Fourth in a series on AI at work. This one’s less about the workplace and more about us.

I listened to a podcast interview with Dr Norman Swan recently and ordered his new book, The Brain Cliff, before it had finished. It’s currently on the bedside table in the queue of books waiting for me to dive in.

Partly I bought it because brain health is a long-standing interest of mine. Partly, if I’m honest, because I’m middle-aged and the topic has started to feel less theoretical than it used to.

The bit that snagged me wasn’t about diet or exercise. It was the idea that one of the things that makes retirement risky is that we stop doing hard thinking. The job goes, and with it the daily requirement to work out things you don’t already know how to do. 

I’ve spent time recently writing about getting people to adopt AI at work. So a slightly uncomfortable thought arrived while I was waiting for the book to turn up. What if the thing that makes retirement risky is something we’re now doing to ourselves, in middle age, on purpose, with considerable enthusiasm?
I went looking to see whether there was anything in it.

The evidence on demanding work is better than I expected

I assumed I’d find something vague and correlational. What’s actually there is more solid than that, and it isn’t really about retirement at all.

A 2024 study in Neurology, drawing on Norwegian data, found that people with the least mentally demanding jobs had a 66% greater risk of mild cognitive impairment and a 31% greater risk of dementia after age 70, compared with those in the most demanding roles.
Adjusting for age, sex and education, the low-demand group had 37% higher dementia risk, and this study used occupational registry data rather than asking people to recall their own working lives, which is a meaningful improvement on much of what came before (read more on this).

A separate analysis of 10,195 people across six countries found that high occupational complexity was associated with a 19% increase in dementia-free survival time, after controlling for education (read more on this).

A large multi-cohort study in the BMJ found lower dementia risk among people in cognitively stimulating jobs, and went looking for the mechanism in blood, finding lower levels of proteins that inhibit the growth of new neural connections (read more here).

And it isn’t only about solo problem-solving. The US National Institute on Aging has pulled together several studies on work complexity and cognitive ageing, including one finding that complex work with people improves episodic memory and promotes brain reserve (National Institute on Aging).

The explanation people reach for is cognitive reserve. The idea is that cognitively enriching activity builds a more efficient neural network, so that even with a significant amount of Alzheimer’s-related pathology present, other pathways remain and different parts of the brain can still communicate with each other (read more).

Put plainly: difficult work appears to build something that protects you later.

The retirement part is murkier

I should be fair about this, because it’s where I started. The retirement evidence is real but mixed.

A Swedish twin study found no memory decline before retirement and significant decline after it, with the drop in processing speed more than doubling post-retirement (read more here).

A Canadian study comparing matched retirees and workers over three years found significant decline on executive function tasks for the retirees, and European SHARE data found a negative but modest effect of retirement on cognition overall (explore this further).

But the counter-evidence is real. An English study of adults concluded there was no cognitive benefit to remaining at work, at least as measured by episodic memory, though it did find that people in manual occupations experienced more rapid memory decline over their working lives than those in higher occupational classes, and suggested employers consider cognitive enrichment programmes for them (explore further).

And a recent meta-analysis found retirement wasn’t associated with global cognitive function at all, with only a slight decrease in memory-related tasks (more detail here).

So: unresolved. Which is fine, because it turns out not to be the important part.
 

The bit nobody has measured

Here’s where I’ve landed. The well-supported finding isn’t about retirement. It’s about cognitive demand.
Complex, effortful, decision-heavy work seems to protect the brain. Low-demand work doesn’t.

And we have just built a technology whose entire value proposition is reducing the cognitive demand of work.

Not the easy parts, either. We don’t outsource the things that are already automatic. We outsource the blank page, the argument that won’t come together, the analysis that needs several things held in your head at once, the email that’s been sitting there because it needs careful wording. That is precisely the effortful thinking the research is pointing at.

So what does the evidence say about AI specifically? Not much yet, honestly.

A 2025 study of 666 participants found a significant negative correlation between frequent AI tool use and critical thinking, mediated by cognitive offloading, with younger participants showing higher dependence on AI and lower critical thinking scores than older ones (more detail).

There’s an older precedent too. Research from 2011 on search engines found that heavy reliance shifted people toward remembering where to find information rather than the information itself — the thing that got labelled the Google effect, or digital amnesia (read more here).

But I’d be careful with all of it, and so are the researchers. Much of this literature relies on cross-sectional self-report designs and attitude-based measures rather than direct, unaided reasoning performance — what people believe about their own behaviour, rather than what they can actually do (full article here).

Cognitive effects play out over decades. This technology has been in most people’s hands for about three years. Anyone telling you confidently that it’s rotting your brain, or that it’s perfectly fine, is ahead of the evidence in both directions.

The honest position is this: we have good evidence that cognitively demanding work protects the brain, we have a technology explicitly designed to remove cognitive demand from work, and nobody has connected the two yet. We’re running the experiment on ourselves in the meantime.
 

So do the workouts

What I took from the interview (and what I gather is the thread running through the book) is that none of this is fatalistic. Swan’s argument is that brain ageing doesn’t have to be about decline and deficit, but about how we help our brains adapt, become resilient and maintain cognitive function — and that it depends on considerably more than what we eat, how we exercise, or our genes. He went and found people in their nineties and hundreds who are still cognitively remarkable, and looked at what they had in common.

Which reframes the whole question for me. It isn’t whether to use AI at work. That ship has sailed, and I’d rather have the hours back. It’s what happens to the effort that’s been displaced.

The muscle comparison isn’t physiologically precise, but it’s useful: you can take the lift every day and still be strong, provided you’re deliberately doing something hard elsewhere. What you can’t do is take the lift every day, do nothing else, and expect the strength to still be there when you need it.

The researchers land in much the same place. If your job isn’t particularly complex, cognitive reserve can be built other ways, such as further education and hobbies, with lifelong learning consistently near the top of the list (read more).

So: learn the instrument badly.
Learn a language.
Read something difficult without summarising it first.
Argue properly with someone who genuinely disagrees with you.
Take up the thing that requires you to be a beginner again.
And given what that study on complex work with people found, make some of it social rather than solitary.

Not as a wellness activity. As maintenance.
 

The part that needs saying out loud

We are adopting this technology faster than anything I’ve seen in twenty-odd years of this work, and nearly every conversation about it is framed around output. Faster, more, less effort.

Almost none of it asks what thirty years of less effort does to us.

I don’t think the answer is to use it less. I think the answer is to notice, and to be deliberate about where the hard thinking goes instead, because it won’t happen by accident, and by the time the effects show up, they’ll have been accumulating for a very long time.

Anyway. The book is in my to-do list. I’ll read it properly, without asking anything to summarise it for me, and I suspect I’ll have more to say afterwards.