AI researchers test models on Einstein's relativity theory

Researchers are conducting an experiment to determine if artificial intelligence can independently replicate the scientific breakthroughs made by Albert Einstein. The goal is to distinguish between mere memory and true scientific discovery by testing whether an AI can generate original insights rather than just retrieving existing data.

To achieve this, scientists presented an idea at the India AI Summit involving a Large Language Model (LLM) restricted to the knowledge available prior to 1911. The objective was to see if the model could derive the theory of general relativity using only the scientific context of that era.

Initial results indicate that current AI models have failed to achieve this goal. Experts, including researchers from Google DeepMind and the Technion Israel Institute of Technology, suggest that these failures highlight significant limitations in the creative reasoning capabilities of current AI architectures, suggesting a need to rethink the fundamental principles upon which these models are built.