The Cognitive Treadmill We're All Running On

There's a metaphor I keep coming back to: AI as a forklift for your brain. You can still lift the weight. You just won't get stronger. And the more we normalize using the forklift, the more we quietly redefine what it means to be a capable thinker.

DeepMind CEO Demis Hassabis said something striking last week. AI, he warned, can sharpen your critical thinking skills — or dull them. The difference depends entirely on how you use it. This wasn't a Luddite warning about technology. It came from someone building the most advanced AI systems on Earth.

The phenomenon has a name: cognitive offloading. It happens when you delegate mental tasks to external tools. GPS navigation offloads spatial reasoning. Calculators offload arithmetic. AI chatbots, increasingly, offload the entire process of reasoning through a problem.

Each offload is individually harmless. But they compound. And new research suggests the compound effect is larger than we thought.

The Evidence Is Piling Up

A study published earlier this year in Societies found something researchers weren't expecting. People who used AI tools more frequently demonstrated measurably weaker critical thinking abilities. The correlation held even when controlling for education, age, and baseline cognitive ability.

The mechanism wasn't mysterious. It was cognitive offloading. The more people delegated their thinking to AI, the less they practiced thinking for themselves. The less they practiced, the weaker the skill became. The weaker the skill became, the more they relied on AI. A feedback loop, spiraling in the wrong direction.

Younger participants showed the strongest effect. The generation coming of age with AI assistance is also the generation showing the steepest decline in independent reasoning. This isn't because they're less intelligent. It's because they're getting less practice.

Think about what critical thinking actually requires. Evaluating evidence. Holding multiple possibilities in working memory. Detecting logical inconsistencies. Constructing counterarguments. These aren't innate talents. They're skills developed through repeated use, like muscles developed through repeated lifting.

The gymnasium metaphor deserves to stick. Every time you skip the workout because a machine can lift for you, the muscle atrophies slightly. Not enough to notice day-to-day. But over months and years, the cumulative effect is dramatic.

What We're Actually Losing

The risk isn't that people will become helpless without AI. Humans adapt. We'll develop new competencies around AI management, prompt engineering, information synthesis. Some of these are genuinely valuable skills.

The deeper risk is subtler: the erosion of the reasoning process itself.

When you use a calculator, you still see the numbers. You still understand what addition means. The calculator handles the mechanical computation, but you retain the conceptual understanding. The offloading is partial.

With AI, the offloading can be total. You describe a problem. AI generates an answer, complete with reasoning. You read it. It sounds plausible. You move on. At no point did your brain actually work through the logic. You consumed the conclusion without participating in the reasoning that produced it.

Do this enough times and something strange happens. You lose the ability to verify whether the reasoning was sound. You can spot obviously wrong answers — sometimes. But subtle errors, unstated assumptions, logical gaps? Those slip through. Your verification muscles have atrophied.

This matters for consequential decisions. Medical information. Financial planning. Political reasoning. Career choices. These all require evaluating complex arguments, not just receiving complex conclusions. And the people making them are increasingly doing so with degraded verification skills.

A doctor using AI to diagnose might miss the subtle pattern that doesn't fit the algorithm's training data. An investor relying on AI-generated financial analysis might not recognize when the model is confidently extrapolating from a historical pattern that no longer applies. A voter consuming AI-summarized political positions might lose the ability to detect when nuance has been flattened into false equivalence.

The errors aren't always obvious. In fact, the most dangerous failures are the ones that look correct. An AI that generates plausible-sounding but flawed reasoning is more dangerous than one that's clearly wrong, because you'll trust it. You'll act on it. And you'll never know you were led astray by reasoning you didn't verify.

The Convenience Trap

There's a seductive argument in favor of offloading: efficiency. Why spend mental energy on something AI can do faster and often better? Isn't this just another technology raising human capability, like writing or the printing press?

The comparison fails. Writing and printing preserved knowledge and extended reach. They didn't replace the process of thinking. A book gives you access to someone else's reasoning. It doesn't reason for you. You still have to read, comprehend, evaluate, integrate.

AI can replace the reasoning step entirely. Ask it whether a particular policy is good. It will give you arguments, evidence, a conclusion. The whole cognitive workflow, outsourced.

This isn't an argument against using AI. It's an argument for understanding what you're trading. Every time you offload reasoning that you could have done yourself, you're choosing short-term convenience against long-term capability. You're taking a withdrawal from your cognitive fitness account.

Sometimes that's the right choice. Not every decision merits deep reasoning. Not every problem is worth solving from first principles. The danger isn't occasional offloading. It's habitual offloading. It's reaching for the forklift before trying to lift.

The Expertise Paradox

Here's where it gets complicated. Research suggests that education and self-confidence provide some protection against cognitive offloading. People with stronger baseline critical thinking skills and higher confidence in their own judgment are less likely to delegate their thinking to AI.

This creates a troubling paradox. The people most equipped to use AI wisely — those with strong reasoning skills — are also the most likely to maintain those skills by using AI sparingly. The people least equipped — those with weaker skills — are most likely to offload everything, accelerating the atrophy.

The gap widens. Those who can think critically think critically more often, getting stronger. Those who struggle think critically less often, getting weaker. AI becomes a magnifier of existing cognitive inequality.

This isn't the technology's fault. It's a usage pattern problem. But usage patterns are notoriously hard to change, especially when the easy path feels so natural.

What Healthy Usage Looks Like

The question isn't whether to use AI. It's how to use it without degrading the underlying capability. There are better and worse patterns.

Use AI for information, not reasoning. Ask it for facts, sources, perspectives you haven't considered. But do the synthesis yourself. Don't ask "what should I conclude?" Ask "what evidence am I missing?"

Force the verification step. When AI gives you an answer, don't accept it. Argue with it. Find the weak points. Ask yourself: what would the strongest counterargument be? If you can't generate one, you haven't understood the position — you've just memorized it.

Reserve the hardest problems for human cognition. The problems that stretch you are the problems that strengthen you. Using AI for easy tasks and reserving human effort for genuinely hard thinking maintains the development gradient that skills require.

Practice switch-side thinking. Before accepting any AI-generated conclusion, spend five minutes constructing the strongest possible argument against it. This isn't being contrarian. It's being complete. An argument you can only see from one side is an argument you don't fully understand.

Maintain the dialogue habit. Talk through problems with humans, not just machines. Human conversation forces you to articulate your reasoning, defend it in real-time, adjust when challenged. These are the conditions under which critical thinking actually develops.

Track your offloads. For one week, notice when you reach for AI help. Ask yourself: could I have done this myself? Was I choosing efficiency or avoiding effort? Just paying attention changes the pattern.

Revisit old problems without assistance. Take something you solved with AI last month. Try to solve it again, alone. Notice where your reasoning feels rusty. That's valuable diagnostic information about which muscles need work.

The Uncomfortable Truth

There's a deeper question here that most discussions avoid. What if the skills we're losing — extended concentration, independent reasoning, working through confusion without external assistance — are becoming obsolete? What if AI really is a forklift, and the future belongs to those who learn to operate it rather than those who maintain the ability to lift?

Maybe. But consider what you'd be optimizing for. A world where humans manage AI outputs without participating in the reasoning that produces them is a world where humans are progressively removed from understanding. Understanding atrophies just like any other skill. And once it's gone, it's hard to rebuild.

The ability to reason independently isn't just a professional skill. It's a civic skill, a personal skill, a protective skill. It determines whether you can evaluate medical advice, financial claims, political arguments, or personal choices. Outsourcing all of that to AI because it's more convenient is a trade that looks rational in the moment and catastrophic in retrospect.

Demis Hassabis was right. AI can sharpen your thinking or dull it. The difference is usage. The technology is neutral. What matters is whether you treat it as a tool that extends your capability or a replacement that removes the necessity of capability.

The gymnasium remains open. The weights are still there. Nobody's stopping you from lifting them. But the forklift is parked right outside, engine running, and it's getting easier every day to convince yourself that lifting is a waste of energy.

The measure of your next six months won't be what you accomplished with AI's help. It will be what you can still accomplish without it.