[allturtles]

> Suppose you grant that justification is probabilistic. What's the extra condition that makes justified belief count as knowledge? Pre-Gettier, one might have been inclined to say that it's that your justified belief is actually right—what you believe is true.

I would say that what makes it count as knowledge is the quality of the justification. If Jones had seen a written job offer to Smith rather than just hearsay that he would get the job, his belief that Smith would get the job would probably count as 'knowledge.'

Requiring that what you believe be actually true to count as knowledge is begging the question, since that requires either already having 'knowledge' that it's true, or having direct access to objective reality that no one has.

[natt941]

If "quality" means something like strength of the evidence, then couldn't you still have Gettier-like cases, so long as the evidence doesn't entail the truth of the belief? (suppose the HR department printed the wrong name on the offer letter, etc.)

If "quality" doesn't mean this... then the worry is that you've just labeled what you were trying to explain, because "quality of justification" might mean something like "that thing, whatever it is, which makes a belief count as knowledge".

About truth as a condition for knowledge... there's a lot of subtlety here. The claim isn't that you have to first know that P is true in order for you to know that P, or that you have to provide the truth of P as one of the reasons why you know P, or anything like that. It's more of a logical claim about what it means when we say "S knows that P"... knowledge is "factive" in that you can't know P if P is false. At least, that's how philosophers use the word "know", and it's at least one of the common ways that ordinary folk use the word.

(For what it's worth, there's a lot of discussion amongst philosophers about whether anyone who knows that P also has to know that they know that P... plenty of philosophers deny this.)

[allturtles]

> If "quality" means something like strength of the evidence, then couldn't you still have Gettier-like cases, so long as the evidence doesn't entail the truth of the belief? (suppose the HR department printed the wrong name on the offer letter, etc.)

It would entail Gettier-like cases if you require knowledge to be true in some absolute sense. But I'm discarding the "true" condition as a requirement of a knowledge. Outside of a few domains like math, knowledge of this kind doesn't exist. Knowledge is always subject to revision.

[natt941]

Often we might think we know something and later conclude that we were wrong... but in this case, it's common to say "I thought I knew such-and-such, but I didn't". Also, for what it's worth, most people don't think that knowing that P requires that you're certain that P.

At least, that's one way of using the word "know". So if we're trying to give an analysis of this thing, we can't just decide to discard truth as a requirement of knowledge... the goal is to give an analysis of this thing that people talk about, not replace it with something else.

That said, it sounds like you're sympathetic to a line of thought on which this so-called "traditional" conception of knowledge isn't very useful, and we should focus on the sorts of things that Bayesian epistemology focuses on... something like degrees of confidence.

[allturtles]

> Often we might think we know something and later conclude that we were wrong... but in this case, it's common to say "I thought I knew such-and-such, but I didn't". ... At least, that's one way of using the word "know". So if we're trying to give an analysis of this thing, we can't just decide to discard truth as a requirement of knowledge... the goal is to give an analysis of this thing that people talk about, not replace it with something else.

Let's consider a case like that. A student receives a graded history test, and says "I thought I knew that James Madison was the 3rd president of the United States, but I didn't."

Under the "justified true belief" one might say that their knowledge was undermined because they learned it wasn' true. But their teacher, born (say) circa 1970, has no more direct access to the truth of who was the 3rd U.S. President than their students.

So we might say instead that the student changed their mind about their knowledge because a rather flimsy justification for believing that Madison was the 3rd president ("I studied really hard") was kicked out from under them by a somewhat sturdier justification for a contradictory belief ("My teacher says that Jefferson was the 3rd President, and she knows a lot about history.")

> That said, it sounds like you're sympathetic to a line of thought on which this so-called "traditional" conception of knowledge isn't very useful, and we should focus on the sorts of things that Bayesian epistemology focuses on... something like degrees of confidence.

Yes, I'd say so.

[natt941]

> Under the "justified true belief" one might say that their knowledge was undermined because they learned it wasn't true. But their teacher, born (say) circa 1970, has no more direct access to the truth of who was the 3rd U.S. President than their students.

I'm not quite sure what you mean and I think a bit more precision is helpful here. If one adopts a JTB conception of knowledge, then one wouldn't say that the knowledge of undermined because the student didn't have knowledge. Yes, it's also true that the (weak) justification for believing Madison was the 3rd president was outweighed by the (new and stronger) evidence that he wasn't... but that's about the justification, not the knowledge per se.

(Quick note: philosophers tend to use "undermined" for when something removes the evidential force, rather than countering the force. For instance, suppose I see someone who looks exactly like you in the library. Ordinarily that's pretty good evidence that you were in the library, and it would be reasonable for me to believe you were in the library. However, if I found out that you had an identical twin, this would undermine the evidence, and assuming this is all the evidence I had, it would not be rational for me to believe you were in the library.)

For what it's worth, the post-Gettier literature (and failure to get much clarify about the concept of knowledge) is one of the things that pushed people towards the more Bayesian approach to epistemology.

(Personally I think the Bayesian approach is avoiding some hard but genuine issues about rationality. Specifically, in Bayesian epistemology everything is relative to your priors. But there are lots of probabilistically coherent priors. Are they all equally good? If so, then there's really not much we can say in general about what's good or bad evidence, what people should or shouldn't believe, it's all pretty relative. I don't think that's right... it seems like one can be irrational yet coherent, otherwise it's hard to say how people like you and me are more rational than people believing massive conspiracy theories. Lots more could be said here, but that's why I and plenty of other folks aren't happy with an "anything goes" view about priors. But then for folks like me who want to say that all coherent priors aren't equally rational, it's hard to say what would make some priors better than other in purely Bayesian terms... so I think we end up back in the realm of traditional epistemology. That said, all this is pretty controversial and this is a pretty quick sketch of an argument, so there's of course lots to be said in response.)

[allturtles]

Thanks for the very thoughtful reply.

You're right, JTB would say they didn't have knowledge. But why? Is it because it wasn't justified or because it wasn't true? If it's the latter, how can you ever ascertain whether something is or isn't knowledge (i.e. how do you know 'truth')? That's my fundamental problem with this account.

I agree, also, that just defining good knowledge as beliefs based on good priors regresses the problem. I tend to think this regression is unsolvable except in pragmatic terms (what is efficacious when tried against the real world is what is good).