Here, I partly take back what I said in an earlier article and argue that the Coherence Theory of Justification is not really refuted by probability theory.* (Warning: somewhat technical.)
Maybe I'm totally misunderstanding but isn't all this stuff with probability kinda downstream of the cohetentism/foundationalism debate. I mean I never thought the cohetentist was committing to any particular claim about how the probability distribution ultimately applies. After all, probability theory is just a fancy way of counting up outcomes of various kinds.
Rather, I would have thought the debate would occur at some sense at the level of what justifies one in having a particular prior in the first place.
Or to put the point differently, I'm not seeing why it's at the level of individual outcomes that one is supposed to apply coherentism rather than at the background theory level itself. I mean, presumably the cohetentist would have beliefs about the nature of witnesses, their truthfulness etc and it's that view to which coherentism applies but why would coherentism not be able to support any theory at all about how to combine witness testimony since, presumably, it's their belief in some broader generalities about witnesses which justifies this not the direct application of coherentism as if one was starting from a position of no knowledge whatsoever about how witnesses work.
(or is thinking about these as witnesses causing confusion here since we can't help but bring prior theory to the table).
I would have thought that these sorts of issues in formal epistemology (and connections with statistics/ML) would be core parts of the undergraduate philosophy curriculum. Instead I have BA and MA in philosophy and never learned about these things.
What on earth is going on? Is it just that some philosophers don't know math so they don't want to put it centrally in the curriculum?
I'm interested in this stuff especially due to it's relevance to normative ethics.
On the one hand, I think utilitarianism has some extremely unintuitive implications. On the other hand, I think that it and its implications are more intuitive, perhaps even by leaps and bounds, than those of any other theory that has been specified to a comparable level of precision. So I am torn on how to resolve that.
I found this book on Bayesian Philosophy of Science which I'd like to read at some point.
The consideration that pushes me most strongly toward utilitarianism is that (1) many of its counterintuitive implications seem to have to do with large numbers (e.g. "a trillion people stubbing their toes is worse than one baby being tortured to death") and (2) most people are *provably* terrible at reasoning with large numbers in domains other than ethics where there's a provably correct answer.
Some of this is a result of difficulty reasoning about large numbers. Huemer makes this point in his defense of the Repugnant Conclusion. Some of the time maximizing pleasure/wellbeing conflicts with other considerations that seem very real.
Infinity, large numbers, and probability can be used to make really counter-intuitive results. Oftentimes, people modify their utilitarianism in order to avoid certain outcomes, especially in the case of population ethics.
A difficult scenario is the 51%-49% deal on world-doubling. Seems like you should take it till you cease existing. It converges on 0 outcome while one narrow low probability event explodes toward infinity. Hard for many to take the deal. But basic expected utility maximization calls for that. Here is a comment I made saying that I found a counter to this unpersuasive: (https://sarahconstantin.substack.com/p/why-infinite-coin-flipping-is-bad/comment/10943226). To me, it looks like inconsistency.
I think that utilitarianism is very appealing for some because the whole view can be reduced to a single sentence, but this also results in the inuintuitive results. The important question is: How much should be value this simplicity? I'm not sure. But physical reality is rather complicated. I think some pre-socratic philosophers had a similar impulse to say "everything is fire" "everything is water." There were wrong but maybe utilitarians are right. Seems more plausible.
Yeah I agree with that mostly. But it's not just that utilitarianism is simpler than other views; it's also that it's more fully spelled out. It's not as though other views require 10 volumes to spell out precisely whereas utilitarianism only requires one sentence; rather, other views are just not fully spelled out all.
I suppose that's only fair since ideas like deservingness or loyalty seem kind of impossible to define precisely in terms of "simple" concepts. I agree with Quine that it is basically impossible to define any concept precisely (you will go in circles), so maybe we shouldn't expect to do so in normative ethics. But I still do think there's something unsatisfying about the ultra-hardcore moral particularlism that many philosophers seem to have adopted.
Another issue is about intuitions for particular cases vs. intuitions about principles.. Are these the same kind of intuition or different? Seems like utilitarianism has intuitive principles but not intuitive particular cases. An alternate theory like "prioritize those people you are close to" might be less intuitive as a principle but more intuitive in actual cases.
But really, my larger point here is that these things should be discussed more by philosopher and students.
I don't want to argue that coherentists are correct, but I do have some concerns about what lessons we can draw from your argument.
In your post "Compassionate Phenomenal Conservativism", you defined phenomenal conservatism as
> PC: If it seems to you that P, and you have no grounds for doubting that appearance, then that gives you at least some degree of justification for believing P.
Now the lack of "grounds for doubting" is playing a role here. You don't spell out what those grounds would be, but I would assume that a justified belief Q that's contrary to P would count as a ground for doubt that P.
But *is* Q a justified belief? Presumably it was before you got this seeming that P. But now? Your belief in Q is fundamentally based on the seeming(s) that Q, which you previously had no (or insufficient) grounds to doubt. But this seeming that P is now a grounds for doubting Q. You may no longer be justified in believing Q.
How do you determine whether you are still justified in believing Q or whether you should jettison it in favour of P?
Well, you're going to examine how your seemings all fit together. If P fits a lot better in the totality of your seemings than Q does, you should make the change. If Q fits a lot better, you should stay put. If it's a toss-up, you should retreat to believing (P v Q) -- or perhaps something weaker.
In other words, you're basing your judgement on how coherent each alternative seems to be. Anything obviously incoherent (such as P & Q) is rejected outright.
It is the (seeming) superior coherence of one set over the other that justifies your belief in the elements of that set rather than those of the other.
Well it's still a seeming, and the moderately strong coherence of a set including P might be grounds for doubting that Q, so it still fits with PC.
But of course if you're not really very good at determining what's coherent and what's not, then your seemings of coherence are not really a good guide to what is and is not coherent. If someone claims that {P, Q} just seems like a more coherent set than either {P} or {Q} on its own, does that really make them justified in holding both P and Q (which, recall, are contraries)?
In fact, someone truly awful at determining what's coherent might (justifiably???) believe (P & ~P).
It seems to me that, in order to be justified in choosing P over Q (or vice versa), the set of beliefs you come up with including P (Q) must *actually* be coherent, and substantially more coherent than the alternatives (so that there are insufficient grounds for doubt).
My conclusion is that if P is a justified belief, then it does cohere with the other beliefs you hold (and with more -- it coheres with the existence of seemings to the contrary, for example).
Furthermore...
In your post "Logical Properties of Warrant", you explained that warrant is not unique. Any property that all knowledge necessarily has (beyond being a true belief) is a warrant property.
I want to talk about justification properties -- the properties that a belief must have in order to be justified. That is, the properties J that satisfy JBsp = Bsp & Jsp.
First off, CBsp (p is coherent with all of s's current beliefs) is not actually a candidate for Jsp. For one thing, P doesn't cohere with what s believes when s believes Q, even when the seeming that P should (rationally) force s to replace Q with P. In order to make a rational decision, s must consider alternate sets of (possible) beliefs. If coherence is a part of J, then it must involve some kind of summation over sets of belief that s does not currently have. CBsp -/-> Jsp.
Nevertheless, it looks like Jsp -> CBsp (*), for the reasons given above.
I'll also note that Ssp & CBsp -/-> Jsp (where Ssp is that it seems to s that p). s might have grounds for doubt that are not full-fledged beliefs. For example, p might be coherent with everything s currently believes, but not with everything s currently suspects.
Next, Ssp is not equal to Jsp. s might have grounds to doubt that p, in which case Jsp might be false even tho' Ssp is true. Ssp -/-> Jsp.
I don't think Jsp -> Ssp, either. It's not that I see a flaw in your arguments (I don't). But it may be possible for Jsp to be true when Ssp isn't -- p might, for example, be a simply derived but unobserved consequence of what s believes.
Of course if Jsp -/-> Ssp, then ~Ssp -/-> ~Jsp. And I think that's problematic for your claim that seemings are "the source" of justification. Sure, ~Ssp -/-> ~JBsp, but that's (I think) because JBsp -> Bsp -> Ssp. Seemings seem to be more the source of beliefs than of justifications for those beliefs. The requirement that s lacks grounds for doubt in your definition of PC shows that coherence is at least as important as seemings. If the above is correct, then it is more important to justification than seemings are.
I hope that the above is worth the effort of an attempt to refute it!
(*) A possible counter-example: Am I justified in thinking that at least one of my beliefs is unjustified? If so, is that an example of a set of a belief that is not coherent with my current beliefs?
Maybe I'm totally misunderstanding but isn't all this stuff with probability kinda downstream of the cohetentism/foundationalism debate. I mean I never thought the cohetentist was committing to any particular claim about how the probability distribution ultimately applies. After all, probability theory is just a fancy way of counting up outcomes of various kinds.
Rather, I would have thought the debate would occur at some sense at the level of what justifies one in having a particular prior in the first place.
Or to put the point differently, I'm not seeing why it's at the level of individual outcomes that one is supposed to apply coherentism rather than at the background theory level itself. I mean, presumably the cohetentist would have beliefs about the nature of witnesses, their truthfulness etc and it's that view to which coherentism applies but why would coherentism not be able to support any theory at all about how to combine witness testimony since, presumably, it's their belief in some broader generalities about witnesses which justifies this not the direct application of coherentism as if one was starting from a position of no knowledge whatsoever about how witnesses work.
(or is thinking about these as witnesses causing confusion here since we can't help but bring prior theory to the table).
I would have thought that these sorts of issues in formal epistemology (and connections with statistics/ML) would be core parts of the undergraduate philosophy curriculum. Instead I have BA and MA in philosophy and never learned about these things.
What on earth is going on? Is it just that some philosophers don't know math so they don't want to put it centrally in the curriculum?
I think that anybody who is reasoning about knowledge needs to understand probability to some extent.
I'm interested in this stuff especially due to it's relevance to normative ethics.
On the one hand, I think utilitarianism has some extremely unintuitive implications. On the other hand, I think that it and its implications are more intuitive, perhaps even by leaps and bounds, than those of any other theory that has been specified to a comparable level of precision. So I am torn on how to resolve that.
I found this book on Bayesian Philosophy of Science which I'd like to read at some point.
https://academic.oup.com/book/36527
The consideration that pushes me most strongly toward utilitarianism is that (1) many of its counterintuitive implications seem to have to do with large numbers (e.g. "a trillion people stubbing their toes is worse than one baby being tortured to death") and (2) most people are *provably* terrible at reasoning with large numbers in domains other than ethics where there's a provably correct answer.
Some of this is a result of difficulty reasoning about large numbers. Huemer makes this point in his defense of the Repugnant Conclusion. Some of the time maximizing pleasure/wellbeing conflicts with other considerations that seem very real.
Infinity, large numbers, and probability can be used to make really counter-intuitive results. Oftentimes, people modify their utilitarianism in order to avoid certain outcomes, especially in the case of population ethics.
A difficult scenario is the 51%-49% deal on world-doubling. Seems like you should take it till you cease existing. It converges on 0 outcome while one narrow low probability event explodes toward infinity. Hard for many to take the deal. But basic expected utility maximization calls for that. Here is a comment I made saying that I found a counter to this unpersuasive: (https://sarahconstantin.substack.com/p/why-infinite-coin-flipping-is-bad/comment/10943226). To me, it looks like inconsistency.
I think that utilitarianism is very appealing for some because the whole view can be reduced to a single sentence, but this also results in the inuintuitive results. The important question is: How much should be value this simplicity? I'm not sure. But physical reality is rather complicated. I think some pre-socratic philosophers had a similar impulse to say "everything is fire" "everything is water." There were wrong but maybe utilitarians are right. Seems more plausible.
Yeah I agree with that mostly. But it's not just that utilitarianism is simpler than other views; it's also that it's more fully spelled out. It's not as though other views require 10 volumes to spell out precisely whereas utilitarianism only requires one sentence; rather, other views are just not fully spelled out all.
I suppose that's only fair since ideas like deservingness or loyalty seem kind of impossible to define precisely in terms of "simple" concepts. I agree with Quine that it is basically impossible to define any concept precisely (you will go in circles), so maybe we shouldn't expect to do so in normative ethics. But I still do think there's something unsatisfying about the ultra-hardcore moral particularlism that many philosophers seem to have adopted.
Another issue is about intuitions for particular cases vs. intuitions about principles.. Are these the same kind of intuition or different? Seems like utilitarianism has intuitive principles but not intuitive particular cases. An alternate theory like "prioritize those people you are close to" might be less intuitive as a principle but more intuitive in actual cases.
But really, my larger point here is that these things should be discussed more by philosopher and students.
Very interesting and well-argued post.
I don't want to argue that coherentists are correct, but I do have some concerns about what lessons we can draw from your argument.
In your post "Compassionate Phenomenal Conservativism", you defined phenomenal conservatism as
> PC: If it seems to you that P, and you have no grounds for doubting that appearance, then that gives you at least some degree of justification for believing P.
Now the lack of "grounds for doubting" is playing a role here. You don't spell out what those grounds would be, but I would assume that a justified belief Q that's contrary to P would count as a ground for doubt that P.
But *is* Q a justified belief? Presumably it was before you got this seeming that P. But now? Your belief in Q is fundamentally based on the seeming(s) that Q, which you previously had no (or insufficient) grounds to doubt. But this seeming that P is now a grounds for doubting Q. You may no longer be justified in believing Q.
How do you determine whether you are still justified in believing Q or whether you should jettison it in favour of P?
Well, you're going to examine how your seemings all fit together. If P fits a lot better in the totality of your seemings than Q does, you should make the change. If Q fits a lot better, you should stay put. If it's a toss-up, you should retreat to believing (P v Q) -- or perhaps something weaker.
In other words, you're basing your judgement on how coherent each alternative seems to be. Anything obviously incoherent (such as P & Q) is rejected outright.
It is the (seeming) superior coherence of one set over the other that justifies your belief in the elements of that set rather than those of the other.
Well it's still a seeming, and the moderately strong coherence of a set including P might be grounds for doubting that Q, so it still fits with PC.
But of course if you're not really very good at determining what's coherent and what's not, then your seemings of coherence are not really a good guide to what is and is not coherent. If someone claims that {P, Q} just seems like a more coherent set than either {P} or {Q} on its own, does that really make them justified in holding both P and Q (which, recall, are contraries)?
In fact, someone truly awful at determining what's coherent might (justifiably???) believe (P & ~P).
It seems to me that, in order to be justified in choosing P over Q (or vice versa), the set of beliefs you come up with including P (Q) must *actually* be coherent, and substantially more coherent than the alternatives (so that there are insufficient grounds for doubt).
My conclusion is that if P is a justified belief, then it does cohere with the other beliefs you hold (and with more -- it coheres with the existence of seemings to the contrary, for example).
Furthermore...
In your post "Logical Properties of Warrant", you explained that warrant is not unique. Any property that all knowledge necessarily has (beyond being a true belief) is a warrant property.
I want to talk about justification properties -- the properties that a belief must have in order to be justified. That is, the properties J that satisfy JBsp = Bsp & Jsp.
First off, CBsp (p is coherent with all of s's current beliefs) is not actually a candidate for Jsp. For one thing, P doesn't cohere with what s believes when s believes Q, even when the seeming that P should (rationally) force s to replace Q with P. In order to make a rational decision, s must consider alternate sets of (possible) beliefs. If coherence is a part of J, then it must involve some kind of summation over sets of belief that s does not currently have. CBsp -/-> Jsp.
Nevertheless, it looks like Jsp -> CBsp (*), for the reasons given above.
I'll also note that Ssp & CBsp -/-> Jsp (where Ssp is that it seems to s that p). s might have grounds for doubt that are not full-fledged beliefs. For example, p might be coherent with everything s currently believes, but not with everything s currently suspects.
Next, Ssp is not equal to Jsp. s might have grounds to doubt that p, in which case Jsp might be false even tho' Ssp is true. Ssp -/-> Jsp.
I don't think Jsp -> Ssp, either. It's not that I see a flaw in your arguments (I don't). But it may be possible for Jsp to be true when Ssp isn't -- p might, for example, be a simply derived but unobserved consequence of what s believes.
Of course if Jsp -/-> Ssp, then ~Ssp -/-> ~Jsp. And I think that's problematic for your claim that seemings are "the source" of justification. Sure, ~Ssp -/-> ~JBsp, but that's (I think) because JBsp -> Bsp -> Ssp. Seemings seem to be more the source of beliefs than of justifications for those beliefs. The requirement that s lacks grounds for doubt in your definition of PC shows that coherence is at least as important as seemings. If the above is correct, then it is more important to justification than seemings are.
I hope that the above is worth the effort of an attempt to refute it!
(*) A possible counter-example: Am I justified in thinking that at least one of my beliefs is unjustified? If so, is that an example of a set of a belief that is not coherent with my current beliefs?