.jpg)
Using ChatGPT 4.0, senior researcher David Balcells asked the model to co-design a computational chemistry project aimed at exploring large combinatorial spaces in chemical catalysis — a problem relevant to drug discovery. Balcells found the model capable of proposing a coherent research plan, though it sometimes made unstated assumptions and omitted important details like relativistic effects for certain metals.
He also tested the AI to critically evaluate its own work and compared responses from other models such as DeepSeek and Gemini, which sometimes disagreed among them, potentially signaling points that would require further verification. Balcells concludes that co-intelligent design is feasible with large language models and that emerging research-tailored models may advance this approach further. While he acknowledges challenges — including copyright issues, skill erosion among experts, and potential misuse — he argues that AI can increase the speed, scale, and quality of research if used critically and correctly.
Read the article (in Norwegian): Kan-KI-bli-kjemiker-forsker-testet-om-ki-kan-desig.html