Probability and Assertion V. H. Dudman


Download 32.07 Kb.
Date conversion05.02.2017
Size32.07 Kb.
Probability and Assertion


It is a conjecture popular with philosophers today that just those things are assertible which seem highly probable, or that things are assertible to just the degree they seem probable. I shall urge that there is no connection between ‘assertibility’ and high subjective probability, whether for atomic statements or for ‘conditionals’.


Might assertion go by probability for atomic statements?1 A test suggests itself at once: contrive a probability just short of certainty and see what happens to assertibility. – In a big lottery, say, where conditions are ideal, probability and assertion both exactly observable.

The result is a knock-down blow to the probabilist programme: I can withhold assent even when the subjective probability is enormous. Full knowing that the chance of my ticket’s winning is extremely small, I nevertheless vigorously dissent from ‘My ticket won’t win’. I would not have bought it unless I had thought it might win, and I still maintain that it may. I do not expect it to win, and agree that it probably won’t, but I disagree that it won’t win simpliciter. Someone has to win: why not me? And similarly when apprised that the lottery has already been drawn. Conceding that the chance of my ticket’s having won is extremely small, and agreeing that it probably didn’t win, I decline to assent to either ‘My ticket didn’t win’ or ‘My ticket won’t have won’, preferring ‘My ticket may/might have won’. It is not just that the probability is never high enough to trigger assertion. An exacter appreciation is that even the smallest uncertainty is enough to cohibit it. Assertibility goes out of the window as soon as the underlying thought is reduced to relying on ‘mere’ probability. Since mere probability leaves open the contrary possibility, assenting on the basis of admitted mere probability would be to treat admitted possibilities to the contrary as if they were impossibilities. This is apparently why we steadfastly withhold assent in the face of rocketing odds. And if so then what is needed for assertibility in this sort of case is not high probability but rather absence of possibility to the contrary.

However that may be, the negative result stands: if there is a ‘rule of assertion’, it cannot be to assert what is highly probable.2 This perception can be obscured by deep questions about probability or by the premature intrusion of explanatory terms like ‘belief’ and ‘knowledge’. It needs remembering that the proposed relationship is directly between probability and assertability, exactly the items the lottery lays open to us.

A simpler argument to the same disconnection is persuasive but inconclusive. As a matter of observation, a speaker who affirms that Booth almost certainly shot Lincoln can still refuse to say that Booth shot Lincoln simpliciter. My explanation is that ‘Booth shot Lincoln’ commits its speaker to the simple factual claim, a commitment that the speaker who interpolates ‘almost certainly’ is at pains to avoid. Presumably the probabilist account is that the probability that Booth shot Lincoln can be simultaneously high enough for ‘Booth almost certainly shot Lincoln’ and not high enough for ‘Booth shot Lincoln’. I invite the reader to decide which explanation is more plausible.

In the same vein, the following exemplifies a stance people actually adopt:

Booth almost certainly shot Lincoln. Indeed the probability that he did not is so small that for practical purposes we may ignore it. We can take it then, that Booth shot Lincoln.
As I understand it, this passage tacitly acknowledges that almost-certainty is strictly insufficient for assertion. On the understanding that the very probable is assertible, it is hard to see what the passage is straining at. – Still, this consideration, too, seems evitable. For the discomfiture of the probabilist programme, I pin my faith in lotteries. To my simple mind, lotteries prove it is not probability that governs assertion.

Probability plays a large role in present-day philosophical thought about ‘conditionals’, perennially fascinating topic. I shall consider Ernest Adams’s suggestion that, for them, assertibility goes by conditional probability. (See [1]. by definition, the conditional probability of C, given A equals the probability of C
& A divided by the probability of A.) ‘Adams’s hypothesis’ is that an indicative conditional ‘If A, C’ is assertible just when the conditional probability of C given A is high. (Thus [2] p. 174; cf. [5] pp. 11. 138.)

Actually, the ‘indicative conditional’ is a bogus category which conflates two very different breeds of message. No matter: I come armed with two counterexamples, one for each breed.

On the positive side, I shall air an alternative proposal: that people’s reactions to ‘conditionals’ are triggered by rational, intelligible, commonsense reasoning whose exact conceptual connections are there to be documented and dissected, because the reasoning is recoverable by patent catechism. If we follow this line, it soon emerges that each breed of indicative conditional has its own style of underlying reasoning, quite different from the other’s (section III).

The two breeds are represented by the natural, single-situation interpretations m1 and m2 of (1) and (2), and I shall consider them in turn:

(1) If Booth did not shoot Lincoln someone else did

(2) If Booth does not shoot Lincoln someone else will.

The argument
(A) Someone shot Lincoln; so if Booth did not shoot Lincoln someone else did
is received on all sides as intuitively cogent: good downright commonsense. But why? Why is (1) assertible given that someone shot Lincoln? Because to a thinker certain that someone shot Lincoln (says Adams) it is exactly as probable that someone other than Booth shot Lincoln as that Booth did not shoot Lincoln. At bottom (say I) because ‘Someone shot Lincoln; Booth did not shoot Lincoln; therefore someone else shot Lincoln’ is a valid argument. (For the full story see Part 4 of [4].)

A decisive test suggests itself: make the conditional problem very high but stay this side of validity. The following experiment was designed by Anthony Appiah ([2], p. 172). Take a plurality of urns containing little coloured balls, among them urn B containing many, many white balls and a single black one. Choose a ball z blindfold from an unknown urn, possibly B, and with nothing more to go on, consider the natural interpretation m3 of

(3) If z
is from urn B, z is white.

m3 belongs, of course, to our first breed of ‘indicatives’. But as a matter of observation, do we assent to it? Is m3 assertible?

Not according to my sources, who are unanimous and vociferous that anyone affirming m3 would have to be ignoring the black ball. They concede that if z is from B it is ‘very probably’ indeed ‘almost certainly’ white, but insist that m3 remains strictly untenable as long as there is a black ball in B, even if the number of accompanying white balls is redoubled.

This accords exactly with my proposal, of course: (3) is not assertible because it is not entailed that z is white, and what nixes the entailment is the one black ball in B. But it is bad news for Adams, who would have (3) assertible because of the high conditional probability.3 – And so to the other breed of indicative conditional.

Over lunch, informant X has become persuaded

(a) that a crack team of professional assassins is now poised to shoot Lincoln tonight, and with every prospect of success;

(b) that Booth, acting alone, is now posed to shoot Lincoln tonight, and with every prospect of success;

(c) that some unknown assailant, acting alone, is now poised to shoot Lincoln tonight, and with every prospect of success.
In each cased, so convincing has been the spiel that X now accepts that someone will shoot Lincoln tonight. She assents, then, to the single-future-event interpretation of ‘Someone will shoot Lincoln’, which we can call m4 for future reference. Now, given that she knows nothing else relevant, will she assent to m2?

My experiments strongly suggest yes, no and no, and I think I can explain all three of these results. When X believes (a) and assents to m2, it is because she expects the machinations of the conspirators to continue unhindered up until evening, imagines Booth then not shooting Lincoln, and, uncaring even whether Booth is on the team, reaches a favourable verdict about someone else shooting Lincoln in those circumstances. When X sat down to lunch she was not expecting anyone to shoot Lincoln. Now that she believes (b) and assents to m4, she still dissents from m2, because the ‘if’-clause imagines thwarted the only machinations against Lincoln she has reason to suspect. If X believes (c), and projects the machinations of an unknown malefactor, it makes a difference whether she also believes that the unknown malefactor definitely isn’t Booth. If she believes this, the ‘if’-clause in (2) achieves nothing: m2 is tantamount to m4 and meets with assent. But given that X knows ‘nothing else relevant’, presumably she is in the other case: for all she knows, Booth could be the would-be assassin. Her reaction? As soon as it becomes a possibility in her mind that the unknown malefactor whose machinations she is projecting might be Booth, X will for that very reason withhold assent from m2, just as she did when she believed (b). I report this as a matter of observation, accessible to the native speaker by self-catechism. But of course: if the unknown malefactor should chance to be Booth then the ‘if’-clause imagines thwarted the only machinations she has reason to suspect.

Adams’s hypothesis predicts the results for (a) and (b), and I pass over them. For an X who believes (c), the conditional probability of m2 depends on how likely it seems to her that the unknown malefactor is Booth. If very, then Lincoln’s being shot by someone else is correspondingly unlikely and m2 is unassertible. But our counterexample lies at the other extreme, where it is genuinely possible but most improbable that Booth is the villain, and the probability of someone else’s shooting Lincoln is correspondingly high. Adams predicts assent in such a case, because of the probabilities. But the fact seems to be that people dissent, as I have averred: ask them to assent to m2 and they insist on appending ‘so long as the unknown malefactor isn’t Booth’. And if this is so, they are evidently impressed by the mere possibility which flaws an otherwise cogent piece of commonsense reasoning.

Those who attempt the same semantics for m1 and m2 are obliged to reckon (B) a valid argument which works just like (A):
(B) Someone will shoot Lincoln; so if Booth does not shoot Lincoln someone else will.
In (A), the proposition that someone shot Lincoln occurs as an independent premiss, given, from which m1 is simply concluded. So a similar construction needs putting on (B). The task is essayed by Frank Jackson, noted exponent of Adams’s hypothesis:
Perhaps someone I know to be very knowledgeable about the plot has assured me that Lincoln will be assassinated by someone acting alone, but she remained silent on who it would be. Booth is just my educated guess.

What do I say beforehand? I affirm (2), that if Booth does not kill Lincoln, someone else will. I expect it to be Booth, but be I right or wrong about that part of the story, what I am confident of is that it will be someone or other, and so, if not Booth, then someone else. ([5], p. 146)

But I am curious about this plot. For its outcome, it has Lincoln dispatched by someone acting alone. Therefore there must be someone whose failure to dispatch Lincoln will result in the failure for the plot. And it is not impossible that this someone should be Booth. Nevertheless Jackson affirms m2. Ye gods; he even expects it to be Booth, and still feels able to ignore the question what if it is. Some plot.

m1, we discern, can never be simply given but has to be got. It is about a future time y, and it has to be got from facts at the point O of speech. And how it is got – from which present facts and by what line of reasoning – determine whether m2 can also be got.4 Everything depends on the story that runs from O to y. Some stories support m2 and m4 in equal degree. Others lend m2 no support at all. No doubt there are further possibilities. But always everything depends (not on how firmly m4 is believed, or on how probable someone’s shooting Lincoln is deemed to be, but) on the detailed workings of the story. Jackson wanted to use m4 without a story. That was the function of the ex cathedra pronouncement from someone ‘very close to the plot’. But the idea sits ill with X’s postprandial deliberations.

I have come this far in order to sketch the sort of examination of commonsense underlying reasoning that I prefer when assertion is in question. But this is far enough. I instance lotteries, Appiah’s urn, and the solitary unknown malefactor as illustrating, in their separate areas, the irrelevance of subjective probability to ‘assertibility’.

[1] Ernest Adams, The Logic of Conditionals (Dordrecht: Reidel, 1975

[2] Anthony Appiah, Assertion and Conditionals (Cambridge: Cambridge University Press, 1985).

[3] Anthony Appiah, ‘”If” Again’, Analysis 47 (1987) 193-9.

[4] V. H. Dudman, ‘Antecedents and Consequents’, Theoria 52 (1986) 168-99.

[5] Frank Jackson, ‘Classifying Conditionals’, Analysis 50 (1990) 134-47.

[6] Allan Gibbard, ‘Two recent Theories of Conditionals’, in Ifs, edited by W.L Harper et al.

(Dordrecht: Reidel, 1981), 153-90.

1 Among ‘atomic statements’ are reckoned not only propositions, such as the single-past-event interpretation m5 of ‘Booth shot Lincoln’, but also certain judgements about the future, including the single-future-event interpretation m6 of ‘Booth will shoot Lincoln’. Now propositions are patently candidates for truth, but according to a popular usage, m6 can never be true but only come true. Such intimations go to the heart of how these messages are understood, and are best respected where possible. Why not teach, conformably, that m6 is assertible just when its coming true seems highly probable? Since m6’s coming true would consist in Booth’s shooting Lincoln, this amounts to saying that m6 is assertible just when Booth’s shooting Lincoln seems highly probable. Meanwhile we can retain the received account of propositions: m5 is assertible when it seems very probably true that Booth shot Lincoln. This revision still has assertibility governed by probability but without attributing either truth or probability to m6 itself. On the other hand, the revised formulation plainly invokes different applications of probability, for the probability of an event of a certain kind’s happening is a quite different conception from the probability of something’s being true, good thing to remember in any case. This distinction is blurred when, tendentiously, the probability of Booth’s shooting Lincoln is processed as the probability that (it is true that) Booth will shoot Lincoln, in line with the maxim (omnes, passim) that, for all atomic statements, ‘probability is probability of truth’. – While tacitly respecting these niceties, my submission in section I are meant to show the untenability of probabilism for propositions and judgements alike, but without adverting to the distinction.

2 A contrary line of persuasion is sometimes encountered. If there is no level of probability less than one at which they would assent to a proposition, asks Anthony Appiah, then
why do Australians have so much to say about ... empirical matters of fact, since quantum theory – which is the best theory we have – predicts that very many future contingencies – the sun’s rising tomorrow, for example – have some finite probability of not occurring? ([3], p. 198)
Verily he bluffeth: from the premisses that the best theory we have assigns a probability less than 1 to the sun’s rising tomorrow and that X asserts ‘The sun will rise tomorrow’, it does not follow that X’s line of thought adverts even remotely to notions like certainty or probability.

A couple of other considerations merit brief mention as fallaciously encouraging probabilism. First, we can always be mistaken about things we say. Whatever my grounds for thinking that Booth shot Lincoln, I could be wrong. But this acknowledgement has no tendency to allow probability into assertion. After all, the fallibility in question is not something we could ever put a figure on. As well , claims about the future are vulnerable in a second way. ‘Booth will shoot Lincoln’, we say, as Booth appears in the presidential box; but before this can happen Booth, or Lincoln drops dead or goes mad or ... , and so Booth does not shoot Lincoln after all. The contingency of events being what it is, this sort of vulnerability is plainly inherent when a judgement about the future is got from present facts. But there is no incentive to probabilism here either, for there can be no question of quantifying sheer proneness to fortuitous upset from whatever quarter.

3 This objection is countered by Appiah. I cannot think that I am fully seized of his position, but from my perspective he seems to defend Adams’s hypothesis by flatly abandoning it.

He claims to have explained people’s resistance to m3 ‘in ways that have nothing much to do with the semantics of “if”-sentences ([3], p. 197), by assimilating it to their resistance to ‘This ticket won’t win’. So he needs to treat the unwelcome result obtained from lotteries as harmless to his theory of assertion because special or exceptional. Instead, it seems to me, he simply revises his old theory of assertion so as to accommodate them as unexceptional and ordinary.

The ‘official’ doctrine of his 1985 book is that ‘you should assert what you believe to be very probably true’ ([3], p. 197). But quite near the end of it he has a footnote in which, with a sang-froid one can only admire, he confides a brand new and wholly different persuasion. ‘I think Slote is right that what matters for assertibility is not belief but knowledge’ ([2], p. 248n.). When we withhold assent from ‘This ticket won’t win’ it is because ‘we have no reason to thing that we know that it is true’ ([3], p. 197). And what matters for assertibility is knowledge. But if assertibility depends on knowledge then it does not depend on probability, for the two are antithetical. To embrace the new doctrine of assertion is ipso facto to forsake the entire program of linking it with probability. A perfect volte face.

This new theory of assertion is explicitly extended to conditionals. The assertibility requirement for ‘If A, C’ is that coming to have minimal evidence warranting the assertion of A would lead you to have grounds for asserting C; ‘reasons, then, for believing that you know it’ ([3], p. 197). The reason that you withhold assent form m3 is that ‘you have no reason to think that evidence that you had urn B would give you knowledge that the ball was not black’. I respond as before. To adopt the new account of assertibility for conditionals is ipso facto to abandon the very terms in which Adams’s hypothesis is cast. No betrayal could be more emphatic. Yet Appiah claims to have argued that, for As and Cs with truth-conditions, his new account has Adams’s hypothesis as a consequence! (p. 197).

4 Given the same set of present facts, different lines of reasoning are possible, leading to opposite reactions to the same ‘future indicative’. Told that Sly Pete can see what is in his opponents hand and is desperate to win, but nothing about what is in either player’s hand, informant X may either assent that if Sly Pete calls he will win or dissent. Both the example and the observation that more than one reaction can be significantly represented are due to Allan Gibbard: see [6], p. 229.


The database is protected by copyright © 2017
send message

    Main page