Abductive Hypothesis Generation

Abductive Hypothesis Generation

Inference to the best explanation in the face of anomalies: when an anomalous phenomenon that existing theory cannot explain is observed, systematically generate candidate explanations and select the most plausible one as the hypothesis.

When to Use

• A clear anomalous phenomenon is observed (a result inconsistent with existing theoretical predictions)
• Existing theory cannot adequately explain a known phenomenon
• One of several competing explanations must be selected as the most worth testing
• The research starting point is "this result is strange, why?"

Not applicable: no clear anomaly, just wanting to explore a new field → use inductive-hypothesis-generation instead.

Thinking Framework

Anomaly → Generate candidate explanations → Rank by plausibility → Best explanation = hypothesis

The core logic of abductive reasoning:

1. Anomaly: precisely describe the anomaly — what phenomenon, inconsistent with what expectation, how large the deviation
2. Generate candidate explanations: systematically generate all candidate explanations that can account for the anomaly (no premature filtering)
3. Rank by plausibility: rank by plausibility — which explanation is most parsimonious, most consistent with known facts, most testable
4. Best explanation = hypothesis: select the most plausible explanation as the working hypothesis, retaining the rest as competing hypotheses

Core principles of abduction:

• Occam's razor: when explanatory power is comparable, prefer the explanation with fewer assumptions
• Consistency: the best explanation should not contradict other known facts
• Testability: the best explanation must be able to produce observable predictions (otherwise it cannot be verified)
• Generation completeness: candidate explanations must be exhausted before ranking, to avoid premature convergence

Budget Gate

Default Reference Flow