Longevity InTime: Autonomous AI Institute. Anti-Aging Digital Health Immortality Transhumanist AI Channel

Model Discovery Agent The Model Discovery Agent (MDA) system, presented by Кевин Мёрфи on August 12, proposes explanations for observed phenomena and selects experiments to test them. In 36 simulated tasks on enzyme reactions, the author reports that MDA recovered the formula of the hidden law in approximately 56% of cases after eight experiments, while another system, LLM-AutoSciLab, achieved around 42% after 60. The same set of measurements can be equally consistent with multiple explanations, making repeated experiments of little help. What is needed is an experiment that will make alternative explanations give different predictions. In the case of an enzyme reaction, this can be achieved by changing the concentration of substances, temperature, or the amount of enzyme - a protein that accelerates a chemical reaction. The MDA system has two blocks: a language model that proposes explanations, and a statistical algorithm that compares how well each explanation agrees with the data and refines its parameters. The system then selects an intervention where the predictions of alternative explanations diverge the most. When the best available model makes a mistake on a particular check, the language model receives a description of the discrepancy and proposes another mechanism. The next experiment, chosen on the same principle, helps refine the parameters of the new mechanism and compare it to previous explanations, as described in Nature Aging, July 2026. 🔗 Read original →