Islands of humanity
In 1810, Wilhelm von Humboldt helped found the University of Berlin on a simple idea: research and teaching belong together. Professors teach what they are discovering, and students learn by doing research themselves. For two centuries, this unity has defined what a university is. I think it may be coming to an end, because a large part of research itself can now be automated.
Agents are getting good at long-horizon research. On my old gaming PC, an agent runs MCTDHF simulations of atoms in CodyFortran. It implements new propagators, tests them, compares them and comes back with a report. That is the kind of work that used to fill a large part of a PhD.
This puts universities in a difficult spot. There is a real case for PhD students doing the research themselves, including the parts that can be automated today. Watching a method fail and figuring out why is the friction that builds the inner cathedral. But a place that deliberately focuses on the human side of research may struggle to also stay at the frontier. Maybe Humboldt’s unity of research and teaching has done its work.
So where does AI-driven research belong? I expect the rise of a new kind of institute: institutes of cognitive symbiosis. In 1960, J.C.R. Licklider described Man-Computer Symbiosis with the image of a fig tree and the one wasp that pollinates it. Neither can live without the other.
A research agent can run convergence checks all day, but it makes false assumptions, for example about which parts of the equations are stiff and force tiny time steps. And it rarely steps outside its own frame. So it comes back and asks. That is cognitive symbiosis: the human asks the AI for help where the AI is better, and the AI asks the human for help where the human is better. The AI brings speed, stamina and endless careful testing. Experienced humans bring intuition, the quick reset of a wrong idea and a feel for what results mean in the real world.
Universities would then focus on the human side of knowledge: digesting it, discussing it, organizing it and, most of all, teaching other humans. They would still do research, but the human side of research: driven by human questions, human incentives and human agency. Not the cutting edge, but islands of humanity. And they would produce what the labs cannot make themselves: people whose judgment was built by friction.
The labs will push the frontier. The universities will make sure humans can still follow.