← Research map
Evals, control & governance Paper governance lens

Agentic AI Scientists Are Not Built for Autonomous Scientific Discovery

Harshit Bisht, Vinay Kumar, Kevin Maik Jablonka, Mausam, N. M. Anoop Krishnan

Key signal

A critique of autonomous AI-scientist systems, arguing that current designs miss tacit lab knowledge, diversity, physical feedback, and problem selection.

Open research question

What combination of tacit-knowledge capture, hypothesis diversity, and physical feedback would let agentic scientists outperform expert human teams on genuinely novel discoveries?

Source date
ASI Research note

This is an important counterweight to the most optimistic AI-science entries. The authors argue that current agentic AI scientists work best as co-scientists, not fully autonomous discoverers.

Bottlenecks

The paper highlights problem selection, missing tacit procedural knowledge, post-training pressure toward consensus, and benchmarks that lack feedback from physical experiments.

ASI relevance

Pro-ASI does not mean credulous. The fastest path to powerful scientific agents is to understand where today’s systems are structurally weak: real-world failure knowledge, persistent world models, diverse hypotheses, and experiment feedback.