Topic: Drafting conclusion

Conclusion type: Summary (plus)

You might view a conclusion as an introduction in reverse: a bridge from the world of your paper back to the world of your reader. The simplest conclusion is a summary of the paper, but at this point you should go beyond mere summary. You might begin with a summary, for example, and then extend it with a discussion of the paper's significance or its implications for future study, for choices that individuals might make, for policy, and so on. You could urge readers to change an attitude or modify behaviour. Certainly, you're under no obligation to discuss the broader significance of your work (and a summary, alone, will satisfy the formal requirement that your paper have an ending), but the conclusions of effective papers often reveal that their authors are "thinking large" by placing their limited subject into a larger social, cultural, or historical context.

Question: From the above notes draft a Summary (plus) conclusion from Rothschild and Spectre's manuscriptA Puzzle about Knowing Conditionals below. Reveal the following separately putting subheadings.

• Summary.
• Extend it with a discussion of the paper's significance or its implications for future study.
• Implications forchoices that individuals might make.
• Implications for policy.
• Urging readers to change an attitude or modify behaviour.
• Reveal that the authors is "thinking large" by placing limited subject into a larger social, cultural, or historical context.

N.B : Make sure to include all the key words that are inboldthroughout the manuscript.

The Manuscript: A Puzzle about Knowing Conditionals

Abstract

We present a puzzle about knowledge, probability and conditionals. We show that in certain cases some basic and plausible principles governing our reasoning come into conflict. In particular, we show that there is a simple argument that a person may be in a position to know a conditional the consequent of which has a low probability conditional on its antecedent, contra Adams' Thesis. We suggest that the puzzle motivates a very strong restriction on the inference of a conditional from a disjunction

One thousand fair coins were flipped one by one yesterday. You have no information about how they landed but, in fact, not all the coins landed heads. It is tempting to think:

(Anti-skepticism) You know that not all the coins landed heads. We take the name from a related thesis in Dorr et al. (2014). The name is apt because if we deny it then we would most probably need to discount any of our knowledge that has a probabilistic evidential basis, which results in a wide-ranging skepticism.

Here is another attractive principle:

(Independence) You should treat each of the coin flips as probabilistically independent. Independence is meant to be a constraint on your probabilistic beliefs about the coins: the probability function representing your credences in the coin flips should make each flip probabilistically independent of each other. This hardly needs motivation: after all, they are causally independent by assumption and you have no special information that would break independence. However, it is worth noting, as Bacon (2014) does, that Independence is incompatible with as signing probability 1 to the proposition that one coin will land tails. So much the worse, we think, for the idea that knowledge requires assigning a proposition probability.

Here are some more general principles:

(Restricted Adams' Thesis). Where A and B are non-conditional statements about coin-flips in the setup, you should assign a conditional statement of the form If A then B as its probability the conditional probability of B given A.

This is just an instance of Adams' Thesis (Adams, 1975), which itself puts no restrictions on A and B. Adams' Thesis assumes that conditionals are not material, as the material conditional can often have a different probability from the conditional probability of B given A. Adams' Thesis is widely assumed to accurately character ize our reasoning and talk with natural language conditionals. For example, saying that it's likely that if A then B seems to be just the same as saying that it's likely that A conditional on B. This observation is explained by the Restricted Adams' Thesis. The main sources of trouble for the unrestricted version of Adams' Thesis stem from Lewis's triviality results (1976) and a certain class of cases where the thesis seems unintuitive (e.g., Kaufmann, 2004). We think these issues are orthogonal to those we are discussing here, and in particular do not apply when A and B are restricted to being statements about coin flips in our setup.

(Restricted or-to-if). If you know a statement of the form A or B but you do not know that A is true or that B is true, then you are in a position to know that if not Athen B.

This is a famous and much discussed inference pattern (e.g., Stalnaker, 1975). Note that Restricted or-to-if is only a substantive hypothesis if the conditional is not the material conditional (as Adams' Thesis implies), since otherwise the disjunction and conditional are logically equivalent. An unrestricted version of the or-to-if principle is more problematic: Suppose you know that it is raining, then you can (perhaps) infer that either it's raining or there's a Martian invasion. In this case, you can use the unrestricted or-to-if principle to infer that if it's not raining then there's a Martian invasion. The restricted version of the or-to-if principle, however, is extremely attractive. It explains many cases of conditional knowledge from inferences. For example, I know Cathy is either in Hong Kong or Sao Paolo, so I know that if she's not in Hong Kong, she's in San Paolo.

(Knowledge & Probability). If you are in a position to know something then you cannot assign it a probability of one-half or less.

This is an uncontroversially weak link between one's probabilities and one's knowledge (much weaker than the doctrine that you can only know things you assign probability 1 to). These principles are in tension. Here is the argument:6 There must be a least number n such that you know that the first n coins did not all land heads. This follows immediately from the setup and Anti-Skepticism (as well as principles of classical logic, which we will consider later). Assuming knowledge to be closed under (known) logical equivalence you know on the current setup the disjunction: either the first n-1 flips did not all land heads or the nth flip landed tails. Since you do not know either disjunct in this case (the first you don't know by the choice of n, the second by the setup and Knowledge & Probability) then by Restricted or-to-if you know that if the first n-1 flips all landed heads then the nth flip landed tails. However, by Independence and Restricted Adams' Thesis you assign this conditional a probability .5. So by Knowledge & Probability you do not know this conditional. Contradiction.

Something has to give. The only plausible candidates seem to be: Anti-skepticism, Restricted Adams' Thesis, Restricted or-to-if and perhaps the background classical logic that we used to derive the contradiction. A few thoughts on these: Restricted Adams' thesis might seem the softest target as the unrestricted thesis is independently problematic and known to have apparent counterexamples. Nonetheless the restricted version of Adam's thesis does not obviously on its own lead to any paradoxical results and we could further restrict it to just the one instance used in the previous paragraph. This use of Adam's thesis does not look anything like the standard apparent counterexamples. Indeed, it seems intuitive to us that the probability of the conditional if the first n-1 flips all landed heads then the nth flip land landed tails is just as Adams thesis states.

A natural reaction to Anti-skepticism is to think that the coin proposition, the first 1000 flips did not all land heads, looks like a lottery proposition, e.g., this ticket will lose the New York State lottery. Many epistemologists think lottery propositions are not knowable (in absence of direct evidence) so we might think that the coin propositions are also not knowable and reject Anti-skepticism. However, it would be a mistake to think that theoretical consistency requires us to take the same attitude toward coin propositions as to lottery propositions. There are many non-lottery propositions that we think have a small probability of being false that we nonetheless want to say we know. For example, reading in the local newspaper that you lost the local lottery with 1/1000 chances of winning, would seem to give you knowledge that you lost, even if the probability that the paper made a printing error and that you have actually won equals the probability of winning the New York State lottery. More tendentiously, you might think that you know that you will not win all the next 100 local 1/1000 lotteries even though the probability of this combination of events can be higher than that of winning an exceptionally large lottery. The present coin case seems much more like the former than the latter. Similar things can be said about an example proposed by Vogel (1990). It seems we know that not all 50 beginner golfers will get a hole-in-one on the Heartbreaker, even if the chance of such an event isn't 0. If every golfer's probability of getting a hole-in-one is stipulated to be independent (perhaps they play on different days and have no knowledge of the others' success, for instance) knowledge does not seem to disappear. In fact, the independence assumption only seems to make us more confident that we know.

Thinking lottery propositions are unknowable, then, doesn't force you to reject Anti-skepticism. There are positive reasons to accept Anti-skepticism as well. As Dorr et al. (2014) show, it is easy to transform skepticism about coin toss cases into skepticism about everyday propositions about the future. Suppose that in each one-hour period in autumn there is an independent chance of 1/2 that a leaf will fall off the tree. If you know the leaf will fall off the tree by the end of Autumn,

you would seem to need to accept Anti-skepticism. Lottery propositions, as single events, do not have an analogous probabilistic structure. So it seems that we can't untangle the rejection of Anti-skepticism from skepticism about the future.9 The argument for the inconsistency of the premises dependsas many argu ments doon assumptions in classical logic. Most obviously, the law of excluded middle (LEM) is necessary to establish the claim that figured in the argument for inconsistency above that there is a least number n such that you know that exactly n coins won't all land heads.10 Many think that the LEM should not be accepted for vague statements, and the relevant knowledge ascriptions do seem vague. We can however, give, another, slightly more cumbersome version of the argument that doesn't rely on the LEM.

Given Knowledge & Probability and the fact that you know that all the coins are fair using modus tolens we can infer that you don't know that any of the coins will land heads (or tails). We can also use these 1000 instances of the Restricted or-to-if principle for each n between 1 and 1000.

Antecedent: You know (the first n-1 coins won't land all heads or the nth coin will land tails) and you don't know (the first n-1 coins won't all land heads).

Consequent: You are in a position to know (if the first n-1 coins all land heads then the nth coin will land tails).

By Independence and Adam's Thesis the consequents in the 1000 conditionals each have a probability .5. Given Knowledge & Probability we can use modus tollens to conclude that you are not in a position to know any of the conditionals in the consequent, so we can derive the negation of each of the consequents. Using modus tollens on the 1000 conditionals we can infer the negation of each of the antecedents. Consider the 1000th antecedent: You know the first 999 coins won't all land heads or the nth coin will land tails and you don't know the first n-1 coins won't all land heads. Given Anti-skepticism the first conjunct is true, so using the inference rule (A&B), A |= B, we can derive that the second conjunct is false. We can now infer, by double negation elimination, that you know that the first 999 coins won't all land heads. By repeating this reasoning we can eventually conclude that you know that the first coin will land tails. This gives us an inconsistency. The proof only relies on Modus Tollens, Double Negation Elimination, and (A&B), A |= B, inference rules which a logician who rejects the LEM can still accept. Of course, a non-classical logician may still find ways to get out of this puzzle, but we have shown that merely eliminating the law of excluded middle is not enough.

The Restricted or-to-if might seem, then, the better target. However, or-to-if reasoning is a critical way of gaining knowledge of conditionals, so without a better candidate restriction it's unattractive to discard it. One modification that might do the work is to further restrict it to cases where you know you know the disjunction.

(Further Restricted or-to-if). If you know you know a statement of the form A or B but you do not know that A is true or that B is true, then you are in a position to know that if not A then B.

You might think that considerations along the lines of Williamson's (2000) safety principle (or his margin for error principles) precludes you from knowing that you know that the first n coins didn't land heads (where n is, again, the least number such that you know that the first n coins didn't land head). If you don't know you know it, this further restricted or-to-if principle won't apply. Of course there might still be a lowest m such that you know that you know the first m coins didn't all land heads. By the Further Restricted or-to-if principle you are in a position to know that if the first n-1 coins landed heads, then one of the next n to m coins landed tails (assuming m<2n-1).

The conditional probability of one of the next n to m coins landed tails given that not all of the first n-1 coins landed heads is just 1-1/2m-(n-1).If n =m-1,thentheconditional probability is .75. In this case knowing the conditional is compatible with Restricted Adams' Thesis and Knowledge & Probability. However, you might think it plausible that Knowledge & Probability is weaker than necessary, and your credence in something should be significantly higher than .75 in order to be in a position to know it. All this shows, though, is that m cannot equal n+1. In particular the gap between cases in which you know and cases in which you know you know needs to be sufficiently large to satisfy a strengthening of Knowledge & Probability. As there is no reason to think such gaps should be small in these cases, this is not a problem with this solution.