abduction
Inferring an explanation for an observation; reasoning to the best explanation, often requiring inventing a new hypothesis.
As defined in 343 | Tom Griffiths on The Laws of Thought
Where it comes up · 3
- 304 | James Evans on Innovation, Consolidation, and the Science of Science4 claims
- 343 | Tom Griffiths on The Laws of Thought2 claims
- AMA | August 20221 claim
Most important claims · 7 of 7Search all claims for “abduction” →
- Charles Sanders Peirce introduced abduction as a third mode of reasoning beyond Aristotelian deduction (deriving necessary consequences from axioms) and Baconian induction (generalizing from new observations): as collective knowledge grows, the best signal for theoretical development is surprise—an experiment or observation that violates one's increasingly accurate world model—which one then resolves by reasoning to the best explanation.304 | James Evans on Innovation, Consolidation, and the Science of Science
- Abductive discoveries are systematically mergers between insiders and outsiders: the people most able to detect a surprise are the 'priest class' who know everything the field expects, but they are least able to resolve it, so disruptive advance occurs when intellectual resources are imported from the most disconnected fields (literary studies, astrophysics, molecular biology) rather than through induction within one's own field.304 | James Evans on Innovation, Consolidation, and the Science of Science
- Deduction is reasoning from certain premises to certain conclusions, whereas induction is reasoning from things we know that may not fully determine the conclusion, yet still making a reasonable inference on that basis—the kind of reasoning scientists use to discover principles of the world.343 | Tom Griffiths on The Laws of Thought
- For abductive inference—coming up with an explanation for something observed—much of the work happens at the algorithmic level rather than being captured by probability math, because you often must invent a hypothesis no one has thought of before in order to entertain it; this is something brains do rather than something the math tells you how to do.343 | Tom Griffiths on The Laws of Thought
- When comparing China and the US, individual Chinese scientists move further across topics over their careers, yet China as a whole focuses on far fewer topics because people follow trends toward the center; the US has many smaller autonomous pockets where individuals move less, but those pockets become a reservoir of diversity from which collective abduction becomes possible.304 | James Evans on Innovation, Consolidation, and the Science of Science
- To produce abductive creativity, AI cannot be a single monolithic model; because expectations come from literature and data come from experiments separately, and deep models internally form little 'schools of thought' that recombine only at prediction time, creativity requires deliberately separating and accentuating that underlying diversity—using some models to generate deep expectations and others to survey experiments/observations and dynamically identify surprises—since training models on their own generated data causes attention collapse and degradation.304 | James Evans on Innovation, Consolidation, and the Science of Science
- In studying Reddit's 'Change My View' forum, deductive arguments (premises to conclusion) almost never change people's minds and may even backfire, because they only work if both parties share premises; personal anecdotes and experiences are what actually change minds, reflecting that humans reason by abduction (inference to best explanation) rather than deduction from foundational beliefs.AMA | August 2022