Eric So, a professor at MIT Sloan, has a term for what's happening to us. He calls it "AI gravity" — the constant, invisible pressure to outsource more of your thinking to machines. Not because you need to. Not because the task demands it. But because the path of least resistance — the easy button — is always, always there.
In his forthcoming book The Collision: What AI Does to Us, So argues that AI isn't just changing how we work. It's changing how our brains operate. And the direction of that change isn't toward greater capability — it's toward managed decline.
He's not alone. In May 2026, Computers in Human Behavior Reports published a cross-domain review of over 40 studies on AI overdependence. The conclusion? The evidence for AI-induced cognitive decline is now robust across education, professional work, and healthcare. We're not speculating anymore. We're measuring.
The finding that keeps researchers up at night: The more positive your attitude toward AI, the worse your ability to distinguish between real and AI-generated content. A February 2026 Nature study on 295 participants found that people who trusted AI the most were the least capable of spotting when it was wrong.
What We're Actually Losing
The data paints a clear picture of what declines when AI does the thinking:
| Capability | What the Research Shows |
|---|---|
| Critical thinking | Measurable decline in analytical effort and independent judgment (Microsoft/CMU, 2026) |
| Creativity | Narrower range of ideas produced — users converge on AI's suggestions (Cambridge, 2025) |
| Memory retention | Significant reduction in recall of work performed with AI assistance (Microsoft, 2026) |
| Decision ownership | Users struggle to explain or defend AI-assisted decisions — can't trace the reasoning (Stanford, 2025) |
| Pattern recognition | Decreased ability to spot errors in AI output — blind trust replaces verification (Nature, 2026) |
The Second-Order Effect Nobody's Tracking
Here's the part that really bothers me. We're not just losing individual cognitive skills. We're losing something more fundamental: the ability to know what we don't know.
When you solve a problem yourself, you build a mental model of the domain. You learn where the edge cases are. You develop a feel for when an answer doesn't smell right. That "smell" — that intuitive sense of wrongness — is one of the most underrated human capabilities. And it's the first thing to atrophy when you outsource problem-solving to AI.
The Dunning-Kruger effect is well-documented: people with low ability in a domain tend to overestimate their competence. What happens when an entire generation has never had to develop competence in the first place? We're looking at a world where everyone is confidently wrong, and nobody has the tools to realize it.
"We are increasingly deferring tasks that our brains are meant to handle to AI systems that think for us, write for us, and create on our behalf. Each time we engage in this sort of cognitive outsourcing, we're participating in dramatic societal change."
— Eric So, MIT Sloan
The Pipeline Problem
There's a pernicious generational angle to this that most commentary misses. The World Economic Forum predicts that 22% of current jobs will be disrupted by AI by 2030. But the entry-level roles — the ones that traditionally trained people to think — are the ones being automated first.
Research analysts used to spend their first year pulling data, finding patterns, and learning to ask the right questions. Now they prompt an AI agent. Junior developers used to cut their teeth on debugging, reading code, understanding why things break. Now they ask an AI to fix it and move on.
The pipeline for developing senior professionals is quietly being dismantled. Not by layoffs, not by offshoring — by AI. And by the time we notice the gap, there won't be anyone left who knows how to fill it.
Question for debate: If every entry-level analyst uses AI to synthesize research, who learns how to do research? At what point does "efficiency gain" become "talent pipeline collapse"?
This Isn't Luddism
Let me be absolutely clear: I'm not arguing against AI. I'm arguing against unconscious dependence.
The calculator didn't destroy mathematics — it changed what mathematicians did with their time. But AI is different in a critical way: it doesn't just automate a narrow mechanical function. It simulates the cognitive function itself. And when you outsource thinking, you don't get the thinking back. You get the output. The output is not the same thing.
We're building a world where expertise is optional. Where you can be a "senior" professional who has never made a senior-level decision without a machine telling you what to decide. Where the people running organizations have spent their entire careers with a digital co-pilot.
And the most dangerous thing about this world? We won't notice we're in it until we're too far in to get out.
The question isn't whether AI makes us individually dumber in any given moment. The question is what kind of society we're building when we systematically remove the need for human judgment from the systems that matter most.
And if your first instinct was to ask an AI what you think about this article — well, you just proved my point.
Have you felt your own thinking changing since adopting AI tools professionally? How do you maintain your "manual flying hours"?
DM this to a colleague and ask them the same question. The best debates happen in the comments.