Serious Theories and Skeptical Theories: Why You Are Probably Not a Brain in a Vat

Here, I explain the difference between serious scientific theories and BIV-like theories.*

[ *Based on: “Serious Theories and Skeptical Theories: Why You Are Probably Not a Brain in a Vat,” Philosophical Studies 173 (2016): 1031-52. ]

This, by the way, is the paper that has caused me to receive the most invitations to contribute to predatory journals and conferences. They all want me to contribute to their neuroscience conferences and journals because they were so “impressed” with my paper about brain science.

1. The Skeptic’s Argument

Imagine that scientists have figured out how to keep a brain alive in a vat of fluid. They read the brain’s electrical activity and stimulate the brain electrically to produce a perfect simulation of life in the early 21st century (before BIV technology was developed). They might even program a simulation of reading a blog post about a brain in a vat.

Skeptics point out that there is no way to refute the hypothesis that you’re a Brain in a Vat (“BIVH”). Anything you experience can just be explained as part of the simulation that the scientists programmed. So you can never know that you’re not a BIV.

A strong version of the argument claims that you can never be justified in believing any contingent proposition about the external world. Let P be some contingent, external-world proposition, like “I have hands” or “there’s a squirrel”:

1. You have justification for believing P only if you have justification for believing ~BIVH.

2. You have no justification for believing ~BIVH. Why:

   1. BIVH predicts that you would have the same sort of sensory experiences that you actually have.

   2. In general, if H predicts E, then E isn’t evidence against H.

   3. Your sensory experiences form the only evidence you have relevant to BIVH.

   4. So you have no evidence against BIVH. (From a, b, c)

   5. You have justification for ~BIVH only if you have evidence against BIVH.

   6. So you have no justification for ~BIVH. (From d, e)

3. So you have no justification for believing P. (From 1, 2)

Other skeptical scenarios work similarly (the deceiving God hypothesis, the dream hypothesis).

I’m going to argue that BIVH is a bad theory because it is impossible to support. By contrast, the hypothesis that we’re perceiving the real world (RWH) is supported by our evidence, leaving RWH much more probable than BIVH.

2. Stuff About Probability

The probability of a proposition (for you) is something like its degree of justification on your evidence, or the degree of confidence you should have in it.

An especially important principle of probability is Bayes’ Theorem:

P(h|e) = P(h) * P(e|h) / P(e)

“P(h|e)” is read “the probability of h given e”. This is the confidence you should have in h after you discover evidence e. This is a function of P(h) (the initial probability of h, before you discovered e), P(e|h) (the probability that e would be true if h were true, or the degree to which h predicts e), and P(e) (the initial probability of e).

3. The Broad BIVH Is Disconfirmed

The Improbability of Coherence

Suppose we read BIVH as claiming merely that you’re a BIV who is being stimulated by the scientists to have some experiences or other. This is the broad reading of the BIVH.

In this case, premise 2a in the skeptic’s argument is false:

2a. BIVH predicts that you would have the same sort of sensory experiences that you actually have.

The broad BIVH doesn’t predict this; it is compatible with your having any set of experiences that the scientists could cause. The set of experiences that would constitute a perfect simulation of a normal life are only a tiny range out of all the possible experiences a brain could be stimulated to have.

You may not realize just how tiny that range is, so let me elaborate. I once programmed my computer to generate random images. I.e., it would assign a random color to each pixel in a 1000 x 1000 grid. I had it generate many of these images, then play a slide show.

What does the slide show look like? It looks like static, of the sort that used to appear on television screens when there was nothing on. None of the images looks like a picture of a puppy, or a nebula, or anything else at all that you would find noteworthy. You could run that program all day for the rest of your life, and in all probability, it would never generate a single image that looked like anything.

That is an illustration of the fact that almost all possible images look like static.

A generalization of this is the fact that almost all sets of experiences fail to appear like perceptions of a real world—or indeed, like experiences of anything. Since the broad BIVH doesn’t predict anything about your experiences, and close to 0% of experiences resemble experiences of a normal person perceiving the real world, the probability of your having experiences like those of a normal person, given BIVH, is close to 0. So, in Bayes’ Theorem, P(e|BIVH) is extremely low, so P(BIVH|e) is also extremely low.

Objection

P(e|BIVH) isn’t all that low, once you think about the motives and abilities of the scientists. Sure, they could program a brain to experience static, but it’s unlikely that they would want to. It’s more likely, a priori, that they would want to make a simulation of living in the real world.

Reply: Fair enough. We could assign probabilities by thinking about the characteristics of the scientists who are responsible for the BIV’s experiences, rather than by thinking directly about the possible experiences a BIV could have. It’s still true that only a very small range of possibilities would result in your having experiences like your current ones.

First, if a group of scientists were to make a BIV, there is a wide range of capabilities that they could have. In only a small part of that range would they be able to make the brain experience a perfect simulation of real life, with no glitches, errors, or shortcomings.

Second, if a group of scientists were to make a BIV, there is only a narrow range of motivations that they could have that would result in their wanting the BIV to just experience a mundane, early-21st-century life.

What could the scientists’ values be, such that this would make sense? If they were benevolent, you’d expect your life (I presume) to be much better than it is, with more enjoyment and less suffering. If they were malicious, there’d be a lot more suffering. If they valued virtue, your life would be optimized for making you develop moral virtue. If they wanted entertainment, your life should be maximally entertaining to watch. Etc. About the only motivation you can ascribe to the scientists is a desire to make everything look exactly as if there weren’t anyone in control of the simulation.

So there are two parameters in the BIV theory (the motivations and capabilities of the scientists) that have to be set to very specific values in order to predict the character of your experiences, and there is no independent motivation for expecting them to have those values, so it’s purely ad hoc.

Those parameters, in fact, let you accommodate pretty much any experience. No matter what your experience was like, you could say, “The scientists wanted me to experience exactly that.”

The Real-World Hypothesis

Maybe the RWH also has a low likelihood (P(e|h)). You might say:

“The RWH just says that we’re perceiving the real world. But this alone doesn’t explain the course of your experiences. To explain your actual experiences, you have to make very specific, ad hoc posits about what objects exist out there and what properties they have. That’s just like what you said about the BIVH!”

In reply, first, the feature of experience I wanted explained was the fairly general fact that your experiences are like those of a normal person—not the specific details of those experiences. The BIVH doesn’t even explain that general fact without making specific, ad hoc assumptions about the scientists. The RWH doesn’t need ad hoc assumptions for that. So they’re not parallel.

Second, you can easily describe experiences that would disconfirm RWH, whereas it’s hard to think of any experiences that would disconfirm BIVH. If your experiences looked random, like static, that would disconfirm RWH. Maybe RWH doesn’t entail that you must have coherent experiences, but it would at least lead you to expect that.

Alternately, you could have experienced “glitches in the matrix”—things that looked like the result of computer errors.

Or you could experience a life that seemed peculiarly well-designed to promote some intelligible value. E.g., if your life seemed optimized for producing pleasure, or virtue, or aesthetic value, or intellectual stimulation, that would support BIVH or the Simulation Theory over RWH.

So RWH really isn’t parallel to the skeptical hypotheses. RWH is falsifiable in a way that the skeptical theories are not.

Why care about falsifiability?

More generally, why care about whether it’s possible to have evidence against a theory?

It’s a theorem of probability that e raises the probability of h only if ~e lowers the probability of h. I.e.,

P(h|e) > P(h) iff P(h|~e) < P(h)

Therefore, if nothing counts as evidence against your theory, then nothing counts as evidence for it either.

The problem with skeptical scenarios is that we have no evidence for them, because (roughly) nothing would count as evidence either for or against them, because they have parameters that enable them to accommodate anything that happens in the same way.

By contrast, RWH could be falsified by the sort of evidence described above. Thus, the failure of those things to happen constitutes evidence for RWH.

4. The Narrow BIVH

Suppose we redefine the BIVH to stipulate that, according to the theory, the scientists have the capacity for making perfect simulations, and they want nothing more than to create an experience just like an ordinary, mundane life. In that case, P(e|BIVH) is essentially 1. So this technically avoids the objection of the preceding section.

But of course that doesn’t get you anywhere. You just trade low likelihood for low prior probability. I.e., by building those stipulations into the hypothesis, you just lower the initial probability of the hypothesis, P(BIVH), by exactly the same factor as you increased the likelihood, P(e|BIVH).

5. Scientific vs. BIV-like Theories

Say you’re living in 1787, and you’re trying to understand the observed motions of the planets in the sky. Kepler has proposed three laws of planetary motion, which include that the planets and Earth travel elliptical paths, with the sun at one focus (plus 2 more laws). But why do they move like this? Newton proposes his Theory of Gravity, plus his general laws of motion, which turn out to explain why the Earth and planets would travel in these paths. So that’s cool.

Now suppose a philosopher shows up and proposes an alternative theory: Maybe God pushes the planets around, directly. Wherever you see them, the explanation is that God decided to push them there. This is a very simple explanation, and it only appeals to causes (God) that nearly everyone in Newton’s time already accepted. So it’s a great theory, right?

No, it’s a crappy theory. Which is why no one in fact proposed it, and no scientist would ever propose something like that.

The problem with the God Theory of Planetary Motion is that it explains too much: no matter what the planets did, you could give the same explanation, “God did it.”

By contrast, the Newtonian theory cannot accommodate just anything (at least, not naturally). If the planets moved in square paths, or jumped around randomly, this would be really hard for Newton to explain. These things wouldn’t strictly entail that Newton’s theory was false, since you could always posit unknown other forces, or teleportation abilities. But Newton’s theory wouldn’t be able to explain these things naturally, nor could it explain them in the same way that it explains the actual observed motions.

Conclusion

What’s wrong with skeptical hypotheses is what’s wrong with many unscientific theories in general: they are untestable, because they could explain any evidence.