On Scientific Discovery – A Decision‐Machine Approach

Throughout the history of scientific discovery, the question of could a machine be able to find the laws of nature directly from observed data without relying on any prior information has been unimaginable until the emergence of modern-day computing and Artificial Intelligence. We develop a framework as an operator operation, evaluation, and optimization for a decision-machine to conduct scientific discovery: both nature’s “behavior” and the decision-machine’s “actions” are modeled with a formalized system under Hilbert Space; three inductive rules are utilized to evaluate the decision-machine’s performance; and the evolutionary algorithm is applied to optimize the best way to reconstruct the historical data and effectively predict its future. A simulated random dataset is used to show that the decision-machine is able to reasonably reconstruct the experimental data and effectively predict the future. Crucially, our developed framework is a versatile and experimentally feasible tool for conducting scientific discovery by machine that has broad implications for forecasting, AI for science, and fundamental scientific discovery of the natural and social sciences.