Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations
arXiv cs.AI · 11 Nov 2026, 04:58 AEST
Contrastive Discovery introduces a method for scientific discovery that utilizes competing explanations rather than evaluating hypotheses pointwise. This approach allows agents to resolve more scientific questions using the same evidence and validation budget.
Key facts
- LLM agents are increasingly used for open-ended, data-driven scientific discovery, yet most of them evaluate hypotheses pointwise, testing an empirical claim $h$ only against its complement $ eg h$.
- To recover that alternative, we introduce Contrastive Discovery, which replaces the single hypothesis with a contrast set, a scientific question paired with mutually exclusive and collectively exhaustive competing explanations.
- Across six domains, $89\%$ of the pointwise candidates with the largest belief shifts are rejections.
- When contrast sets guide search, the agent resolves $67\%$ more questions than pointwise surprisal under the same experiment budget.
- Together, these results support explicit competing explanations as a way for discovery agents to resolve more scientific questions from the same evidence and validation budget.
Read aloud
Contrastive Discovery introduces a method for scientific discovery that utilizes competing explanations rather than evaluating hypotheses pointwise. This approach allows agents to resolve more scientific questions using the same evidence and validation budget. L L M agents are increasingly used for open-ended, data-driven scientific discovery, yet most of them evaluate hypotheses pointwise, testing an empirical claim $h$ only against its complement $ eg h$. To recover that alternative, we introduce Contrastive Discovery, which replaces the single hypothesis with a contrast set, a scientific question paired with mutually exclusive and collectively exhaustive competing explanations. Across six domains, eighty nine dollars\%$ of the pointwise candidates with the largest belief shifts are rejections. When contrast sets guide search, the agent resolves sixty seven dollars\%$ more questions than pointwise surprisal under the same experiment budget. Together, these results support explicit competing explanations as a way for discovery agents to resolve more scientific questions from the same evidence and validation budget.
Sources in this cluster
- Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations
arXiv cs.AI · 11 Nov 2026, 04:58 AEST