The Raven Paradox

The paradox

Consider the hypothesis: "All ravens are black." Observing a black raven seems to confirm it. Each new black raven is a data point in favor of the generalization.

Now notice that "All ravens are black" is logically equivalent to "Everything that is not black is not a raven." These two statements are true in exactly the same possible worlds. They say the same thing in different words.

Observing a red apple confirms "Everything that is not black is not a raven." The apple is not black, and it is not a raven, so it fits the pattern. But if a red apple confirms that statement, and that statement is logically equivalent to "All ravens are black," then the red apple also confirms that all ravens are black.

Your morning fruit bowl is evidence about ornithology.

The main responses

Hempel accepted the conclusion and tried to defuse the surprise. A red apple does confirm that all ravens are black, just very weakly. The confirmation is so small, given the vast number of non-black things in the world, that it's practically negligible. The paradox feels wrong because we're comparing a negligible confirmation against our strong background knowledge, not because the logic fails.

The Bayesian treatment adds more precision. Whether a red apple confirms the raven hypothesis depends on how you sampled it. If you randomly selected an object from the world and found a non-black non-raven, that does raise the probability of the hypothesis slightly. But if you specifically went looking for non-black things, the apple tells you almost nothing, because the sampling procedure was biased toward finding confirming instances.

What this reveals about induction

The paradox exposes a gap between the logical form of confirmation and our actual scientific practice. We treat black ravens as evidence for the raven hypothesis and red apples as irrelevant, but that asymmetry is hard to ground purely in logic.

The evidence that feels meaningful depends on background assumptions about what we're sampling, how we're looking, and what the hypothesis is really about. A purely formal account of confirmation that ignores those background conditions will keep generating paradoxes like this one.