DeepMind, founded by Demis Hassabis, aims to “solve intelligence” and then use that intelligence to “solve everything else.” Sam Altman and Dario Amodei also believe that AI has the potential to greatly benefit society, with Altman predicting that AI will drive faster scientific progress and Amodei suggesting that AI progress could lead to a “country of geniuses in a data center” by 2026. One of the foundational myths driving the AI boom is the hope that AI can help humanity better understand the universe.
FrontierScience, a new benchmark published by OpenAI, indicates that AI models are making progress towards this goal of advancing scientific capabilities. The benchmark includes questions in physics, chemistry, and biology at two levels of difficulty, with Olympiad-level questions pushing the limits of what young minds can do and Research tier questions testing open-ended reasoning and the ability to support real-world research. Some of the questions are so complex that they could take days or even weeks for human scientists to answer.
For example, one research question in chemistry asked about “meso-nitrogen atoms in nickel(II) phthalocyanine,” while another in physics required a derivation of “electrostatic wave modes” in plasma. These questions are challenging even for expert researchers, with one PhD researcher in plasma physics noting that similar analyses have taken him weeks to complete. The benchmark aims to rigorously measure how AI models can improve scientific capabilities and potentially accelerate scientific discovery.
The difficulty in testing AI models as they become more competitive with human scientists underscores the rapid progress being made in the field of AI. The hope is that AI can eventually help humanity unlock new insights into the universe and drive scientific progress at an unprecedented pace. As AI continues to advance, researchers and developers will need to grapple with the ethical implications of this technology and ensure that it is used for the betterment of society as a whole.
