Ernest Sosa has argued that we do not form beliefs when we dream. If I dream that a tiger is chasing me, I do not really believe that a tiger is chasing me. If I dream that I am saying to myself "I'm awake!" I do not really believe that I'm awake. Real beliefs are more deeply integrated than are these dream-mirages with my standing attitudes and my waking behavior. If so, it follows that if I genuinely believe that I'm awake, necessarily I am correct; and conversely if I believe I'm dreaming, necessarily I'm wrong. The first belief is self-verifying; the second self-defeating. Deliberating between them, I should not choose the self-defeating one, nor should I decline to choose, as though these two options were of equal epistemic merit. Rather, I should settle upon the self-verifying belief that I am awake. Thus, dream skepticism is vanquished!
One nice thing about Sosa's argument is that it does not require that dream experience differ from waking experience in any of the ways that dreams and waking life are sometimes thought to differ (e.g., dream experience needn't be gappier, or less coherent, or more like imagery experience than like perceptual experience). The argument would still work even if dream experience were, as Sosa says, "internally indistinguishable" from waking experience.
This seeming strength of the argument, though, seems to me to signal a flaw. Suppose that dreaming life is in fact in every respect phenomenally indistinguishable from waking life -- indistinguishable from the inside, as it were -- and accordingly that I could easily experience exactly *this* while sleeping; and furthermore suppose that I dream extensively every night and that most of my dreams have mundane everyday content just like that of my waking life. None of this should affect Sosa's argument. And suppose further that I am in fact now awake (and thus capable of forming beliefs about whether I am dreaming, per Sosa), and that I know that due to a horrible disease I acquired at age 35, I spend almost all of my life in dreaming sleep so that 90% of the time when I have experiences of this sort (as if in my office, thinking about philosophy, working on a blog post...) I am sleeping. Unless there's something I'm aware of that points toward this not being a dream, shouldn't I hesitate before jumping to the conclusion that this time, unlike all those others, I really am awake? Probabilities, frequencies, and degrees of resemblance seem to matter, but there is no room for them in Sosa's argument.
Maybe we don't form beliefs when we dream -- Sosa, and also Jonathan Ichikawa, have presented some interesting arguments along those lines. But if there is no difference from the inside between dreams and waking, then my dreaming self, when he was dreaming about considering dream skepticism (e.g., here) did something that was phenomenally indistinguishable from forming the belief that he was thinking about philosophy, something that was phenomenally indistinguishable from forming the belief that was affirming or denying or suspending belief about the question of whether he was dreaming -- and then the question becomes: How do I know that I'm not doing that very same thing right now?
Call it dream-shadow believing: It's like believing, except that it happens only in dreams. If dream-shadow believing is possible, then if I dream-shadow believe that I am dreaming, necessarily I am correct; if I dream-shadow believe that I am awake, necessarily I am wrong. The first is self-verifying, the second self-defeating. The skeptic can now ask: Should I try to form the belief that I am awake or instead the dream-shadow belief that I am dreaming? -- and to this question, Sosa's argument gives no answer.
Update, 3:28 pm:
Jonathan Ichikawa has kindly reminded me that he presented similar arguments against Sosa back in 2007 -- which I knew (in fact, Jonathan thanks me in the article for my comments) but somehow forgot. Jonathan runs the reply a bit differently, in terms of quasi-affirming (which is neutral between genuine affirming and something phenomenally indistinguishable from affirming, but which one can do in a dream) rather than in terms of dream-shadow believing. Perhaps my dream-shadow belief formulation enables a parity-of-argument objection, if (given the phenomenal indistinguishability of dreams and waking) the argument that one should settle on self-verifying dream-shadow belief is as strong an argument as is Sosa's original argument.