Current progress and boundaries
Automated experiment platforms and AI-assisted scientific discovery are active research areas, but work is concentrated on specific tasks and a complete loop still requires substantial expert constraints.
Why it remains difficult
Experimental noise, instrument failures, and theoretical limits are difficult to infer from text. Failed experiments can waste resources, and scientific conclusions must be independently reproduced.
Paths worth exploring
Begin with standardized, low-risk, repeatable experiments. Include safety rules, lab logs, negative results, and independent replication in system design.
How to evaluate success
Assess whether hypotheses are falsifiable, experimental success and replication rates, resource efficiency, and the ability to interpret unexpected results, with independent peer review.