Scientists make AI play Battleship to help it do science better

Artificial intelligence (AI) is being tested with board games like Battleship to improve decision-making skills, which could revolutionize scientific research. Scientists, including Valerio Pepe, are exploring how AI can make better choices with limited resources, similar to how researchers decide which experiments to conduct. In a study presented at the International Conference on Learning Representations, AI models played a collaborative version of Battleship, where they had to locate and sink ships efficiently. Initially, humans outperformed AI models like Meta’s Llama-4-Scout, but OpenAI’s GPT-5 showed better results. By optimizing AI to ask strategic questions and communicate using code, Llama-4-Scout eventually surpassed GPT-5 and human players in efficiency. Although Battleship is simpler than real scientific problems, the strategies used in the game could be applied to scientific decision-making, potentially transforming how research is conducted. QUESTION: How might AI’s ability to make better decisions with limited resources impact the future of scientific research and discoveries?