CAMPUS NOTICE: IIT Delhi and FSU Jena study reveals AI models excel at basic tasks but struggle with scientific reasoning. Research published in Nature Computational Science. Team developed MaCBench to evaluate AI in chemistry and materials science. Models showed 77% accuracy in equipment identification but only 46% in safety hazard assessment. Findings suggest AI relies more on pattern matching than scientific understanding. Human oversight crucial for complex reasoning and safety decisions. Study led by NM Anoop Krishnan (IIT Delhi) and Kevin Maik Jablonka (FSU Jena). Implications extend beyond chemistry to other scientific fields. Researchers call for advances in AI training to emphasize genuine understanding. (Updated 14 Oct 2025, 15:29 IST; source: link)
Key Points
- CAMPUS NOTICE: IIT Delhi and FSU Jena study reveals AI models excel at basic tasks but struggle with scientific reasoning
- Research published in Nature Computational Science
- Team developed MaCBench to evaluate AI in chemistry and materials science