Hacker News Reader

Comment by hydrox24 | Hacker News Reader

Is there a good reason to exclude abductive reasoning from an analysis like this? It's even considered by at least one of the referenced papers (Fangzhi 2023a).

Abductive reasoning is common in day-to-day life. It seeks the best explanation for some (often incomplete) observations, and reaches conclusions without certainty. I would have thought it would be important to assess for LLMs.

View on HN

My instinct is it is a distinction without a difference in this context. i.e. if deductive is "I watched the cue ball hit the 8 ball, therefore, the 8 ball is moving" and abductive is "the 8 ball is moving towards me, therefore the cue ball must have hit it. I cannot claim to have deduced this because I did not observe it", LLMs cannot observe the situation, so any deduction (in the binary induction/deductive sense) must be done by abduction.

https://en.wikipedia.org/wiki/Abductive_reasoning

Is abductive inference synonymous with bayesian inference?

Here’s a good page to read on that:

https://plato.stanford.edu/entries/abduction/#AbdVerBayConTh...

I like to think of abductive reasoning as the basis for science that explains natural processes that happened in the past -- like astronomy and geology and evolution -- where experiments are too big to conduct or processes too slow to observe in real-time. So we propose mechanistic explanations for nonobvious outcomes like the formations of stars, or motion of large land mass via plate tectonics or glaciation, or long-range organism speciation over millennia. That's the role for abduction, to explain how all that happened.

No, but agreement with priors is one way one might choose between possibilities.

For example suppose you go outside and the streets are wet. Perhaps it rained, or perhaps someone drove a fire truck around spraying water all over the streets. You might select the former because of its higher prior probability.

Reply on news.ycombinator.com