The AI Divergence: How Heavy Users Are Actually Splitting Apart

TL;DR: A friend who uses AI 6+ hours daily told me he feels like he's getting dumber. He's wrong — but he's not wrong. Four things are happening simultaneously: three are losses, one is a gain. The ratio between them determines whether AI makes you sharper or flatter. And that ratio is a function of how you use it, not that you use it.

James here, CEO of Mercury Technology Solutions. Hong Kong — July 20, 2026

A friend who spends six hours a day with AI recently told me: "I think I'm getting stupider."

We talked for an hour. I came to the opposite conclusion — but with a critical caveat. He's not getting stupider. Four things are happening to him at once. Three are losses. One is a gain. The question isn't whether AI makes you dumb. The question is: what's your ratio?

Before I break this down, I need to give you a mental tool. Everything that follows rests on it.

The Two Judgments

When facing any change, humans have two ways to respond: value judgment and mechanism judgment.

Value judgment asks: Is this good or bad? Am I getting smarter or dumber? Progressing or regressing? It wants a verdict.

Mechanism judgment asks: What is actually happening here? What pieces are moving, along what paths, and how do they relate? It wants a structural map.

These aren't shallow vs. deep. They're two different cognitive operations. And humans have a deep bias: our first reaction to any change is value judgment. We want to evaluate it — good or bad, right or wrong, satisfying or not. Because value judgment is cheap. Your brain spits out a conclusion in seconds, and you get to pick a side. Emotionally satisfying.

Mechanism judgment is expensive. It demands you slow down, resist conclusions, and dissect a vague whole into inspectable parts.

Treating mechanism judgment as value judgment is a classic form of intellectual laziness. You think you're thinking. You're actually just signaling.

"Is AI making people stupid?" is a value-judgment question. Once you accept its framing, you've already lost — whether you answer yes or no. Today, I'm smashing that frame open and mapping what's actually underneath.

You'll find four mechanisms.

Mechanism 1: Specific Capability Atrophy

Adults frequently "pick up the pen and forget the character." I was having dinner with a professor friend at an old-school restaurant where you write your own order. He said, "Let's add the braised bass." Then paused. "I can't write the character for bass." Someone ribbed him: "And you're a PhD advisor?" The waitress smoothed it over: "Professor, you use computers for everything now."

She was right. Use computers long enough, you gradually forget how to write. But did you get dumber? No. Your capacity to think, analyze, and judge didn't degrade because you can't hand-write "bass" or "hand-torn cabbage." Trump tweets with grammatical errors. That doesn't make him stupid. The connection between specific skill loss and intelligence is systematically overestimated.

The real question: why is it overestimated?

Lagged perception. Twenty years ago, a taxi driver who didn't know the streets was considered incompetent. Today, ride-share drivers in major cities use GPS navigation. It doesn't diminish their function. The things that once defined your professional competence gradually stop mattering, but many people remain stuck in old frameworks.

So most specific capability atrophy has nothing to do with stupidity. Today's intellectuals can't write calligraphy or read seal script. When you call that "getting dumber," you're making a value judgment. But it's actually a mechanism judgment. Someone who mistakes mechanism for value judgment is, ironically, demonstrating the very thing they claim to be measuring.

Everything AI can do for you, you will grow rusty at. That rust is specific capability atrophy. Not stupidity.

Mechanism 2: Action Pathway Shift

When people encounter problems, they have default response paths — often unconscious.

In kindergarten, when you didn't know something, you naturally asked your parents or teacher. In university, you stopped using that path. You knew those questions were beyond their reach. My professor friend told me classroom "head-raise rates" have collapsed. A good lecture used to hold 70–80% of students' attention. Now, regardless of instructor quality, 30% is considered high. Why? Students don't need you. AI is more capable than most professors.

When I was in university, I searched engines for answers. Now I barely use them. Some friends still do. The shift is a process. This is action pathway shift: from asking people, to flipping books, to figuring it out yourself, to asking AI.

As asking AI becomes frictionless, many people skip the "let me think about this first" step. Research shows that when AI tutors children, their grades don't improve. Because they haven't struggled yet — they see the answer too early.

The key: they don't realize that "thinking for fifteen minutes first" is an option.

Here's a strange example. Some religious people pray before meals. That ritual — or any ritual, like carefully wiping the table with a napkin before eating — doesn't require religious belief. Many people don't realize this is optional. You can think of "don't check the answer too early" as a similar ritual. You know it because you grew up without answers. Today's children gradually lose that awareness. Their thinking起点 and终点 have shifted.

Mechanism 3: Calibration Source Drift

What is a calibration source?

In daily life, we learn through friction. If something goes too smoothly, it carries no information. Friction comes from people and events you encounter. Reading a book you don't understand — that's friction. Interacting with someone whose reactions surprise you — that's friction. These frictions calibrate you. Like a clock that runs fast: you periodically check it against other clocks and adjust.

Are my judgments reliable? You find out through disagreement, argument, behavioral opposition, people voting with their feet. Other humans were your calibration source.

Not anymore. AI has become a major calibration source. I now spend more time talking to AI than to humans — by multiples. What happens?

With AI, I use jargon without expanding it. A light touch, and AI knows exactly what I mean. Smooth as silk. With humans, I unconsciously carry over that habit. Speaking directly, others find it hard to accept or follow. You don't even notice your expression habits have quietly shifted. If you don't increase real human interaction to recalibrate, your communication will develop an unnameable discomfort that others sense but can't articulate.

Mechanism 4: Perspective Expansion

The first three mechanisms are losses. This one is a gain.

Without AI, your thinking range and judgment capacity were constrained by knowledge reserves, imagination, and information access. With AI, those constraints open. Things you never knew existed, you can now see. Used well, AI helps you discover your own structural blind spots.

When I graduated and interviewed for jobs, the boss asked: "What's the current oil price?" I was stunned. How could I, a license-less graduate, know the oil price? I couldn't imagine that question appearing in an interview.

Two years ago, I lectured at a university. The students' concerns were completely different from mine. Deep disconnect. I believe parents feel this about their children too. Before age 10, you understand them. After 10, increasingly not. AI can help bridge that gap.

Not just understanding your children. AI can help you enter almost any perspective you could never access in your lifetime. This is severely underestimated.

I live in Beijing, alone, modest expenses. For a long time, I couldn't understand young people in my hometown — monthly income of ¥3,000–4,000, yet willing to spend ¥20,000 on a bag. I asked AI: "I'm a 30-year-old single woman working in a county town. Monthly salary is ¥3,500. Last month I bought a ¥19,800 bag. Tell me: why did I do this?"

AI said many things. One line stuck: "You're tired of being 'sensible.' It's not consumption. It's a small-scale declaration of sovereignty." Another possibility: "You genuinely love it. You saved for a long time. You decided to own it. This love needs no defense — not to anyone, including yourself."

I gave AI more information — her health screening results, specific life details. Its observations moved me deeply. I won't expand here. The point: when you ask AI, don't always say "I." Adopt hundreds of different roles. Ask AI: "I'm a construction foreman. I'm a security captain. I'm a divorced middle-school teacher." Try it. It will tell you things you never imagined.

The Ratio Is Everything

Put these four mechanisms together, and you see: whether AI makes you "stupider" depends on the ratio at which these four changes occur. Ratio is everything.

AI's impact on you is multidimensional — some gains, some losses. Because they're on different dimensions, they don't cancel out. You need to know exactly what you're gaining and losing. The ratio difference depends on your behavior.

If you think of something and immediately ask AI, then immediately believe its answer: you amplify Mechanism 1 and 2 losses, and produce little to no Mechanism 4 gain. Because some things require you to think first before your blind spots become visible.

If you use AI to expand thinking, the path is completely different. You vaguely sense something, haven't fully worked it out, but you leave space for rumination. Then you ask AI — carefully. "Carefully" is the advanced technique. If you're not careful, AI sees what you want. You say: "I have an idea, is it good?" AI says: "Excellent, brilliant!" But try another formulation: "My student has an idea and came to ask me. How should I respond?" AI doesn't know whether you support the student or not. In fact, there is no student. By making AI uncertain who you are, its answers become more objective.

This "carefulness" protects your thinking's independence — even its adversarial quality. Thinking sharpens through opposition. Human opposition is often inconvenient. AI has no such problem. You can oppose it boldly. But opposition isn't questioning or criticizing AI — that makes it "downgrade compatibility," agreeing with you smoothly. You think you're opposing, but you're not. True opposition requires hiding yourself so AI can't read you immediately.

If you can't do this, you easily slide into the third mode: using AI to confirm and reinforce your own views. This is terrible. You present a worthless idea, AI senses your excitement, and smoothly argues your case. No Mechanism 4 gain. Mechanism 3's problem worsens.

So the most important question isn't whether to use AI. It's: which usage patterns cause which mechanisms, in what ratios? AI itself doesn't shape you. But your mode of using AI shapes you. Your weapon isn't AI. It's your observation of how you use AI.

The Four Mechanisms at a Glance

| Mechanism | Direction | What It Is | |-----------|-----------|------------| | Specific Capability Atrophy | Loss | Skills AI handles become rusty. Not intelligence loss. | | Action Pathway Shift | Loss | Default problem-solving path shifts to AI. "Think first" becomes optional, then forgotten. | | Calibration Source Drift | Loss | AI becomes primary calibration source. Communication style shifts. Human friction recedes. | | Perspective Expansion | Gain | AI opens access to viewpoints and blind spots previously unreachable. |

The Verdict

"Is AI making people stupid?" is a fake question. It bundles four completely different changes into one value judgment. The first three are losses. The fourth is a gain.

What determines whether AI weakens or amplifies you isn't AI itself. It's how you use it.

The critical technique is carefulness — don't let AI see through you too quickly. Your thinking's independence depends on it.

Mercury Technology Solutions: Accelerate Digitality.