Philosophical Studies

, Volume 148, Issue 2, pp 249–252

Sosa’s dream


    • Department of PhilosophyUniversity of Arizona
    • Department of PhilosophyUniversity of Arizona

DOI: 10.1007/s11098-008-9309-y

Cite this article as:
Ballantyne, N. & Evans, I. Philos Stud (2010) 148: 249. doi:10.1007/s11098-008-9309-y

Ernest Sosa has lately posed a new skeptical problem and a solution. Setting out a novel account of dreaming, Sosa worries that “the possibility that we dream now threatens not only our supposed perceptual knowledge but even our supposed introspective knowledge, our supposed takings of the given” (2007, p. 9). In what follows, we outline Sosa’s dream argument and evaluate his solution.

According to the “orthodox” model of dreaming, waking states are intrinsically just like dream states (2007, p. 1). In contrast, Sosa claims that there is an intrinsic difference between waking and dreaming: “in dreaming we have no real beliefs” (2007, p. 9).1 On Sosa’s “imagination” model, “[d]reams seem more like imaginings, or stories, or even daydreams, all fictions of a sort, or quasi-fictions” (2007, p. 7). Dreamers are, says Sosa, more like moviegoers than hallucinating subjects.

With the imagination model in play, Sosa reveals his new dreaming problem: “I may ask: do I now think I see a hand? Well, might it not be just a dream? Might I not be only dreaming that I think I see a hand?” (2007, p. 9). Indeed. For “[a] vivid and realistic dream is … subjectively very much like its corresponding reality” (2007, p. 11). However, “[i]f I am only dreaming, then I do not really think I see a hand …” (2007, p. 9).

Sosa’s solution to this problem takes two steps: first, to secure unreflective animal knowledge (2007, pp. 9–13); and to then acquire reflective knowledge (2007, pp. 13–19). Animal knowledge consists in true belief that is safe: “[a] belief that p is safe provided it would have been held only if (most likely) p” (2007, p. 25; cf. 26).2 But on the imagination model, our beliefs are safe with respect to dream worlds. For suppose that Green forms the true belief that p. Green’s belief is safe because there is no nearby dream world where he believes p and p is false. That is the first step of Sosa’s solution. It will be uncontroversial, granting the imagination model. The second step deserves scrutiny, however.

Sosa wishes to keep a further threat at bay: “How … can I non-arbitrarily take myself to be awake, when I cannot distinguish my state internally from that of a realistic dream?” (2007, p. 16). Absent a reflectively defensible argument for thinking he is awake, reflective knowledge eludes him. Such an argument, Sosa thinks, will depend on establishing a difference between waking and dreaming—despite their subjective indiscernibility.

The difference he plumps for is this: “[w]hat enables us to distinguish the two content-identical states is just the fact that in the dream state we do not affirm anything,” whereas in the waking state we can and do affirm many things (2007, pp. 17–18). Facing the new dreaming problem, Sosa claims we have three options: believing we’re awake, disbelieving we’re awake, and suspending belief whether we’re awake. He writes: “If I know that only one of my options is epistemically undefective, making it the best option, that then would seem the rational option for me to take” (2007, p. 18). Critically, if one believes that one is awake, one believes truly. If one disbelieves that one is awake, one believes falsely. And if one suspends belief whether one is awake, one thereby misses the truth (while knowing it). These options and outcomes are summarized here:





True belief


False belief


No belief

Sosa’s point is that once someone reflects on the options, only believing is not defective. Downstream of reflection, then, it is rational to believe that one is awake. Sosa has thus furnished an argument for thinking he is awake. Reflective knowledge follows and the skeptical problem is discharged.

We object. Sosa’s solution supposes that the three options are in fact available. Note well: on the imagination model, if one is dreaming, these options are not available—believing included. But then believing would not be the rational option. It would be no option at all. Therefore, in order for one to conclude that believing is the rational option, one must have a reason to think one is awake. Sosa cannot non-arbitrarily take himself to be awake when he cannot distinguish his state internally from that of a realistic dream (2007, p. 16). He sought a reflectively defensible argument for thinking he is awake. The argument he supplied, though, must assume that he is awake; it takes as a premise precisely what it seeks to establish.

Is Sosa’s dream dead? Jesse Jackson once cautioned us: “No one should negotiate their dreams. Dreams must be free to flee and fly high.” In that spirit, we will see whether Sosa’s dream may come true by considering some replies to our argument.

Reply 1: Sosa does take himself to be in a position to distinguish waking from dreaming states, where “two scenarios are indistinguishable if, and only if, one can tell neither that one is in the first and not the second when that is so, nor that one is in the second and not the first when that is so” (2007, p. 18, fn. 16). Dreaming and waking, says Sosa, are distinguishable in just this way. He relies on “your ability to tell that [you are awake], when you [are], despite your inability to tell that [you are dreaming] when [you are]” (2007, p. 17, fn. 15). When awake, one can tell that one is awake and therefore that one has the three options available.

Objection 1: We are unsure what an “ability to tell” that one is awake amounts to. Perhaps it means that when one is awake, one is rational to so believe. If so, Sosa illegitimately takes as a premise what he needs to prove, as we argued above. If not, it is unclear how Sosa is rational to believe that the three options are available. Either way, invoking the “distinguishability” of waking from dreaming brings no relief.

Reply 2: Sosa may offer an argument that does not include as a premise the claim that he is awake.
  • (P1) I have three options.

  • (P2) Believing is the rational option.

  • (C) I am rational to believe that I am awake.

Can’t Sosa enjoy non-inferential knowledge that P1 and then employ it in the above argument?

Objection 2: Sosa may run P1–C. It is a broadly externalist solution to the dreaming problem. Sosa shouldn’t take it, however, for very plausibly, the following principle falls out of his view: if a premise presupposes that one is awake, then it will not serve as a non-arbitrary reason for thinking one is awake. P1 runs afoul of this principle and is thus unhelpful to Sosa. To put the point otherwise, recall that Sosa is engaging with the skeptic; he seeks a reflectively defensible argument. But surely no skeptic will grant that Sosa is entitled to P1 as a premise.

Reply 3: If Sosa’s imagination model is correct, then the belief that one is awake has a cogito-like status—viz., necessarily, if one believes that one is awake, then one is awake. And once one recognizes that the belief that one is awake can’t be mistaken, one has reason to think that belief is rational.

Objection 3: There is a critical difference between the status of Descartes’ cogito and Sosa’s dream-cogito. It isn’t possible for it to seem to one that one exists unless one exists. But it is possible for it to seem to one that one has a belief even when one doesn’t. That difference makes all the difference.

Let us conclude. Sosa has argued that if he is awake, then believing so is the best option. But we have observed that if Sosa is dreaming, then believing is not the best option, for it is not so much as possible. Without a non-arbitrary reason to think he is not dreaming, Sosa has no reason to think he is awake.


Though we’re unsure what “intrinsic” amounts to here, the contrast between the two models is clear enough.


For Sosa’s more nuanced definition of safety, see (2007, p. 26). The nuances don’t matter for what we say here.



The authors, listed in alphabetical order, are grateful for comments to Stew Cohen, Terry Horgan, Keith Lehrer, and Josh Rasmussen. Thanks are due to Ernest Sosa for helpful correspondence. NB would like to thank the Social Sciences and Humanities Research Council of Canada for support.

Copyright information

© Springer Science+Business Media B.V. 2008