Shell games, information, and counterfactuals

Abstract

Cohen and Meskin 2006 have recently proposed a novel counterfactual account of information. I argue that it is a step down from its intended target, namely Dretske's 1981 theory of information. Thinking of the information carried by signals in terms of counterfactuals leads to falsely diagnosing bona fide instances of information transmission as not being instances of information transmission at all, with major loss of explanatory power.

Notes

¹I want to thank Eddy Nahmias, Gualtiero Piccinini, Ben Sheredos and two anonymous referees of this journal for their very useful feedback on previous drafts of this paper. I also want to thank an anonymous referee of The Philosophical Quarterly for excellent comments on an expanded version of this paper that have helped with this shorter version as well.

²In the absence of the non-vacuity proviso, anything would carry information about y's being G whenever it is necessary that y is G. In such case, ‘if y were not G then x would not be F’ would have an impossible antecedent, and counterpossibles are commonly assumed to be vacuously true (Lewis [1973]; but see Brogaard and Salerno [2007] for a dissenting opinion).

³Dretske [1981: 245] also formulated an alternative theory of information, according to which x's being F carries the information that y is G iff there is a nomic regularity such that, given that x is F, y must be G by virtue of a law of nature or logic. I will not discuss this second formulation in what follows, but I want to emphasize that the virtues of the Dretskean account extolled in this paper do not apply to the nomic regularity version.

⁴There are two assumptions here. The first is that the two recipients know that the peanut is under one of the four shells. The second is that shells 1 and 2 are put back down after having been turned over, without changing the peanut's location.

⁵The problem here is not merely that we are not told whether the counterfactual is ultimately true or false in the circumstances described in the Shell Game, but that we are not told what exactly would make the counterfactual true or false and how we could go about figuring out its truth-value.

⁶According to Cohen and Meskin [2006], another advantage of their account is that it does not make use of conditional inverse probabilities. Their use in the context of Dretske's [1981] project has been criticized on the grounds that neither a frequentist nor a propensity nor a degree of belief interpretation of such probabilities appears to be compatible with the overarching objectives of Dretske's project [Loewer 1983].

⁷An additional problem, acknowledged by Cohen and Meskin [2006: 342] but not considered fatal, is that it is unclear whether the doxastic element is entirely absent from Lewis's [1973] semantics for counterfactuals. This is because ‘the relative importance of respects of comparison that underlie the comparative similarity of worlds … [is] … a highly volatile matter, varying with every shift of context and interest’ [1973: 92]. I will not discuss this further threat to the applicability of Lewis's semantics to Cohen and Meskin's [2006] project.

⁸This is just one of the innumerable contextually relevant scenarios in which the signal (shell 3 is empty) might occur—without any gratuitous departure from actuality—when the state of affairs the signal is about (the peanut is under shell 4) does not occur.

⁹This difference in neural firing would have to be explained differently depending on whether we assume the world to be deterministic or indeterministic. The distinction between deterministic and indeterministic cases is not central to my main point, so I will disregard it in what follows.

¹⁰It is primary because anything else information users do with information presupposes that such information is learned.

¹¹I argue in Scarantino [2007] that Dretske's [1981] theory ultimately fails as a naturalistic theory of information, but for reasons different from those suggested by Cohen and Meskin [2006].