The Intelligence Test

ROBIN GREEN  |  June 3, 2026

Every major technological revolution eventually becomes a mirror. We build a marvelous piece of technology and then spend a generation staring into it, discovering not what it’s capable of but what we are.

The printing press didn’t merely spread information. It exposed what happened when authority over knowledge was decentralized: reformation, revolution, and eventually the Enlightenment.

The industrial revolution didn’t just increase output. It forced a reckoning with what labor meant, who it belonged to, and what human beings were worth beyond their productivity. 

Today, artificial intelligence is doing the same thing, at a speed and scale those earlier revolutions could never have imagined. The central question AI forces us to answer is the old and familiar: Can intelligence be governed by conscience?

I’ve spent several years researching and writing about this question. I first started paying attention to AI in the early 1980s, when expert systems were going to change everything, which they did, just not everything anyone expected. The answer I’ve arrived at now is a cautious, stubborn yes. But only if we choose it.

Predictably, the choosing is harder than the building.

What the Research Is Actually Telling Us

Start with something counterintuitive. In January 2026, a team led by Professor Karim Jerbi of the Université de Montréal published the largest-ever direct comparison of human and AI creativity in Scientific Reports, drawing on results from more than 100,000 participants. Their findings weren’t what either the optimists or the pessimists expected. Generative AI can now match or exceed the average human on certain divergent thinking tasks. But the most creative individuals consistently outperform even the strongest AI models, and by a widening margin.

As machines absorb routine cognitive work, the human capacities for imagination and ethical judgment become more central to what the economy — and human culture — actually need.

Jerbi’s conclusion is worth sitting with:

Generative AI has above all become an extremely powerful tool in the service of human creativity. It will not replace creators, but profoundly transform how they imagine, explore, and create.

This vital piece is sometimes missed in current AI debates. The question of whether machines can replicate human performance is becoming less interesting as the answer becomes clearer: On a growing list of tasks, they can. 

The more consequential question is what happens at the frontier where they cannot, the territory that requires not just pattern recognition but judgment, not just generation but meaning. That territory turns out to be expanding, not shrinking.

The World Economic Forum’s Future of Jobs Report 2025, compiled from more than 1,000 employers across 55 economies, found that creative thinking has risen to the top of the skills employers demand and that automation is the reason why. As machines absorb routine cognitive work, the human capacities for imagination and ethical judgment become more central to what the economy — and human culture — actually need. 

A great example is Ferrari, which uses generative AI on Amazon Web Services and now runs design simulations 60 percent faster than before. Yet, the engineers didn’t disappear. They gained time for the judgment calls no simulation can make: the tactile sense of whether a car feels right, the experience to know which number to trust.

The paradox, in other words, is real. AI makes human creativity more valuable precisely by doing everything that doesn’t require it.

The Part Nobody Wants to Talk About

The creativity story is genuinely encouraging. The governance story isn’t.

The Organisation for Economic Co-operation and Development’s Digital Education Outlook 2026 documented what it called the “crutch effect.” Students using AI tools performed significantly better on tasks in the moment, but when the AI was removed, their performance dropped sharply. It’s a precise metaphor for a risk running much wider than education. When we outsource cognition without developing the judgment to oversee it, we don’t become more intelligent. We become more dependent.

And the decisions being made right now about what AI systems optimize for, whose data trains them, and who has recourse when they go wrong aren’t primarily technical decisions. They’re moral ones, and we’re making them at speed.

History offers a useful anchor here. In February 1975, roughly 140 molecular biologists gathered at Asilomar, Calif., and agreed to a voluntary moratorium on certain recombinant DNA experiments until safety protocols could be established. The moratorium held. What came out of it shaped bioethics for the next half-century. It wasn’t legally mandated. It was a community of scientists who decided, together and voluntarily, that the first duty of creation was not to see how far it could go but to remember why it began.

The AI field hasn’t produced its Asilomar moment yet. Voluntary commitments exist: Major labs have published safety frameworks, and the first binding international AI treaty was signed in 2024. These aren’t nothing. But they operate in an environment where competitive pressure and commercial incentive push hard in the other direction. The gap between the sophistication of the technology and the maturity of its governance is, at present, considerable.

The Mirror, Honestly

Let’s come back to our reflection. When we look into the AI mirror, we see both things at once.

The COMPAS algorithm, used in U.S. criminal sentencing, exhibited racial bias in its recidivism predictions. This wasn’t a malfunction. The system learned from historical data that encoded decades of human prejudice and reflected that prejudice back with the added authority of a confidence score. That’s the mirror working exactly as designed.

But the mirror shows the other side too. A 2025 study from Harvard’s physics department, led by Gregory Kestin and Kelly Miller, found that students using a carefully designed AI tutor learned more than twice as much in less time as students in traditional active-learning classrooms. The crucial word is “carefully.” The AI that worked was built around pedagogical principles: designed to guide, not to deliver answers, to prompt thinking rather than replace it. The technology didn’t determine the outcome. The intention behind it did.

The machine doesn’t choose which version of this we build. That part is still ours.

The Question Underneath the Question

Here is what I think fellow Mensans will find interesting: The AI debate, for all its technical complexity, keeps collapsing back into a very old philosophical question about the relationship between intelligence and purpose.

A 2025 Gallup study with nearly 4,500 working adults found that employees with a strong sense of purpose are 5.6 times more likely to be engaged in their work than those without it. Not marginally more likely — 5.6 times more. 

MIT Sloan researchers Roberto Rigobon and Isabella Loaiza, writing in 2025 on which tasks AI is least likely to automate, found the answer clustered around ethics, empathy, and what they called “hope” — the human capacity to imagine futures that don’t yet exist. Not because machines can’t approximate these things, but because the approximation requires the real thing to evaluate it.

When machines can do everything that doesn’t require genuine moral imagination, genuine moral imagination becomes the rarest and most valuable thing we have. That’s not consolation. That’s an observation about what kind of investment actually makes sense right now: in education, in leadership, in whatever we decide to call wisdom.

The Duet

The framework that feels truest to me is not a contract but a covenant. A contract specifies what each party delivers. A covenant commits both parties to the flourishing of the relationship itself.

In August 2024, MIT composer Tod Machover premiered his FLOW Symphony at Seoul Arts Center with the Sejong Soloists: a work for string orchestra and a custom AI system that responds in real time to the live players, adding hybrid sonorities drawn from Machover’s own recordings of a Vermont river. Neither the AI nor the musicians control the music unilaterally. Each shapes what the other does. What results is something neither could produce alone.

That’s the model I keep coming back to. Not the machine as servant, not as master, but as a genuine collaborator with its own perspective and its own limitations, and the human as the one who decides what it all means.

I’ve been watching AI long enough to know the hype cycle better than I trust it. Expert systems were going to replace doctors. Neural networks were going to replace judgment. Now, large language models are going to replace thinking. None of it has worked out quite that way because none of it accounted for what turns out to be irreducible about human cognition: not the processing speed, not the pattern recognition, but the caring about what happens next.

The path ahead will test us. There will be choices made for profit that should be made for people, and voices insisting that the only measure of progress is speed. They’ll be wrong, and they’ll be loud.

But there will also be teachers who stay late because a student is finally beginning to understand. Engineers who pause a deployment because something doesn’t feel right. Leaders who choose people over efficiency and discover the two aren’t opposites.

That’s always been the story. AI just makes it more urgent.