DISCERN: Can AI Agents Work Like Scientists and Guide Discovery?
What this work does
Builds controlled data-integrity, analysis-verification and hypothesis-revision tasks around real datasets, including constructed conflicting-evidence dossiers.
Connection to discovery
A benchmark for evidence discipline and scientific judgment, complementary to task-completion or equation-recovery metrics.
Evidence
Inspected the three evaluation levels and failure analysis showing that recognizing a limitation need not change a final claim.
Scope and limitations
Controlled injections and review rubrics do not constitute new physical experiments. Model/judge choices and domain composition affect scores; counterfactual dossiers are constructed evaluation cases.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.