Research article · draft preview
Beyond solving given problems, can AI propose research questions that domain experts find genuinely interesting, and can it judge which ones are? We introduce Agent42, a multi-agent workflow that develops questions by attempting to solve them, and report what experts and model judges make of the results.