data. It measures the share of deaths attributed to Alzheimer’s, not disease prevalence. If taxi drivers die earlier from cardiovascular disease or road accidents, they never reach the age when Alzheimer’s typically manifests. Statisticians call this competing risk, and it can produce exactly this pattern out of nothing.

Second, reverse causation. Someone in the early stages of cognitive decline doesn’t take a job that demands complex wayfinding — or quits the one they have. It may be that healthy brains sustain long taxi careers, not that taxi careers build healthy brains.

Third, occupational coding is crude. A death certificate records one job: the last one. Someone who spent 30 years as an accountant and drove a cab for three years after retiring gets counted as a taxi driver.

Fourth, confounders went largely unaddressed. Income, education, smoking rates, frequency of social contact — all vary wildly by occupation. Spending your entire shift in conversation with strangers might build cognitive reserve on its own.

And critically: the study never distinguished drivers who use GPS from those who don’t. The back half of that 30-year window is squarely in the smartphone navigation era. You cannot get “GPS wrecks your brain” out of this dataset. It simply isn’t in there.

But Cognitive Offloading Is a Real Problem

The BMJ study failing as causal evidence doesn’t void the navigation-and-brain question. That has its own separate research lineage.

Véronique Bohbot’s team at McGill reported that drivers with heavier GPS dependence had less hippocampal gray matter — and over three years of follow-up, lost it faster. In a separate experiment, participants who followed turn-by-turn directions were dramatically worse at sketching routes through the same area afterward than participants who had navigated on their own.

Cognitive science calls this cognitive offloading. When you delegate storage to an external device, your brain declines to hold the information itself. The photo-taking impairment effect — people remember museum objects worse after photographing them — belongs to the same family, as does the Google effect, where subjects remember where to find a fact rather than the fact.

There are two ways a brain finds its way somewhere. One is the spatial strategy: building a map in your head. That runs on the hippocampus. The other is the response strategy: follow instructions, right at the convenience store, second left. That runs on the caudate nucleus. The moment you start navigation, you switch to mode two automatically. You arrive precisely on time. Your hippocampus did nothing all day.

Why This Keeps Getting Dragged Back Up in the AI Era

The reason a 25-year-old MRI study is trending again in 2026 is that navigation is no longer the only thing we’ve outsourced.

AI writes the code. It summarizes the meeting, drafts the email, does the background research. There’s a specific anxiety that shows up constantly in developer forums: after several months on AI coding tools, people report they can’t sit with a hard problem the way they used to. The worry got louder after MIT Media Lab published an essay-writing study in 2024 showing weaker brain connectivity metrics in the LLM-assisted group. Worth noting, though: n=54, and it went public without peer review. It’s a data point, not a verdict.

Honestly, nobody knows yet. Calculators didn’t eliminate mathematicians. Socrates warned that writing would destroy human memory, and writing turned out to be the thing that let us go further than memory ever could. Whether a tool replaces cognition or frees it for higher-order work depends almost entirely on how you use it.

So What Do You Actually Do

We’re not at the stage where anyone can prescribe a scientifically validated prevention protocol. But there are habits with no obvious downside.

Turn the navigation off sometimes in neighborhoods you know. Heading somewhere new? Glance at the map before you leave and rough out the shape of the route in your head. Give yourself 30 seconds to think before you read the AI’s answer. Muscles grow under load, and the accumulated evidence points the same way for brains.

What the taxi driver study left behind isn’t a conclusion. It’s a question. For everything that got easier, what did we stop using? And of those things, which ones should we keep using anyway? I’m driving home without the GPS tonight.