Blog powered by Typepad

« Discussions elsewhere | Main | Nida-Rumelin on grasping phenomenal properties »

December 11, 2006


Joachim Horvath

Since no finite rational subject should accept that it is omniscient or infallible, and since to accept Moore-paradoxical sentences really seems irrational, the natural conclusion would be that no rational subject should use the Ramsey test. Or, more cautiously, rational subjects should regard the test merely as a heuristic device. But was there ever any reason to take it as more than that, i.e. as a one-hundred percent reliable method to settle questions about the acceptability of indicative conditionals?

Joachim Horvath

Some second thoughts: Are you really entitled to draw conclusions about omniscience and infallibility? In (1) and (2) you only talk about 'belief'. So wouldn't it be more precise to say that a rational subject should 'only' regard itself as omnidoxastic and doxastically infallible. Both of these could still be cosmic accidents and therefore would not automatically count as instances of knowledge or justification. So, the rational subject in question may still radically overestimate its cognitive capacities, but at least it wouldn't fall victim to the hubris of equating itself with God.

I guess that a lot turns on what we say about a conclusion like:
'It is rational for me to accept that if p, then I believe that p.'
Is the rational force of my acceptance of the latter conditional strong enough to make it rational for me to believe that I know everything that is the case - or does it 'merely' make it rational to regard myself as omnidoxastic?

And a similar question can be asked for:
'It is rational for me to accept that if I believe that p, then p.'
Does my rational acceptance here really warrant the conclusion that I am an infallible knower - and not 'merely' a doxastically infallible subject?


oh, nice. Seems to show that in an important class of cases belief update doesn't go by conditionalisation: upon accepting that it's currently raining in Berlin, I rationally come to accept that I believe that it's raining in Berlin. But my prior conditional credence in "I believe that it's raining in Berlin" conditional on "it's raining in Berlin" is quite low.


Joachim: Re your first point, I take the moral to be that one should reject the version of the Ramsey test given in the paper. Other versions might still work. Re your second point, you're right that strictly speaking the argument doesn't entitle us to a conclusion about epistemic powers. I think one could make the case that the corresponding beliefs with "know" instead of "believe" are also irrational, though. The analogs of (2) are obviously bad, and the analogs of (1) are arguably bad. Alternatively, one could just change the conclusion to concern "the doxastic powers" (or "the cognitive powers") of a god. Maybe we'll do that.

Wo: You're right that conditionalizing on 'It's raining in Berlin' doesn't work. But arguably this is because one isn't conditionalizing on total new evidence. That is, when one updates on the evidence 'It's raining in Berlin', one thereby gains new (introspective) evidence 'I believe its raining in Berlin', so one also needs to conditionalize on that, and so on. Versions of this point also arise with the Thomason conditionals that we cite.

Joachim Horvath

Dave: The corresponding beliefs with "know" don't seem to be so clearly irrational or problematic to me. Consider a rejection of (1) first: "p and I don't know that p". O.K., in everyday talk this would sound a bit odd, but there certainly is no pragmatic maxim in place that one should only express pieces of knowledge by one's assertions. The utterance will be rational, for example, if I think that I have good evidence for p, yet not enough evidence to constitute knowledge. Now consider (2), the conditional itself: "If I know p, then p". Isn't this just a trivial conceptual truth about knowledge, i.e. that knowledge implies truth?

Kosta Calfas

A thought experiment: remove all instances of "rationally requires" and "rational" and see if the argument is nearly as coherent without the question-begging. OK, That ends the facetious part, now on to the logic. Regarding Joachim Horvath's last comment it would not be at all odd to say "p and I don't know that p" depending on how you define "know." I would imagine there are a number of things that are p and you don't "know that p." Again, "know" is one of those terms, like "rational" that cannot simply be dropped into an argument as-if uncontroversial or settled. A case can be given where "rational requirements" for a belief cannot be assessed prior to holding that belief such that one needn't accept neither premise in David's argument. Imagine a man believes there is an external deity who is responsible for his fortunes, and meanwhile works to ameliorate his disposition, believing "god helps those who helps themselves." It is part of his belief system that the fruit of his efforts coincides with the beneficence of his deity. Now the question is: "Was his belief in p rational?" What might be called the "rational" or "knowledgeable" part of his behaviour (working to ameliorate his suffering) is in a sense parasitic upon holding the belief. Were our believer not to hold his belief in a deity in the first instance, we would not be able to evaluate the rationality of his beliefs in the second. It therefore strikes me as wrongheaded to say one is "rationally required" to accept premise (1) and (2) in David's paper since, in some cases, it is putting the proverbial cart before the horse.


Joachim: Well, plenty of people (e.g. Williamson) think it is a norm that one should only assert what one knows. Some also hold that there are related norms on belief, and that there's something wrong with believing P when one takes oneself not to know that P. I don't know that these views are correct, which is why I said "arguably". In the reverse direction: yes, 'If I know p, then p' is a triviality.

Kosta: Of course the argument doesn't work without "rational", as it turns on Moorean principles about rationality. The paper doesn't endorse either the Ramseyan or Moorean principles -- it just makes an observation about what they entail. Still less does it endorse the conclusion. I'm inclined to think that the Ramseyan principle (as stated) is to blame, but of course there's a lot to be said about the Moorean principles as well.

Joachim Horvath

Dave: Thanks for your patient responses first! But there's still something I don't quite get. Above you say that "analogs of (2) [with 'know'] are obviously bad" and when I say that such an analog, i.e. "If I know p, then p", is trivial, you agree. But since it is trivially true and it doesn't imply anything like infallibility, how can it be irrational to accept it? It seems, quite on the contrary, that it would be irrational to deny it. Am I missing something here?


Sorry, I meant that the knowledge-involving analogs of the Moorean beliefs for (2) are obviously bad. I.e. it's bad to accept "I know p" and to reject or to suspend judgment about p. What follows is that one should rationally accept all instances of the knowledge-involving version of (2) -- which in this case is obvious, though not very interesting.

Joachim Horvath

May I just sum up briefly how I see the issue then. If we try to reason just like in your paper with "know" instead of "believe" we get: (1) If p, then I know p.
This is bad, of course, but - pace Williamson - it is not as implausible that it can be rationally denied, i.e. without holding a Moorean belief, as it is in the case of its analog with "believe".
Secondly, we get: (2) If I know p, then p.
This is not bad at all and therefore need not and should not be denied on rational grounds.
In conclusion, it seems that (1) should and arguably can be rationally denied, while (2) is unproblematic and should be accepted. But then, these are much more palatable results than the ones you derive for "believe" in your paper!

Joachim Horvath

Footnote: But one really does get very bad results in both cases if one uses "believe with justification" instead of "know"!

Miguel Hoeltje

Nice paper! But the conclusions do seem a little strong to me; from "x accepts every *instance* of "p --> I believe that p" he happens to consider", it doesnt follow that x should accept that he is ominsicent. This seems to require that x accepts "Ap (p --> I believe that p)" - which he might (reasonably) deny. (The same seems to go for being infallible.)

It seems there can be schemata which are such that you cannot reasonably deny any of their instances, while you can reasonably deny their universal generalization.

What am I missing?

Ben Benjamin

you are missing an answer from the oracle of Adelaide.


I think the conclusion is a result from confusing a hypothetical world with the real one.

Suppose Nancy is dubious about some proposition p. Let's apply the Ramsey test to see if Nancy believes "if p, then Nancy believes p". First we have to enter the hypothetical world where Nancy believes p. But this isn't really Nancy, its some strange Bizarro Nancy who has come to believe p.

To apply the test, we see if Bizarro Nancy accepts "Nancy believes p". But Bizarro Nancy, if she is at all rational, doesn't accept "Nancy believes p", because Nancy doesn't believe p -- only Bizarro Nancy does! Therefore, by the Ramsey test, Nancy would reject "if p, then Nancy believes p", i.e., Nancy doesn't believe she is godlike afterall.

The comments to this entry are closed.