Anthropic arguments based on selection effects for observers have been claimed to succesfully explain the measured value of the cosmological constant.In this talk I review the fundations of such claims in the context of probability theory and show that different (and equally legitimate) ways of assigning probabilities to candidate universes lead to totally different anthropic predictions. As an explicit example, I discuss a weighting scheme based on the total number of possible observations that observers can carry out over the entire lifetime of the Universe. I show that this leads to an extremely small probability for observing a value of the cosmological constant equal to or greater than what we now measure, in marked contrast with the usual result.
I also discuss principles of consistent probabilistic reasoning, showing that the anthropic principle as applied in most of the literature is logically inconsistent. I conclude that current implementations of the anthropic principle display a worrysome lack of predictivity, and cannot be used to explain the value of the cosmological constant, nor, likely, any other physical parameters.