A New Framework for the AI Age
The Risk We’re Not Talking About
The greatest risk of AI is not that machines become intelligent. It is that humans stop developing their own intelligence.
TL;DR — The Argument in Full
The greatest risk of AI is not that machines become too intelligent. It is that humans stop developing their own intelligence. We have built entire institutions to prevent AI from harming us. We have built almost nothing to prevent AI from diminishing us. This essay introduces a new framework — Human Capability Safety — and argues that every breakthrough in AI capability must be matched by a deliberate breakthrough in human development.
—
Educators across America are reporting cognitive surrender: students who simply stop thinking and defer entirely to AI.
—
When AI is removed, learning gains disappear or reverse. Students are performing with AI, not becoming more capable because of it.
—
The world invests billions in AI Safety. Human Capability Safety has no funding, no research agenda, and no coordinating body.
—
The new operating principle: AI Amplification + Human Amplification = Human Flourishing. Both must scale together.
—
AI that performs for people is not the same as AI that develops people. We have built the former. We have barely begun thinking about the latter.
The Blind Spot
We are investing billions to make AI safe.
We have forgotten to keep humans capable.
A new report from educators across America describes something alarming: writing is collapsing. Critical thinking is weakening. Students who once wrote 20-page papers now struggle to produce 750 coherent words. Researchers have a name for what’s happening — cognitive surrender — the moment a person stops thinking altogether and simply defers to the machine.
This is not a story about lazy students. It is a story about a civilization making a quiet, catastrophic choice — and not realizing it.
We have built almost nothing for
Human Capability Safety.
AI Safety asks: how do we prevent AI from harming humans? It is a vital question. But there is a second question — equally urgent, almost entirely ignored: how do we prevent AI from diminishing humans?
These are not the same question. And we cannot afford to conflate them.
The Evidence Is Already Here
Cognitive decline was already underway.
AI arrived at the worst possible moment.
85%
of high school and college students admit using AI for coursework
8×
more novel ideas in human-written college essays vs. AI-generated ones
11
developed countries — including the U.S. — saw literacy scores significantly decline in the last decade
Stanford reviewed over 800 academic papers on AI in K-12 education. The finding: when AI is removed, learning gains disappear — or reverse. Students are performing with AI. They are not becoming more capable because of it.
This is the crucial distinction. A tool that performs for you is not the same as a tool that develops you. We have spent decades building the former. We have barely begun thinking about the latter.
Where the Money Goes
The asymmetry is staggering.
AI Safety
Billions in annual investment from governments and foundations
Dedicated research labs at Oxford, MIT, Berkeley, and Anthropic
UN resolutions, national AI strategies, regulatory frameworks
Cover stories. Senate hearings. TED talks.
Human Capability Safety
No agreed definition, no coordinating body
No research agenda, no dedicated funding stream
No policy frameworks addressing human cognitive development in the AI age
Almost no one is even asking the question.
The Trump administration just dismantled the Department of Education and cut nearly $900 million in education research grants — at the exact moment we need more research, not less, into how AI shapes human cognition.
We are navigating a civilizational shift with instruments designed for a different era. And the people best positioned to see this — educators — are reporting that they lack the resources, the training, and the policy support to respond.
A New Framework
AI Amplification requires Human Amplification.
Here is the idea I believe our industry has been missing: every breakthrough in AI capability must be matched by a deliberate breakthrough in human development. Not as an afterthought. Not as a corporate social responsibility initiative. As a design principle.
AI Amplification
More capable systems
Human Amplification
More capable people
Human Flourishing
The actual goal
The problem is not that AI is too powerful. The problem is that we have treated it as a replacement for human effort rather than a stimulus for human growth. The calculator did not stop humans from learning mathematics — but only because we decided it wouldn’t. We made that choice deliberately. We haven’t made that choice with AI.
One teacher in the survey put it simply: “AI is a tool that needs to be age-gated. In the same way we required students to learn long division before giving them a calculator, we need students to develop cognitive strength before giving them cognitive shortcuts.”
That instinct is correct. But it is insufficient. We cannot just restrict AI. We have to actively build the human capacities that AI might otherwise erode.
What Human Capability Safety Means in Practice
I
Every AI tool deployed in learning must be evaluated not only for what it does, but for what it builds — or erodes — in the human using it.
II
AI that performs for people is fundamentally different from AI that develops people. Both have a place. We must stop confusing them.
III
Cognitive struggle is not a problem to be solved. It is the mechanism through which humans develop. Removing struggle removes development.
IV
Investment in human capability must scale with investment in AI capability. This is not charity. It is systems thinking.
V
The students who cannot read a 20-page article today will be the voters who cannot read a bill tomorrow. Human Capability Safety is democratic infrastructure.
The Question the Industry Keeps Avoiding
We have asked: How do we make AI more intelligent?
We have asked: How do we make AI safer?
We have barely asked:
“What kind of humans
will we become
because of AI?”
That question is not philosophical. It is operational. It determines what we build, how we deploy it, what we measure, and what we refuse to optimize away.
The answer, right now, is not encouraging. And the window to change it is narrowing.
Questions & Answers
The questions this framework raises.
What exactly is Human Capability Safety?
Human Capability Safety is a framework for ensuring that the rise of AI does not come at the cost of human cognitive development. AI Safety asks: how do we prevent AI from harming humans? Human Capability Safety asks: how do we prevent AI from diminishing humans? These are not the same question. The first focuses on what AI does to us. The second focuses on what AI does to our capacity to think, learn, create, and judge. A civilization that produces powerful AI while allowing human capability to atrophy has not made itself safer — it has made itself more fragile.
How is this different from existing concerns about AI and education?
Most existing concern focuses on specific harms: cheating, misinformation, bias, job displacement. Human Capability Safety is a broader and more structural concern. It asks not whether AI produces bad outputs, but whether it produces less capable humans. A student who uses AI to write a flawless essay has not been harmed in the conventional sense — but if they have also stopped developing as a writer and a thinker, something important has been lost. The framework shifts the evaluation from the quality of the AI output to the development of the human who used it.
What does “AI Amplification requires Human Amplification” mean in practice?
It means that every organization deploying AI — in schools, in workplaces, in products — should ask a second question alongside “does this work?” The second question is: “does this develop the human using it, or does it replace the development?” In practice, this might mean: designing AI tutors that ask questions rather than give answers; requiring students to demonstrate mastery before accessing AI assistance; measuring not just task completion but capability growth; and investing in human development research at the same scale as AI capability research.
Isn’t this just a conservative argument against new technology?
No. This is not an argument against AI. The calculator did not destroy mathematical thinking — but only because society decided it wouldn’t. We made deliberate choices about when and how calculators entered education. The argument here is not to restrict AI, but to be as deliberate about human development as we are about AI capability. A hammer does not make a carpenter. A powerful AI does not make a more capable human. The tool and the person who uses it must both be developed — and right now, we are developing only the tool.
What would it look like if we took Human Capability Safety seriously?
It would look like a field. Dedicated research into how AI shapes human cognition over time. Funding streams. Policy frameworks that evaluate AI deployments not just for safety and efficacy, but for human development impact. It would look like product principles: every AI tool in a learning environment evaluated for what it builds in the human, not just what it produces. It would look like a cultural shift — from optimizing for performance to investing in becoming. We are not there yet. But the window to make that choice deliberately is still open.
The future belongs
to humans who
kept becoming.
AI will keep advancing. That is not the question.
The question is whether the humans directing it
are developing as fast as the systems they’ve built.
Right now, they are not.
That is the risk worth talking about.
