[tiberius_p]

First order logic can only detect formal logic fallacies. Informal logic fallacies like ad hominem, strawman, red herring, etc. are cast in language. They can't me defined and resolved mathematically. The model should be fine tuned with examples of these informal fallacies and counter-arguments to them. Even so it won't be able to detect them in all cases, but it will at least have some knowledge about them and how to reply to them. This knowledge could be further be refined with in context learning and other prompt engineering strategies.

[jfengel]

I would expect a true logical fallacy detector to take any natural text and spit out "unsupported assumption, unsupported assumption" over and over and over.

[grandempire]

> ad hominem, strawman, red herring

These aren’t logically incorrect, people who study rhetoric have just identified these as common patterns of poor persuasion.

[Quarondeau]

Couldn't they be classified as non-sequiturs, given that the conclusion doesn't follow from the premises?

[grandempire]

Take ad hominem. It’s true that there is no logical connection between who is saying something and whether it’s true.

But in practice, that’s one of the most relevant factors of whether you should be listening to someone. Does this person have a solid track record? Do they have your interest in mind?

So it is relevant information. It’s just that, “well once this guy kicked a dog” is usually done in bad faith.

So I wouldn’t consider it a non-sequitor, except in its most crude forms.

[taeric]

In this vein, one of the more insipid traps of these fallacies is that they do not lead to a conclusion, on their own.

Ad hominem continues to be a good example. If you know that someone is a liar, you don't know that everything they say is false. You just know that they lie and are likely saying something to affect listeners. Could be based on some truth. Could not.