Coel has put up a post on one of philosophy’s favourite topics: Brain in a vat (BIV). I’ll try to get to other sources, such as the Massimo Pigliucci post Coel refers to, but for now I’ll respond to Coel’s post only.
So, “[the BIV problem] supposes that we are a brain kept alive in a vat, being fed with a stream of inputs generated by an Evil Genius. Everything that we experience as sense data is not real, but is artificially simulated and fed to us.”
First, a quick comment on the Tatsuya Ishida cartoon. There should have been one brain in a vat, and both speech bubbles should have come from the same brain: one taking the part of the brain that is supposed to be the main character, and the other is the internal auditory perception of what the brain presumes to be the other character. Of course it could be that two brains in vats are being manipulated in conversation, but that then seems too much like an actual conversation between brains, and sort of misses the main point of BIV: all perceptions, whether thoughts of the lone BIV, or stimulated auditory perceptions that convince the BIV that it has heard another speak (and visually aware of the other, etc.). But that’s just nit picking.
Back to the problem. How can you tell if you are a BIV?
You can’t. Or rather I should say that so far we don’t know if you could tell or not. The significant point is that we can’t positively know we could tell the difference, between being a real brain in a human body, situated in the physical world we perceive, or being a BIV being stimulated to so perceive the world.
Coel’s parsimony point is right, but I disagree with his use of it. Parsimony is merely a pragmatic tool in an epistemically uncertain world. It assures us of nothing. The proper use of parsimony is that if there are two explanations (or theories) that fit the same evidence then we might as well use the simplest. Since both theories fit the evidence then by definition we can’t tell which is correct, and at that point it doesn’t actually matter.
Example: Theories X and Y fit evidence A, B, C. X is simpler than Y, so use X.
New evidence D also supports X and Y, so continue to use X.
Then, new evidence E comes along that supports Y, but not X. All of a sudden Y is looking interesting. E might be so convincing that it clearly falsifies X and supports Y. Then Y is the theory to use.
Later, evidence F, G might come along, some of which shows that the experiments that affirmed E were faulty, and F and G also support X and not Y. X is back on the front foot.
That’s how it works. In the case of the BIV problem we have zero evidence either way. We have no evidence supporting BIV, and not supporting the falsity of BIV. We cannot resolve it with any evidence we have.
So, what should we do?
Well, think about the consequences of perceiving the real world and being a BIV perceiving a stimulated fictional world.
What if we really are brains in vats? Or, following my criticism of the cartoon above, what if I really am a BIV, and you and everyone else I perceive are stimulated fictions? What can I do about it?
Bear in mind that I can’t even be sure I’m an actually brain in an actual vat either. I might be merely a computer simulation: of a BIV being stimulated to perceive a real world.
This scenario is so fucked up compared to my perception of a real world that there really isn’t much I can do about it. I might want to get all revolutionary on the Evil Genius’s ass; but really, if he’s controlling my brain (or simulating the control of a brain) then he’s controlling my revolutionary thoughts too. He’s actually making my brain wonder if it’s a BIV! The cruel bastard!
In the end I’m left with the correct use of parsimony. It seems like I’m perceiving a real physical world containing other real physical people, so why not simply use that as a working conclusion? Work in progress. Just like all the rest of science.
Note that this has nothing at all to do with thinking about a finite subset of infinite possibilities, or any of that crap.
“Any statement not supported by evidence is most likely to be wrong and thus should be discarded.”
No! This is not how it is. Any statement not supported by evidence is NOT most likely to be wrong! We simply can’t say if its true or false at all. We can’t say anything much about it at all, except that if we don’t have evidence to support it then it’s indistinguishable from a statement that is false, and as such is as useless as a statement that is false. Using terms like ‘most likely’ is totally inappropriate.
Take the statement, “Praying to Jesus gets results, sometimes.” Well, people pray often, and most times they get no result. Sometimes people pray and get the result they pray for. This is true for many sports competitions, since many people seem to pray to Jesus for victory in sports. But since someone wins and someone loses (barring the anathema of US sports – a draw) the actual result is indistinguishable from coincidence. Prayer, as a determiner of sports results might actually work. Maybe Jesus really does dish out results according to prayer and some other very complicated cosmic data he uses. But to us the use of prayer is indistinguishable from useless. [Note, that’s not to say there are not other psychological consequences of prayer.]
“Regardless of whether the “world” is real or simulated, the standard scientific “real-world model” gives the greatest parsimony and predictive power in describing that world.”
Not so. The BIV, if that is the state of affairs, only ‘thinks’ the real-world model gives predictive power at all. The actual laws of the universe that the Evil Genius inhabits might be quite different from the one that the BIV thinks it inhabits, with very different laws. Of course the BIV can’t imagine what those laws might be, because the BIV is only aware of the ones it is programmed to perceive. All the ‘predictions’ the BIV thought it observed never happened. Have you ever noticed in a dream how natural and obvious some of that weird shit appears to be, how convincing it is, when in the dream? Well, why should that not be the case for the BIV? If you are a BIV then you haven’t ‘woken up’ yet to observe how false all your perceptions are; you are still in the ‘dream’ believing you can make predictions.
“Any departure from that standard model would result in a worse account (one that is less parsimonious and less predictive about that stream of experiences). That’s because science’s models do work very well about our world and are the best that we have.”
Not they don’t, if you are a BIV! They only appear to work very well.
“One might object that in a “brain in vat” scenario the Evil Genius could feed us any stream of experience he liked, with no rhyme or reason to it, no regularities and no predictability. That is indeed possible, but then the stream of experiences could only be described essentially as a streamed video tape, which is the most information-hungry type of model and one that would be totally useless at predicting anything.”
It doesn’t matter how information hungry the theory is from the perspective of the brain, real or BIV. This has nothing to do with it at all. The computer simulation of rolls of a dice are more information hungry than rolling an actual device since computer simulations are built on information hungry computers. But that doesn’t mean we should make a statement, “All simulations of dice in computers are actually real dice, since simulating a dice is more information-hungry, and by the rule of parsimony we should not believe that computers are ever simulating dice.”
“In contrast the “real world” model is very compact in that all one needs are the basic laws of physics and all else follows from that. This model has a vast amount of predictive power, as we know from the fact that engineering works, and planes we build fly, and predictions we make for solar eclipses come true.”
No! You, the BIV, only THINK planes fly, if indeed you are a BIV.
“The brain-in-vat model has no capacity to make any predictions at all about the stream of experiences, unless we make all sorts of assumptions about the Evil Genius and why he is feeding us the stream of experiences.”
Well, where in the BIV notion does it require us to be able to say anything at all about the Evil Genius? If I am a BIV then any assumptions I might presume to make are untestable bollocks.
But, the BIV model does make predictions. It predicts that the brain so programmed can make predictions. It can make predictions about its imaginary world, and it can make predictions about being a brain in a vat. They simply aren’t testable predictions.
“Yet, ex hypothesis, we can never have any information about the Evil Genius or his doings.”
Exactly. Including never having the information to know that you are not a BIV.
“The only way of regaining any degree of parsimony or predictive power…”
No, no, no! If you are a BIV you get only the predictive power that the system controlling you, or simulating you, gives you; which, if you are a BIV, is this very prediction that you are or could be a BIV. And it may all be fake. As a BIV you don’t get any say in anything! You don’t get to know anything. The very fact that you are wondering if you’re a BIV is a consequence of being that very BIV, but only because the stimulations or simulations produced by the Evil Genius allow it. That you are predicting the consequences of being a BIV is predicted by being a BIV, if you are a BIV.
Even if the Evil Genius attached some physical eyes onto the brain and allowed you to see him waving at you from outside the vat, you would have not way of knowing which of the following just occurred:
1) You are a BIV and just saw the actual Evil Genius.
2) You are a BIV hallucinating.
3) You are a real embodied brain in a real physical world hallucinating.
This is just like revelations from God that some religious people report, or when some ‘crazy’ person hears the voice of Jesus and neurologists see his auditory cortex buzzing away as if hearing real voices. Religious personal revelations might be real or they might be imagined. These are indistinguishable events. We reject them only because they don’t seem convincing to us most of the time, and because there is not convincing evidence. Though remember, to many religious people, belief is the parsimonious result.
And this is how we deal with the BIV.
Imagine I am a BIV. I am stimulated to perceive a physical world. That fake physical world model results in me learning some physics (maybe fake physics) and evolution (maybe fake evolution) and I conclude that I am evolved from animals that didn’t have brains and animals that had simple brains. With all that I conclude that I am a human animal. one of many, that has a brain that has quite a fertile imagination, and imagines it might be a brain in a vat.
Or, imagine I’m a real human in a physical world. That physical world results in me learning some physics and evolution and I conclude that I am evolved from animals that didn’t have brains and animals that had simple brains. With all that I conclude that I am a human animal. one of many, that has a brain that has quite a fertile imagination, and imagines it might be a brain in a vat.
These are truly indistinguishable. Otherwise we wouldn’t be having this debate. But note that one is a subset of the other. The simpler one, the latter, seems to work in just the same way as the former, except that the former has some features (me being a BIV) that I can’t actually observe or test. So, according to the principle (i.e. suggestion, recommendation) of parsimony I might as well go along with the latter.
Note that this is almost an arbitrary choice; but not quite. I could live my life as if I’m a BIV. Suppose I did so in one of two ways:
1) I live my life as if I’m a BIV, but I keep it to myself. I never tell anyone. I even try to fool myself that I don’t believe it. Other than a few personal psychotic moments how would this be different from believing I’m a real person in a real world?
2) I live my life as if I’m a BIV, but I try to tell everyone? But hold on, why would I? Since I’m a BIV what good would it do telling my imaginary fellow humans that don’t actually exist except as figments of my BIV? But suppose I did, and those very same figments of my BIV locked me up in an asylum. Well, it’s only an imaginary asylum.
The problem with the BIV perspective is not that it’s not real, but that it’s fucked up to think it is real. It just seems to add a layer of complexity that is unevidenced and unnecessary. I might as well go along with the apparent reality I perceive, whether it’s a real reality or a stimulated one or a simulated one.
Note that this is also how I see religion: an unevidenced and unnecessary level of complexity added to our experienced life; pretending we are God’s brains in his vat of a universe. So fucked up that it’s no wonder we get the craziness. But note also that it could all be true – all of it! Islam AND Christianity – all true but we poor brains in God’s vat simply don’t get the big picture; and William Lane Craig is right that all the dead children have been done a great favour. Religion is as exactly as fucked up as imagining we’re brains in vats. Once you start adding imaginary stuff on top of evidenced stuff, anything goes.
“The in-a-vat wrapper to the real-world model is along the same lines, a vast and needless complication that doesn’t in any way improve the model’s fit to evidence. We could just as well imagine any number of other parallel meta-realities that make no difference to what we experience.”
That’s right. That’s all it boils down to. A rather arbitrary choice, made because it makes life simpler, even if only apparently so.
“Since there are an infinite number of such possibilities the chances of any random one of them being actually true is infinitesimal.”
This is totally irrelevant. It wouldn’t matter if there were only two possibilities: 1) real world, as we actually perceive it; 2) BIV, stimulated to perceive the world as we actually perceive it.
“Thus, by the adoption of the usual scientific method, invoking Occam’s razor and principles of parsimony and the need for predictive power, we can reject brain-in-a-vat scenarios.”
No. Only the principle of parsimony is necessary. Predictive power is of no use, since a BIV makes the same predictions. We cannot reject the BIV with any conviction. Instead we simply choose to reject it because that’s easier. There is no other convincing reason.
“That, of course, does not mean that they can be absolutely ruled out, any more than we can rule out any other hypothesis designed to leave no discernible trace at all on our experience of the world (apophatic theologians are particularly good at inventing these), …”
Agreed.
“… but the chances of any such suggestion being true is too low to merit taking it seriously.”
No. It’s nothing to do with chance, probability, possibilities, or anything else. It’s simply taking the simpler option that doesn’t require that we sustain belief in things (Evil Genius, vat, disembodied stimulated brain) that we cannot test.
Coel refers to Descartes and his Cartesian Doubt, whereby Descartes digs deep into a reasoned sceptical investigation to figure out what lies at the bottom when all our uncertainties are stripped away. His working conclusion is cogito ergo sum, I think therefore I am. From there Descartes cocks it all up, because he already believes in God and he just can’t drum up enough doubt about that.
But his Cogito is about the only place I can see we can start from. Coming up from there I can’t see any difference between our perceived reality and solipsism, or a BIV. That’s the only starting position I have, and that’s were I take it from here: Contingency of Knowledge.
Ramblings 
Great post, as usual. You’ve got the right of this one.
If I remember correctly, Pigliuicci was arguing that BIV=false was not more parsimonious, but he wasn’t making clear (as you did here) that parsimony doesn’t have anything to say about what we can or can’t know, which is really the issue. This is why science generally (and wisely) eschews metaphysics.
“From there Descartes cocks it all up, because he already believes in God and he just can’t drum up enough doubt about that.”
[applause]
Thanks Asher. I’ll try to get to back to Pigliuicci if I can. I need a re-read before commenting on it. It seems overly burdened with unnecessary philosophical detail on first reading.
Hi Ron,
Thanks for your analysis of my post. Here’s some responses:
I still maintain that my claim there is right, and don’t see how what you’ve said counters my argument.
OK, but that’s one statement. Now let’s write down some more statements:
Praying to Zeus gets results, sometimes.
Singing to Baal gets pineapples, always.
Jumping to Ra paints peanuts, occasionally.
… and we continue to list all possible such statements. I still maintain that the vast majority of them will be wrong, and thus that my claim (“any statement not supported by evidence is most likely wrong”) is true.
If you are a BIV, then the “standard scientific real-world” model *does* give good predictive power about the BIV’s stream-of-experiences, which is what I called the BIV’s “world”. And that is true, our scientific model does give good predictive power about what we experience. That is what I claimed.
You’re right that, in the BIV scenario, the “standard scientific real-world” does not give good predictions about the meta-reality of the Evil Genius, but that was not my claim.
If we’re talking about the “stream of experiences” then appearing to work well and working well are the same thing. Again, the claim was about the S-of-E, not about the meta-reality.
Ditto last answer. The claim is about what happens in the stream-of-experiences, not in the meta-reality (and in the S-of-E planes do indeed fly).
Well maybe, but it is a fact that the S-of-E supplied by the Evil Genius does indeed contain regularities, and thus we do (as a matter of fact) have predictive power about the S-of-E.
Agreed, and this is the central point of my post. It gives a rationale for rejecting the superfluous BIV “wrapper”.
Agreed.
The BIV can only have the same predictive power about the S-of-E if you hypothesize that the Evil Genius chooses to feed us with a consistent “standard scientific real world” model (and thus the BIV model adopts the predictive power of our science). At that point parsimony cuts in. The two together are required for the argument.
We’re mostly in agreement, in that we both appeal to parsimony to dispense with the BIV wrapper. I’m then going one step further and trying to justify parsimony in terms of a probabilistic argument. That’s the big difference between us, since you don’t regard my argument as valid. However, from what you’ve said here I don’t see how you are countering my argument.
Essentially the BIV wrapper is one of an infinite number of conceivable meta-realities for which there is zero evidence, since they are hypothesized as making no discernable difference to what we experience.
For any one of these meta-realities that is actually true (if any), I can list a very large number of alternative meta-realities (possibly an infinite number of alternatives) that are not true. Would you agree with that?
If so, then I list all the possible meta-realities, pick one at random and say, “that meta-reality is indeed true”. Would you agree that, more likely than not, that statement is false (which follows from the previous paragraph)?
Hi Coel,
Thanks for the response. I take your point that predictions by a BIV are valid within that imaginary context that the BIV is experiencing – though like the content of dreams that might well appear to be the case for irrational thoughts: a BIV may really think true=false under some conditions. Maybe our predictions make the Evil Genius giggle at the irrationality of them. But because that is indistinguishable to each of us, whether a real brain in a real world or a BIV, the predictability tells us nothing. And of course our reliance of the usefulness and predictability of science is heavily influenced by the shared nature of science. But there is no shared nature for the sole BIV. All this goes no way to assuring us we are not a BIV.
On to the use of probabilities…
Praying to Zeus gets results, sometimes.
Singing to Baal gets pineapples, always.
Jumping to Ra paints peanuts, occasionally.
Predicting from Newton’s laws gets results, mostly (not at high speeds, and not if making experimental errors)
…
The point is that from the range of infinite possibilities, if it were not that we have come to find Newton’s laws useful, they too could be labelled as highly improbable (if probability were appropriate). It’s only our experience that convinces us that Newton’s laws are reliable in certain ranges. We could say the same about praying to Jesus or claiming we’re a BIV – experience fits them just as well as Newton’s laws. Every time I claim that when we launch rockets what’s really happening is that the Evil Genius is making me think Newton’s laws hold. Every time I pray to Jesus, Newton’s laws hold when we launching rockets, it works. This is how I imagine many theistic scientists continue to believe in Jesus.
Try these:
1) I state that I have a bag of ten balls or 5cm diameter, equal in all respects, except 1 is white and the others are black. I ask you to give a a measure of the likelihood of picking the white one. From our statistically inductive experience of the world, so far, we’d both agree that there’s a 1:10 chance.
2) I state I have a bag of a billion balls, with one white and the rest black. We can still apply probability, because I’ve given data we can work with.
3) I state I have a bag of infinite capacity, that, if it contains anything at all, contains balls of equal size, mass, etc., but not necessarily colour or labels. I don’t know how many balls might be in the bag, or what colours they are. There might be zero, or an infinite number. We cannot say anything useful about the probability of picking out a white ball. The actual probability depends on data we don’t have: how many balls are in the bag, and the proportion that are white, and so on. The space of possibilities, reds, whites, striped, labelled, spotted, …, doesn’t tell us anything useful, and cannot be used to determine the likelihood of picking any colour ball out of the bag.
Case (3) is were we are with multiverses, variations on solipsism and BIV scenarios, gods and prophets, and all manner of other imagined things.
Suppose in (3): on several attempts we do actually pick out a white ball, and discard it each time. Occasionally we pick out and discard a black ball, and sometimes a red one. We are still in no position to say what the probability is of picking a white one next time.
We might pretend that we are confident that most swans, sorry, balls, are white. But that’s only our experience so far, dipping into the locality of the mouth of the bag. This is the sort of problem some physicists ponder about the localisation of our cosmological perspective.
But that’s not a problem. It would still be parsimonious to presume most balls are white, until such time we find that they are not. We might ponder the many possibilities, but that will not inform us of anything without more data. Parsimony is not shown to work by any further reasoning. It only appears to work, so far. And when it stops working in some specific case, we change our minds about what we conclude.
If we pondered how many balls in bag (3) are labelled Zeus, Baal, Ra, the correct answer is we don’t know. We really don’t know. We might intuitively say that they are all unlikely, given the infinity of possibilities. But possibilities are merely a numeric space from which data is logically possible, not an actual space of probabilities. It is incorrect to presume that there is some specific and actual distribution that makes any particular unevidenced possibility either likely or unlikely.
I think the confusion about this distinction between the space of logical possibilities and the space of possible actualities comes from our experience with simple small spaces. Coins and dice are already loaded such they have a small actual space of probabilities: 2, 6. Of course coins do sometimes land on edge, and dice do roll up against a wall and fail to land on one face, but we know why these other possibilities are unlikely. With BIV and other imagined spaces we don’t.
Again, thinking about cosmology, a lot is placed on the rather useful values (from our perspective) of fundamental constants. But theists (or physicists) that suggest they are unlikely in the infinite possibilities of cosmologies are making a specific claim that they cannot support. It may be that in the business of universe creation (by gods or by some wider scale natural process in some hyperverse) that all universes always have these constants, and the laws we find always work the way we find they do here. Or maybe there are just two stable states: just like ours, of completely flat an uninteresting. We don’t know anything about universe creation to be able to say anything as general as: our constants are unlikely or special.
In (3) we may be finding lots of white balls because there is only a localised supply that we have not exhausted, but compared to the finite content the picking of a white ball is infinitesimally small. Or maybe all the balls except a few are white, in which case picking white ones is highly likely. Or maybe all but those few within reach are labelled Jesus, just as the Church of Jesus’s Balls have been telling us, and only when we dive right into the bag and accept Jesus’s Balls will we discover the truth.
All we can say is that so far, from our experience, and using the principle of parsimony of not adding complexity when some simpler explanation will do, when the simpler explanation is a subset of the grander explanation, then we find this works well. We not only can’t appeal to probability of the unknown, but nor do we need to. For now, at least, we use parsimony as a heuristic, with no foundational proof of it, no appeal to probabilities. It is sufficient. And that it isn’t some more certain rule also allows us to ditch it when it’s not useful.
So I think my example (3) and my explanation of it does show that the use of probabilities in thought experiments about possibilities cannot apply.
Hi Ron,
Predictability about the stream-of-experiences is indeed important here, since any overall explanation needs to include an explanation of the S-of-E, and predictability about the S-of-E is the same as parsimony in that it allows a description with reduced information content.
Yes, exactly. Everything is a priori highly improbable. It is evidence that tells us which of the highly improbable things are actually a good match to reality/S-of-E. Being “useful” here is the same as there being evidence for them, since they are useful because they model reality/S-of-E.
But, when we have zero evidence, any statement plucked at random is highly unlikely to be true.
Sure, this is the possibility of the Evil Genius feeding us the “standard scientific real-world” model. And that is indeed the most parsimonious way of describing the S-of-E.
Agreed, case 3 is when we have no evidence at all about what there is. You seem to be now agreeing that any claim about what we will next pick out of the bag is very unlikely to be true (since we have no evidence to guide us to the claim). Agreed?
Here I don’t agree. We are indeed in no position to say *for* *sure* what we will next pick, but we do have some information about what is in the bag, since we have taken some samples from it. That does give us some information, though only limited information. For our next pick we would be justified in regarding “white” as more probable than “purple”.
Yes, and if you were offered a million dollars if you guessed the colour of the next one, you’d go for white, surely? On the information we have, it is more probable than some random other colour.
Except that here you have posited an infinite bag with a possibly infinite number of balls, which makes a difference (we might expect everything with infinite balls).
Let’s restrict to 10 balls, each of which has at most one label. Now ask the question, what is the probability that at least one ball is labelled Ra? Surely the answer is that it is very low, given the number of possible labels?
We can consider the set of all possible “specific and actual distributions”, and work out the probability from that super-set.
I’m baffled here. If you’re not going to justify parsimony in some fashion, on what basis are you asserting that it is “useful”?
Reblogged this on oogenhand and commented:
“Well, think about the consequences of perceiving the real world and being a BIV perceiving a stimulated fictional world.”
Consequences are the idea behind Bernstein’s Wager, superior to Pascal’s Wager.