From Artificial Intelligence to Autonomous Discovery: The Next Architecture of Scientific Power

This strategic concept paper argues that scientific advantage will increasingly depend on the capacity to convert AI-generated proposals into independently verified, reusable knowledge. Drawing on developments at Google, Microsoft, IBM, Quantinuum, and academic laboratories, it proposes an integrated architecture for autonomous discovery: an experiment allocator guided by information value, an independent replication network, and a portfolio of validated discovery assets. A fourth, exploratory proposal examines experiments designed to expose failures in scientific models. Each proposal specifies its contribution, evaluation approach, and conditions for failure. Quantum computing is treated as a conditional component whose value must survive a complete comparison with strong classical alternatives. A twelve-month pilot offers a practical path to testing the architecture. The paper distinguishes reported capabilities from future hypotheses and explicitly acknowledges the substantial contribution of AI to its development. It is a strategic working paper, not an experimentally validated system or a claim of established technical priority.