Medicine wasn't invented by someone trying to invent medicine. It emerged from humans being curious about why people got sick, because they lived in bodies that got sick. Engineering came from needing shelter, wanting to cross rivers, watching things fall. The discoveries came from embodied, situated, motivated minds encountering problems that mattered to their survival.
We take for granted that the pipeline of human discovery will continue. That future generations will invent disciplines we can't yet imagine, solve problems we don't yet know exist. But that pipeline depends on something fragile: the cognitive capability of humans to notice, wonder, and reason their way into entirely new territory.
What happens to that pipeline when we hand the cognitive work to machines?
The augmentation narrative
The dominant conversation in 2026 is about knowledge augmentation. AI making us smarter, faster, more productive. And it's not wrong. AlphaFold predicted protein structures nobody had solved. AI is accelerating drug discovery and materials science. These are real breakthroughs.
But they extend from a foundation that humans built. AlphaFold needed decades of painstaking, human-generated protein data to train on. Every AI breakthrough stands on a body of knowledge that was created by people doing the slow, embodied, often unglamorous work of observation, experimentation, and failure.
The question nobody is asking: who generates the next generation of that foundational knowledge if humans have been trained out of doing it?
Beyond knowledge collapse
Important work is already emerging on the risks. Acemoglu's research on knowledge collapse (NBER Working Paper w34910, February 2026) demonstrates how AI dependence can erode existing knowledge within organisations. Klein and Klein write compellingly about the "hollowed mind," the cognitive atrophy that follows when humans stop exercising judgment. The deskilling literature is growing across multiple disciplines. The BCG/Harvard research (Randazzo et al., HBS Working Paper 26-036, December 2025) reveals how different patterns of AI use create fundamentally different cognitive outcomes.
All of this matters. But I believe these researchers are each describing a piece of something bigger.
Knowledge collapse is about losing what we already have. What I'm calling discovery extinction is about losing what we haven't found yet.
If we offshore the cognitive work that builds the muscle for discovery, we don't just lose current expertise. We lose the breeding ground for future expertise that doesn't exist yet. Entire disciplines that would have emerged from human curiosity and embodied experience simply never happen.
And here is the feedback loop that makes it self-reinforcing: the AI systems that depend on human-generated knowledge to train on lose their raw material too. As humans produce less original foundational knowledge, AI has less to learn from. The well doesn't just stop being replenished. It dries up.
The measurement problem
This is where traditional research design hits a wall. You cannot run a controlled experiment on discovery extinction. There is no control group for the future. You cannot measure the discipline that was never founded or the cure that was never stumbled upon. By the time you have trailing evidence that discovery has slowed, the cognitive foundations needed to reverse it may already be gone.
Some risks are so systemic and so slow-moving that by the time they are measurable, they are irreversible. Climate science faced exactly this challenge for decades: the evidence lag meant that by the time sceptics accepted the data, the window for easy intervention had narrowed considerably.
The answer is not to wait for proof. It is to measure the leading indicators.
What I have seen in 30 years
I have spent three decades leading technology transformation in organisations. Long before AI, I watched this pattern play out with earlier waves of automation. Organisations hollowed out capabilities not through malice, but through the quiet assumption that if technology could do it, people didn't need to learn it any more.
And when the need returned, it was usually in the form of a retired contractor who still had the knowledge and capabilities required. The organisation had to buy back, at premium rates, the expertise it had allowed to atrophy.
AI accelerates this pattern by orders of magnitude. Previous automation waves replaced manual tasks. AI replaces cognitive tasks: the reasoning, judgment, and pattern recognition that are the raw materials of discovery.
The critical question
The question in 2026 is not whether your organisation is ready for AI. Every maturity model on the market will help you answer that.
The question is whether AI is leaving your people ready for the future.
I am building CHART (Cognitive Health Assessment for Responsible Technology), a maturity framework to help organisations answer it. The methodology will be openly published. It measures whether AI deployment is preserving or eroding the human cognitive capabilities that all future progress depends on. Not AI readiness. Not AI governance. Not AI ethics. The raw human capacity to think, judge, and discover.