Key Info

Anthropic's Science Blog published a guest post by Jakob Macke, Daniel Gedon, and Stefan Wahl describing how AI is beginning to assist with scientific modeling, including writing, running, fitting, and revising models as code.

Highlights

  • Scientific models explain and predict real-world systems, from single neuron firing to neutron star collisions.
  • Recent advances have changed how scientists fit complex models to data, but models are still typically written manually.
  • Researchers often must weigh several competing model versions that explain data differently.
  • AI can explore many candidate models by treating them as code, automating parts of the modeling process.