BoxingGym: Benchmarking Progress in Automated Experimental Design and Model Discovery
What this work does
Lets a scientist agent query a hidden probabilistic world, revise a model and communicate an explanation to a novice whose predictions evaluate that explanation.
Connection to discovery
A useful decomposition of evidence acquisition, model quality and reusable communication, rather than scoring only a final answer.
Evidence
The body separates information-gain/regret measures, predictive error and communication-based model discovery across generative domains.
Scope and limitations
Information gain is defined relative to the supplied generative model and parameterization. Communication scores also depend on the receiving agent; these are simulated rather than prospective human or laboratory measurements.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.