Pharmacophore an International Research Journal
Artificial intelligence in drug discovery is moving from bounded prediction toward systems that can formulate plans, select computational and laboratory tools, revise strategies, and initiate experiments. This transition changes the scientific governance problem because an agent may influence not only what is predicted but also which evidence is gathered, which hypotheses are pursued, and which physical actions are undertaken. Conventional evaluation based on predictive accuracy, task completion, or benchmark performance cannot determine whether such agency is scientifically justified, appropriately bounded, reproducible, or accountable. This article develops a proposed scientific constitution for agentic drug discovery. The constitution is conceived as a structured allocation of scientific authority, evidence requirements, permissions, prohibitions, challenge rights, escalation conditions, and human intervention powers. It distinguishes scientific roles from technical capabilities, tool availability from tool suitability, model confidence from calibrated uncertainty, molecular plausibility from experimental confirmation, and successful task execution from pharmaceutical usefulness. The proposed construct connects four governance layers: role-bound authority, evidence warrants, rules for planning and action, and accountability through logging, challenge, override, and explicit reauthorization. It further argues that autonomous experimentation should be governed through consequence-sensitive permissions rather than a binary distinction between autonomous and non-autonomous systems. The contribution is theoretical and has not been empirically validated, standardized, or demonstrated to ensure safe deployment. Its applicability will depend on the task, laboratory environment, evidence maturity, chemical hazards, institutional responsibilities, and effectiveness of human oversight. The scientific constitution is therefore offered as an organizing framework for designing and evaluating bounded agency, not as a regulatory instrument or authorization for operational autonomy.
