Human involvement in fully automated ML-driven scientific discovery

Determine whether human scientists need to be involved in the identification and discovery or validation of new causal relationships when pursuing fully automated scientific discovery with machine learning.

Background

The review discusses prospects for fully automated science driven by ML, particularly in identifying causal relationships. It flags uncertainty about whether humans must remain in the loop for identification tasks, citing broader concerns about automation in scientific practice.

Clarifying this question would shape research agendas and system designs for AI scientists, determining the extent to which human judgment is indispensable in concept discovery and causal inference.