Why do I believe that the table in front of me exists? Well, I have a sense perception of light reflected from the table entering my eyes, and I hypothesize the existence of the table as an explanation of what caused that perception.
Now suppose I have a perceptual intuition about the existence of, say, a hell world in which people who steal are punished. What do I conclude? I assume that the belief is most likely a figment of my imagination, invented by my brain as a result of various cultural images and ideas that have been developed over the ages.
There are important, practical differences that may justify the divergence here. For instance, nearly everyone's perception of the table in front of me agrees to a large extent, while intuitions about punishment in the hereafter diverge wildly, if they're present at all. But suppose it were otherwise; suppose 99.9% of humans shared an intuitive impression that, yes, people who steal will be punished in a hereafter of a particular sort. Does this make it justified to conclude that such a hereafter exists, and that its existence is impressed upon human minds by some physical mechanism like the photons which impress upon our eyes the existence of the table? What if we found a region of the brain that seemed to produce the hell-world intuitions? Would we take that to be the receptor of this other-worldly signal, in the same way that the eyes and vision-processing regions of the brain are taken to be the receptors of the table signals?
Perhaps it's more complicated to assume that another world exists and sends these signals than to assume that they're generated by our own minds. That may be. But why is it also not more complicated to assume that an external table exists than to assume that our brains are confabulating that sensory input as well? (I suppose this is the usual question, Why assume that I'm not dreaming?)
Perhaps we're in a simulation, and the table impression is being fed to me by the program that's running my brain. But then why not suppose that the hell intuition is also being fed to me by that program, rather than supposing that it's something my own brain generated on its own? In that case, why couldn't the hell world be, from the perspective of things-inside-the-simulation, as "real" as the table? My intuition would just be a warning from the simulator about what he has in store for my future. If we think a simulator exists for other reasons, then such a conclusion doesn't seem outlandish or terribly complicated.
I suppose I've essentially just recapitulated Alvin Plantinga's argument that certain religious beliefs can be properly basic, so I waive any claims to originality here. Unlike Plantinga, though, I would make no distinctions about what sorts of beliefs can be properly basic. There's no difference in principle between other-worldly intuitions and an overwhelming impression that, say, an invisible Great Pumpkin flies over the pumpkin patch each Halloween.
What exactly would a good Occam-abiding Bayesian say here? Is the distinction between table impressions transmitted through photons and hell impressions transmitted through some other physical means merely one of degree (e.g., wider agreement on what the table impressions tell us, better understanding of what photons are for other reasons) rather than kind? I would guess so, but perhaps there's another point I'm missing.
Maybe we need to make our beliefs pay rent by anticipating particular future outcomes. For afterlife questions, we have to wait a while to test some of the predictions, but perhaps there are other predictions that the hell-world hypothesis makes which could be examined more immediately. For instance, if we think the hell world will be run by our simulator, we might imagine that he would issue forewarnings about that world whenever we steal, by causing us to execute a cognitive algorithm for the sensation of guilt. I guess it's more simple to postulate that our own minds trigger the guilt algorithm on their own, without needing external activation. But then consider this analogy. When I see someone's fingers hit keys on a piano, I perceive a sensation of sound. My hypothesis to explain this is that the piano sends sound waves to my ears, which receive the signal and trigger cognitive operations that we refer to as "hearing the sound." But why not suppose instead that, whenever I see someone hit a key, my brain automatically triggers those "hear the sound" operations -- with appropriate confabulations to account for the fact that, say, I don't see every key that the piano player's fingers hit, or I may hear the music on the radio instead?
It seems the difference with the hell scenario comes principally from the fact that sound waves are a useful hypothesis in other contexts; for instance, their existence accounts for the results of certain physics experiments. Absent such data, belief in "sound waves" would, apparently, be just as absurd as belief in ghosts due to an overwhelming sense impression that they exist. (I wonder: Are there entities or phenomena in which most scientists do believe that have not been shown to be useful theoretical constructs in other ways? Skeptics claim that so-called supernatural beliefs are of this type, but are there others -- like sound waves or photons -- that similarly don't pass this test?)
On the other hand, what if we have other theoretical reasons for believing in the existence of a simulator with motives for punishing people and a desire to forewarn them about it. Then couldn't pangs of guilt and associated fear fairly naturally be interpreted as messages from the simulator? Of course, the "your brain did it on it's own" explanation is also strong, but perhaps not overwhelmingly so? Could responding directly to such warnings be a safer course of action than acting in response to more elaborate scenarios about other, stranger conditions for punishment that our simulators might impose?
Subscribe to:
Post Comments (Atom)
The reasoning here is illogical. I mean that in a formal sense: it violates Bayes' Theorem.
ReplyDeleteFirst, prior probability derives from actual facts, not conceivable fictions. The idea that UFO's are spacecraft flown by extraterrestrials is entirely explicable with known scientific understanding, yet the prior probability that any given UFO we see actually is a spacecraft flown by extraterrestrials is exceedingly small, because of past experience with UFO observations (in which every corroborated cause turned out to be something else, going on hundreds of cases to date, if not thousands). So, too, with all theories relying on causes never once ever observed before, e.g. this universe being manipulated by mad scientists running it as a simulation, or any other cosmic intelligences sending us signals (such as regarding imaginings of hell and feelings of guilt). Not that these things are impossible, but rather they have never once been corroborated as the cause of anything, whereas everything we have observed has had numerous other causes multiply confirmed for it. Thus, picking any particular set of observations, the prior probability that it's cause derives from a set of never-corroborated theoretical causes, rather than from a highly-confirmed set of known causes, is exceedingly small (and thus you need some exceedingly strong evidence to overcome those implications of prior experience).
Second, the observed phenomenon must be taken wholly into consideration--you can't arbitrarily exclude evidence. The explanatory power of competing theories is derived from the ability to explain all the relevant evidence, not isolated patches of it. Indeed, a theory that explains more evidence often by virtue of that very fact has greater explanatory power. In this case, the evidence includes conflicting intuitions regarding guilt held (a) by other people and (b) even yourself at different stages of your life. The ultra-punisher theory predicts consistent intuitions; the guilt-is-neurocultural theory predicts inconsistent intuitions; we observe the latter, not the former; therefore, the guilt-is-neurocultural theory has greater explanatory power, while the ultra-punisher theory has basically been disconfirmed. Likewise, the evidence includes guilt responses to highly specific cultural constructs (e.g. a Jew feeling hell-guilt about eating shrimp; a Muslim woman feeling hell-guilt about wearing pants), which are massively implausible concerns of an ultra-punisher (why would anyone care whether we eat shrimp or wear pants?), but entirely explicable as outcomes of neurocultural causes. Furthermore, most cultures in history had no conception of hell, and still today there are countless people who have no pertinent experiences or intuitions regarding it, and those who do have entirely contradictory ones (not being able to agree what sort of hell it is, who goes, and how long victims of it stay there), which is again predicted by neurocultural theory, and contra-predicted by ultra-punisher theory.
Thus, ultra-punisher theory fails every inductive logical test we have, whereas neurocultural theory is a powerful, highly-confirmed explanation of all the relevant observations.
Thanks for the comments, Richard!
ReplyDeleteFirst, prior probability derives from actual facts, not conceivable fictions.
When we have data, then yes, we incorporate frequency information into our prior to get a new prior. But what happens when we have no data? What prior do we use with zero previous confirmed observations either way?
In the case of whether the table exists, for instance, we don't know one way or the other about previous instances. We can't tell whether there have existed tables before, or what fraction of tables are mind-confabulations vs. physical objects, because such knowledge begs the question. Similarly, we've only seen this universe -- we don't have data on what fraction of universes actually are simulations vs. "basement-level" realities. We can make guesses based on other arguments, but we have no hard frequency data to go on.
Thus, picking any particular set of observations, the prior probability that it's cause derives from a set of never-corroborated theoretical causes, rather than from a highly-confirmed set of known causes, is exceedingly small
Yes, this is my point that, say, light waves have better support than hell worlds because we have a wide variety of observations that are all explained by the existence of photons (including many physics experiments).
I agree with almost everything you say in the second main paragraph. I would just point out that it needn't necessarily be the case that revelations about punishment be consistent -- maybe the simulator just wants to have fun, and so intends to punish certain people for certain things and other people for others. (Of course, with a simulator just having fun, it's hard to tell whether following the commands or disobeying them is the safer option; a masochist simulator might really enjoy punishing those who expected to avoid hell for responding to feelings of guilt.)
Alan Dawrst said... But what happens when we have no data? What prior do we use with zero previous confirmed observations either way?
ReplyDeleteBut we do have data. Tons of it. Just as with UFOs. The fact that we've never seen an actual alien spaceship does not constitute having zero data.
First, as I said, we have vast data confirming naturalism and disconfirming bizarro hypotheses of the sort you imagine. Thus, for example, we may never have seen a thriving faerie people made of paper, but we've seen tons of data that render the hypothesis of their existence extremely small. In other words, the data we do have makes that kind of hypothesis very improbable. Thus, it is not necessary to have "seen" the absence of a hell to conclude that we have massive data against it. Likewise superbeings manipulating our minds.
Second, as I also said, the hypotheses you propose entail predictions, e.g. that everyone would have consistent intuitions about hell, which are falsified by abundant data, whereas alternative hypotheses, e.g. the neurocultural explanation, make several predictions that are well confirmed in the evidence. Thus, even if the prior probability were a perfect 50/50 (as it would be in a case of having absolutely no pertinent prior knowledge, even though, as just noted above, this isn't such a case), the epistemic probability would still be near zero, because the balance of posterior probabilities is heavily weighted against your theory.
Alan Dawrst said... In the case of whether the table exists, for instance, we don't know one way or the other about previous instances. We can't tell whether there have existed tables before, or what fraction of tables are mind-confabulations vs. physical objects, because such knowledge begs the question.
ReplyDeleteNo it doesn't. Because the theory of tables explains all the evidence perfectly, whereas your theories do not. That's fatal to your theories, not to table realism. No question is begged, because the hypothesis adduced has passed every test and is less complex than every alternative explanation of the same results (it thus survives Occham's Razor).
If you want to propose a Cartesian Demon scenario, then you can get the same predictions, but that contradicts your thesis. For a supermind who wanted us to know hell would not then hide that fact from us, thus such a supermind cannot behave like a Cartesian Demon. And even if you convoluted your way into a thesis that somehow unites Cartesian Demon behavior with active worldwide hell-intuition-production, and further explained all the contradicting evidence (like the inconsistency of hell intuitions worldwide), the result would be a theory so complicated, that by virtue of the logic of set theory alone it would have a very small prior probability.
That is, every ad hoc thing you add to "force" your theory to fit the evidence that otherwise would flatly contradict it (or offer no particular support to it), makes that theory inherently less probable than a theory without that feature. Thus, prior probability falls with every attempt to escape the fact that the posterior probabilities refute you. The net result is exactly the same--unless and until you have actual evidence supporting your theory, which does not also support the contrary theory exactly as well. And you have none.
Similarly, we've only seen this universe -- we don't have data on what fraction of universes actually are simulations vs. "basement-level" realities. We can make guesses based on other arguments, but we have no hard frequency data to go on.
This has no bearing on your argument. Even if we're in a sim, your theory of hell-intuition-creation still fails on the exact same Bayesian logic.
Of course, we have no reason to believe sims like our world are even probable. But I won't go into Bostrom's fallacies here, since it doesn't matter.
I agree with almost everything you say in the second main paragraph. I would just point out that it needn't necessarily be the case that revelations about punishment be consistent.
But that's an irrelevant observation. The only pertinent observation is that it would probably be the case--in fact, to an extremely high probability, unless you "doctor" your thesis with a big set of ad hoc assumptions, every one of which halves the prior probability of your original theory. You can't bootstrap your way out of this.
Maybe the simulator just wants to have fun...
Maybe the universe is a McDonald's toy. Maybe invisible dinosaurs live under your bed. Maybe the moon is made of cheese. "Maybe" isn't worth jack. You have to ask what's probable. Divide a pie into even slices for every possible explanation of the facts, and one of those slices will be your theory of a supermind sending hell intuitions. Already, we're talking about a tiny, tiny slice, not even half the pie, because there are so many other equally possible explanations of the same evidence.
Now if you add ad hoc bullshit like "it just wants to have fun by pretending to send random people the wrong intuitions" then you have to cut that one tiny slice even more. Thus, no matter how you turn, you just make the theory less probable, not more. You can only escape this result by having evidence for your theory. No theory can be made more probable by just stacking more maybes onto it.
First, as I said, we have vast data confirming naturalism and disconfirming bizarro hypotheses of the sort you imagine. [...] Thus, it is not necessary to have "seen" the absence of a hell to conclude that we have massive data against it.
ReplyDeleteI don't lump explanations into either a "natural" bucket or a "supernatural" bucket. If an explanation provides a mechanism, it's a "natural" explanation. (I don't know what "supernatural" could even mean.) One explanation of why I see a table is that photons brought light from it to my eyes. One explanation of why my brain runs an algorithm corresponding to guilt when I do particular things is that a simulator set up the parameters of his program with a rule triggering the guilt algorithm in those cases for the purpose of warning me about punishment. Both are entirely sensible, physical explanations -- fairies have nothing to do with it. The difference comes from Occam's razor: Postulating a simulator with those motivations is a hugely complicated mechanism for such a trivial observation. But Occam's razor is a principle, not data. Maybe your point was that we have lots of experience with the fact that assuming Occam's razor tends to work pretty well? In that case, I agree.
Because the theory of tables explains all the evidence perfectly, whereas your theories do not. That's fatal to your theories, not to table realism.
Exactly -- I agree. All I'm saying is that we have no "count" data like
(# of times tables I've seen have been real) = 100
(# of times tables I've seen have been confabulated) = 0.
Rather, we have 100 observations of things that look like tables, all of which are more elegantly explained on the assumption that tables exist and interact with the world according to particular, simple physical principles. This is basically the argument against our being Boltzmann brains, actually.
Thus, prior probability falls with every attempt to escape the fact that the posterior probabilities refute you.
Exactly. I agree with your entire discussion here. I think our disagreements are just terminological about what counts as "prior probability" vs. "data."
This has no bearing on your argument. Even if we're in a sim, your theory of hell-intuition-creation still fails on the exact same Bayesian logic.
Yes, exactly. Evidence about contradictory intuitions greatly diminishes the probability of the sim-intuition hypothesis. My point was merely that we may have other reasons to think we're sims, in which case that particular portion of the hypothesis (the existence of a simulator) may not have vanishingly small probability.
The only pertinent observation is that it would probably be the case
Indeed. I was only observing that the probability doesn't equal 100%.
not even half the pie, because there are so many other equally possible explanations of the same evidence.
No, of course not. I don't claim the sim-intuition hypothesis is very likely at all. But very improbable things can still worry me.
I have a new blog on food and ethnics and I was doing reseach on suffering, trying to answer the question "how do you know a chick suffers?. Now, I guess I won't do anything else today than reading your papers.
ReplyDeleteThanks.
Élise
Hi Élise -- thanks for the question. There are a number of good articles that outline the evidence for pain in birds like chickens -- perhaps the bibliography of the Wikipedia page on the subject would be a good place to start. I personally first realized that animals like chickens could feel pain by reading this passage from Peter Singer's Animal Liberation. I think one of the strongest arguments is the evolutionary and physiological one noted in the introductory paragraph to Jane A. Smith's "A Question of Pain in Invertebrates":
ReplyDelete"Quite apart from philosophical considerations, practical and scientific evidence may lead us to assume that all mammals can experience something analogous to (though most likely qualitatively and quantitatively different from) the human experience of pain. Humans, after' all, are mammals; and although the details may differ, we share our basic physiology with other mammalian species. There is also a reasonableness, it seems, in extending this view to include other members of the Vertebrata. The further we move away from the mammalian plan, the more difficult it becomes to infer pain in other species. But vertebrates, at least, have similarities in basic anatomy and physiology, including similarities in nervous organization [...]."
In any event, I wish you luck with the blog posts! It's great that you're tackling these important topics.
Alan, going a little off-topic for this post, I'm keen for your feedback on a question which is challenging my ethical intuition. You put a compelling case for caring about wild animal suffering. Let's assume for the sake of argument that such suffering far outweighs the net happiness of all sentient beings on earth (incidentally another question I'm interested in your take on). If that is truly the case, might a "doomsday scenario" (say a nuclear winter leading to the death of 80% of sentient beings) be the most promising route for increasing utility on earth for the next several thousand years? Obviously this is a repugnant outcome, but I'm interested to know why.
ReplyDeleteJust to clarify my question above. The reason I ask is that I am new to utilitarian thinking -- and I'm keen to know how one approaches a paradox like this, where seeking to reduce suffering leads to a ridiculous outcome that in fact massively increases it.
ReplyDeleteHi Jay, Thanks for the question! It's great to think about cases that challenge our intuition.
ReplyDeleteI'm uncertain as to whether the net balance of suffering in nature is positive or negative. I do think there are strong arguments to be made in favor of the negative side, notably the fact that most species give birth to far more offspring than survive past the early stages of life. (Humans are abnormal in this respect. And even human infant mortality during, say, the middle ages wasn't trivial.) However, I remain open to evidence on both sides.
Suppose the net balance of suffering vs. pleasure is negative. Does this mean doomsday would be best? I actually happen to think the answer is "no," because I'm hopeful that humans, through technological advances in the future, may find ways to significantly reduce suffering elsewhere in the cosmos. I think human-type empathy is probably rather rare in the universe, so that it would be a shame for humans to go extinct before having the chance potentially to prevent wild-animal-type suffering on a wider scale. As it turns out, the Singularity Institute for Artificial Intelligence is currently working on some papers about this very topic, so keep on the lookout for them in the coming months. (I may post a link on this blog when they're ready.)
I should add that even if the net balance of suffering on earth over the next few million years is negative, this might be outweighed by the vast numbers of potentially very happy post-human life-years that would be prevented by doomsday. I'm personally not a complete negative utilitarian, so this point does matter to me, even though I do care a lot about preventing suffering.