The Do's and Don'ts of Moorean Shifting

Help, where are my hands?!

“Here is one hand!” The man yells excitedly, as those around him marvel at his achievement: having disproved external world skepticism.

Probably one of the first responses to external world skepticism you will come across is G. E. Moore’s quite straightforward proof: *raises one hand* “here is one hand” *raises the other* “and here is another.” While this is pretty funny, it’s also pretty shit. But although this particular argument is about as persuasive as doing a Peter Hitchens and leaving the conversation, I think the general form of the argument can be quite useful in other spheres than responses to skepticism. To get there, though, let’s first look at why the Moorean shift is a terrible way to respond to skepticism:

Imagine that Richard “Dick” Dickens (or Dick for short) is taking part in a medical trial. The medical researchers are testing a new drug that will let hand-amputees experience the joy of having hands again by inducing a hyper realistic hallucination as of having hands, upon waking from a quick nap. To make the experience fully immersive, the drug also makes the user forget that they have ever lost their hands. For some reason, though, the researchers are simultaneously doing a control trial with handed people who have just received a placebo pill.

Dick now wakes up and finds himself with a set of beautiful, smooth hands. Being a little groggy, he assumes that he does in fact have hands. But as his head clears up, he remembers what situation he is in, and that he might very well be one of the drug users. Luckily for Dick though, he remembers his days as a philosophy student, and he begins chanting that magic phrase that his professor taught him all those years ago: “Here is one hand”… he pauses, and savors the feeling of his peace of mind returning to him… “and here is another!” He breathes a sigh of relief, having now assured himself that his hands are, in fact, the real deal.

Now, I obviously presented this in a, let’s say, colorful way (your amusement is my duty), but I hope you will agree with me that Dick is not exactly acting rationally in this scenario. He would have exactly these beliefs, and perform exactly the same actions, even if he was one of the amputees, meaning he has a clear defeater for his belief in his having hands. It seems clear to me that whether Dick is justified in believing he has hands will depend on the proportion of actual trials to control trials.¹ If he has no idea about the proportion, he should simply withhold judgement.

I assume it’s pretty clear what the analogy is supposed to be here. When Moore tries to refute skepticism by insisting on having hands, it’s the same thing: he’s pointing to some piece of evidence that he would have (or at least think he has), even if the skeptical scenario were true. Another, clearer, way to frame the move is with the classic “one man's modus ponens is another man's modus tollens.” The skeptic provides some argument (here S is some skeptical scenario):

1. If I don’t know I’m not in S, I don’t know I have hands.

2. I don’t know I’m not in S.

3. Therefore I don’t know I have hands.

The Moorean then presents a variation over this argument, granting 1:

4. If I don’t know I’m not in S, I don’t know I have hands.

5. I do know I have hands.

6. Therefore I know I’m not in S.

Since I’m more certain that I have hands than in any abstract philosophical premise the skeptic can come up with for 2, I will be more certain of the second argument than the first, and can thus reject S. I think it’s quite clear that if we here imagine that we are Dick and S is the scenario described before, this is a terrible move to make. The question is how to diagnose the mistake. I think the problem is that this style of argument cannot work when my believing the evidence in question is predicted equally on both the skeptical scenario and the non-skeptical scenario.

It might be useful here to distinguish a prior and posterior logical moment. The prior moment is when I believe things to be roughly as they appear, and the posterior moment is when I come to consider the possibility of a skeptical scenario (analogous to when Dick is groggy, vs when he remembers he’s in an experiment). It’s true that my prior credence in having hands is very high, and so I priorly have very strong evidence against skepticism. But when I come to consider the possibility of a skeptical scenario, I realize that I would have believed the exact same thing priorly, regardless of whether I were in a skeptical scenario or not, meaning I can no longer use my having hands as evidence for not being in a skeptical scenario.

This may be a bit unnecessary, but It’s not analytic philosophy without a bit of unnecessary formalism. We denote the real world hypothesis with R and the skeptical hypothesis with S, and say that B denotes “I believe E”, where E is some piece of evidence that favors R over S (e.g. my having hands). Then if we posteriorly come to see that P(B|S)≈P(B|R), that means that E can no longer be cited as evidence in favor of R. This is because any justification you have for E given R, you would also have for E given S (or you would at least think you had as strong justification). And so, even though E is evidence against S, and even if you end up believing E when the dust has settled, you cannot use E to adjudicate between R and S.²

In other words, you shouldn’t use Moorean shifts when you would still believe the evidence you are citing just as strongly, assuming the thing you’re shifting away from were true.³

But Moorean shifts aren’t all bad. Aspiring philosopher is presented with a strong argument. The conclusion is untenable, so they justifiably Moorean shift. Many such cases! And as you might expect, I think these cases are those not described earlier, i.e., those where your belief in the evidence cited is not equally predicted on the negation of the conclusion. In fact I have even used this sort of move to argue that christians shouldn’t think homosexuality is wrong.

One area where I think Moorean shifts are often very useful is in ethics. For example you may be having a casual conversation with Immanuel Kant, as one so often does, when he presents you with this argument:

1. Lying to a Nazi about where some Jews are is treating a person as merely a means to an end.

2. It is always immoral to treat a person as merely a means to an end.

3. Therefore it is immoral to lie to a Nazi about where some Jews are.

You of course say “what the hell, Kant?! Why do you always have to bring Nazis into this?” After this, you present him with the following argument:

4. Lying to a Nazi about where some Jews are is treating a person as merely a means to an end.

5. It is not immoral to lie to a Nazi about where some Jews are.

6. Therefore it is not always immoral to treat a person as merely a means to an end.

Notice that nothing Kant is saying is providing an error theory for your belief in 5, meaning that you are free to use your higher credence in 5 than in 2 to shift away from 3. Used in this way, the Moorean shift becomes a powerful philosophical tool; we don’t have to go into the weeds with Kant and discuss the merits of the Categorical imperative—so long as some step in his argument is less certain than 5, we are justified in dismissing it.⁴

But it’s of course not only in ethics that this is useful. After having spoken with Kant, you meet Robert Sapolsky, who gives you the following argument:

1. We can statistically predict human actions.

2. If we can statistically predict human actions, we don't have free will.

3. Therefore we don't have free will.

Again, you simply counter with a variation over the argument:

4. We can statistically predict human actions.

5. We do have free will.

6. Therefore it is not the case that if we can statistically predict human actions, we don’t have free will.

You know the gist: here again, Robert Sapolsky’s strange sort of incompatibilism doesn’t provide any error theory for your belief in 5, so it makes sense to infer 6, assuming you actually think 5 is more plausible 2.

Maybe Sapolsky comes up with some theory for why you would be equally likely to believe in free will, even if his view is true. This now has to be treated as a new hypothesis, which we cannot Moorean shift away from. Instead the work simply lies in showing whether this hypothesis or the free will hypothesis is more plausible, without resorting to our strong appearance as of having free will (or whatever reason you have for believing 5). Maybe the hard determinist hypothesis is very parsimonious here, or maybe it’s a much better hypothesis that we simply have free will. But remember that this is now a new discussion—in the original case we were justified in Moorean shifting, it is simply that the shiftee(?) can tweak their hypothesis to make this move no longer justified.

In fact, I think this is just how we should generally think about skeptical scenarios. The skeptical hypothesis is built to predict exactly the same things as the real world hypothesis, so in adjudicating between the two, we cannot point to any experience. Rather, we must see whether there is independent reason to favor one over the other. For Dick this might be that he knows that there are 3 times as many control trials as real trials, and so he is justified in believing he has hands. And with skeptical scenarios, it might be that the skeptical hypothesis ends up having to be much more complicated than the real world hypothesis, to be able to predict our experiences.

1. Well, I actually tend to think that belief-talk is just a vague way of talking about credences, so he should simply proportion his credence in having hands to the proportion of control trials.

2. The reason why I’m not simply citing P(E|S)≈P(E|R) is because some pesky externalist will say that my justification in E depends on whether E is actually the case (or maybe that in R, my evidence is not the appearance that I have hands, but that I actually have hands). But by only pointing to my having the belief that E, we sidestep that issue. The externalist may still think that their view provides a response to skepticism, but that would be a separate argument from the Moorean shift and misses the point I’m trying to make, hence why I just want to sidestep that here.

3. I have made similar points regarding inflationary metaphysics previously.

4. It may sound like this presupposes moral realism, but it doesn’t. If you think moral statements are about something else than objective facts, then that’s of course just what we are talking about here. If you think that we are trying to talk about actual objective facts in moral discourse, and that the truth or falsity of these facts don’t predict anything about our beliefs in them, then you should think that we are never justified in believing any moral proposition.