AI is already choosing experiments
Autonomous laboratories combine machine learning, robotics and feedback loops to select experiments, measure outcomes and update the next search step with limited human intervention.
The materials example
Deep-learning systems have predicted large numbers of candidate crystal structures, while robotic laboratories can synthesise and test selected materials. This compresses search time but does not make every prediction experimentally useful.
Novelty must be audited
High-profile materials claims have prompted debate over how many predicted compounds were truly new, stable or practically synthesizable. Databases, literature checks and physical experiments remain essential.
What changes if the loop closes
The strongest case for AI discovery is not a chatbot proposing an idea. It is a system that generates a hypothesis, designs a discriminating test, gathers new data and produces a result that independent researchers can reproduce.
Research record
Nature (2023) — autonomous laboratory for inorganic materialsPeer-reviewed · open sourceNature (2023) — GNoME and materials discoveryPeer-reviewed · open sourceNature — debate over novelty in AI materials claimsScientific news · open sourceCan AI Make Genuine Scientific Discoveries?
Direct answer: Yes, AI can contribute to genuine discoveries and in some systems can autonomously drive parts of the experimental loop. It does not remove the need for validation or scientific accountability.
Discovery partner, increasingly autonomous tool—not an oracle.