Open-source AI program can answer science questions better than humans
Associate Professor Kan Min-Yen from the Department of Computer Science at NUS Computing was quoted in a Science Magazine, feature examining a new open-source AI system designed to answer complex scientific questions with a high degree of accuracy.
Associate Professor Kan commented on both the promise and risks of using such tools in scientific work, cautioning that researchers must carefully judge how much to trust AI-generated answers, particularly when they are used as substitutes for primary sources.
"If you're using these tools to [substitute] for the primary sources, that can be dangerous because there could be nuances that are lost.," he said. While such tools may be more acceptable in fast-moving fields like AI — where they can help researchers keep pace with rapidly expanding literature — Associate Professor Kan noted that greater caution is required in domains such as psychiatry, where patients' health is at stake.
The article discussed how the open-source system was evaluated against both human experts and leading proprietary AI models, and how it performed strongly on tasks requiring cross-paper reasoning and literature synthesis. Associate Prof Kan noted that while these systems show promise in supporting scientific research, they should be viewed as tools that augment — rather than replace — human expertise and critical judgment.