It is not really a quote, but a good quip from an otherwise lame recent internet discussion:
Matt:
Ok, for all of the people responding above who admit to not having a soul, I think this means that it is morally ok for me to do anything I want to you, just as it is morally ok for me to turn off my computer at the end of the day. Some of us do have souls, though.
Igor:
Matt—I agree that people who need a belief in souls to understand the difference between killing a person and turning off a computer should just continue to believe in souls.
Of course, my original comment had nothing to do with god. It had to do with “souls”, for lack of a better term as that was the term that was used in the original discussion (suggest reading the original post if you want to know more—basically, as I understand the intent it simply referred to some hypothetical quality that is associated with consciousness that lies outside the realm of what is simulable on a Turing machine). If you think that humans are nothing but Turing machines, why is it morally wrong to kill a person but not morally wrong to turn off a computer? Please give a real answer...either provide an answer that admits that humans cannot be simulated by Turing machines, or else give your answer using only concepts relevant to Turing machines (don’t talk about consciousness, qualia, hopes, whatever, unless you can precisely quantify those concepts in the language of Turing machines). And in the second case, your answer should allow me to determine where the moral balance between human and computers lies....would it be morally bad to turn off a primitive AI, for example, with intelligence at the level of a mouse?
If you think that humans are nothing but Turing machines, why is it morally wrong to kill a person but not morally wrong to turn off a computer?
Your question has the form:
If A is nothing but B, then why is it X to do Y to A but not to do Y to C which is also nothing but B?
This following question also has this form:
If apple pie is nothing but atoms, why is it safe to eat apple pie but not to eat napalm which is also nothing but atoms?
And here’s the general answer to that question: the molecules which make up apple pie are safe to eat, and the molecules which make up napalm are unsafe to eat. This is possible because these are not the same molecules.
Now let’s turn to your own question and give a general answer to it: it is morally wrong to shut off the program which makes up a human, but not morally wrong to shut off the programs which are found in an actual computer today. This is possible because these are not the same programs.
At this point I’m sure you will want to ask: what is so special about the program which makes up a human, that it would be morally wrong to shut off the program? And I have no answer for that. Similarly, I couldn’t answer you if you asked me why the molecules of apple pie are safe to eat and the those of napalm are not.
As it happens, chemistry and biology have probably advanced to the point at which the question about apple pie can be answered. However, the study of mind/brain is still in its infancy, and as far as I know, we have not advanced to the equivalent point. But this doesn’t mean that there isn’t an answer.
Because people are really annoying, but we need to be able to live with each other.
We need strong inhibitions against killing each other—there are exceptions (self-defense, war), but it’s a big win if we can pretty much trust each other not to be deadly.
We’d be a lot more cautious about turning off computers if they could turn us off in response.
None of this is to deny that turning off a computer is temporary and turning off a human isn’t. Note that people are more inhibited about destroying computers (though much less so than about killing people) than they are about turning computers off.
if i believed when i turned off my computer it would need to be monitered by a specialist or it might not ever come back on again, i would be hesitant to turn it off as well
And indeed, mainframes & supercomputers are famous for never shutting down or doing so on timespans measured in decades and with intense supervision on the rare occasion that they do.
It certainly doesn’t put a halt to brain activity. You might not be aware of anything that’s going on while you’re under, or remember anything afterwards (although some people do,) but that doesn’t mean that your brain isn’t doing anything. If you put someone under general anesthetic under an electroencephalogram, you’d register plenty of activity.
Hmm, I don’t happen to find your argument very convincing. I mean, what it does is to pay attention to some aspect of the original mistaken statement, then find another instance sharing that aspect which is transparently ridiculous.
But is this sufficient? You can model the statement “apples and oranges are good fruits” in predicate logic as “for all x, Apple(x) or Orange(x) implies Good(x)” or in propositional logic as “A and O” or even just “Z”. But it should really depend on what aspect of the original statement you want to get at. You want a model which captures precisely those aspects you want to work with.
So your various variables actually confused the hell outta me there. I was trying to match them up with the original statement and your reductio example. All the while not really understanding which was relevant to the confusion. It wasn’t a pleasant experience :(
It seems to me much simpler to simply answer: “Turing machine-ness has no bearing on moral worth”. This I think gets straight to the heart of the matter, and isolates clearly the confusion in the original statement.
Or further guess at the source of the confusion, the person was trying to think along the lines of: “Turing machines, hmm, they look like machines to me, so all Turing machines are just machines, like a sewing machine, or my watch. Hmm, so humans are Turing machines, but by my previous reasoning this implies humans are machines. And hmm, furthermore, machines don’t have moral worth… So humans don’t have moral worth! OH NOES!!!”
Your argument seems like one of those long math proofs which I can follow step by step but cannot grasp its overall structure or strategy. Needless to say, such proofs aren’t usually very intuitively convincing.
(but I could be generalizing from one example here)
No, I was not trying to think along those lines. I must say, I worried in advance that discussing philosophy with people here would be fruitless, but I was lured over by a link, and it seems worse than I feared. In case it isn’t clear, I’m perfectly aware what a Turing machine is; incidentally, while I’m not a computer scientist, I am a professional mathematical physicist with a strong interest in computation, so I’m not sitting around saying “OH NOES” while being ignorant of the terms I’m using. I’m trying to highlight one aspect of an issue that appears in many cases: if consciousness (meaning whatever we mean when we say that humans have consciousness) is possible for Turing machines, what are the implications if we do any of the obvious things? (replaying, turning off, etc...) I haven’t yet seen any reasonable answer, other than 1) this is too hard for us to work out, but someday perhaps we will understand it (the original answer, and I think a good one in its acknowledgment of ignorance, always a valid answer and a good guide that someone might have thought about things) and 2) some pointless and wrong mocking (your answer, and I think a bad one). edit to add: forgot, of course, to put my current guess as to most likely answer, 3) that consciousness isn’t possible for Turing machines.
if consciousness (meaning whatever we mean when we say that humans have consciousness) is possible for Turing machines,
This is the part where you’re going astray, actually. We have no reason to think that human beings are NOT Turing-computable. In other words, human beings almost certainly are Turing machines.
Therefore, consciousness—whatever we mean when we say that—is indeed possible for Turing machines.
To refute this proposition, you’d need to present evidence of a human being performing an operation that can’t be done by a Turing machine.
Understanding this will help “dissolve” or “un-ask” your question, by removing the incorrect premise (that humans are not Turing machines) that leads you to ask your question.
That is, if you already know that humans are a subset of Turing machines, then it makes no sense to ask what morally justifies treating them differently than the superset, or to try to use this question as a way to justify taking them out of the larger set.
IOW, (the set of humans) is a subset of (the set of turing machines implementing consciousness), which in turn is a proper subset of (the set of turing machines). Obviously, there’s a moral issue where the first two subsets are concerned, but not for (the set of turing machines not implementing consciousness).
In addition, there may be some issues as to when and how you’re doing the turning off, whether they’ll be turned back on, whether consent is involved, etc… but the larger set of “turing machines” is obviously not relevant.
I hope that you actually wanted an answer to your question; if so, this is it.
(In the event you wish to argue for another answer being likely, you’ll need to start with some hard evidence that human behavior is NOT being Turing-computable… and that is a tough road to climb. Essentially, you’re going to end up in zombie country.)
You wrote:
“This is the part where you’re going astray, actually. We have no reason to think that human beings are NOT Turing-computable. In other words, human beings almost certainly are Turing machines.”
at this stage, you’ve just assumed the conclusion. you’ve just assumed what you want to prove.
“Therefore, consciousness—whatever we mean when we say that—is indeed possible for Turing machines.”
having assumed that A is true, it is easy to prove that A is true. You haven’t given an argument.
“To refute this proposition, you’d need to present evidence of a human being performing an operation that can’t be done by a Turing machine.”
It’s not my job to refute the proposition. Currently, as far as I can tell, the question is open. If I did refute it, then my (and several other people’s) conjecture would be proven. But if I don’t refute it, that doesn’t mean your proposition is true, it just means that it hasn’t yet been proven false. Those are quite different things, you know.
Well, how about this: physics as we know it can be approximated arbitrarily closely by a computable algorithm (and possibly computed directly as well, although I’m less sure about that. Certainly all calculations we can do involving manipulation of symbols are computable). Physics as we know it also seems to be correct to extremely precise degrees anywhere apart from inside a black hole.
Brains are physical things. Now when we consider that thermal noise should have more of an influence than the slight inaccuracy in any computation, what are the chances a brain does anything non-computable that could have any relevance to consciousness? I don’t expect to see black holes inside brains, at least.
In any case, your original question was about the moral worth of turing machines, was it not? We can’t use “turing machines can’t be conscious” as excuse not to worry about those moral questions, because we aren’t sure whether turing machines can be conscious. “It doesn’t feel like they should be” isn’t really a strong enough argument to justify doing something that would result in, for example, the torture of conscious entities if we were incorrect.
So here’s my actual answer to your question: as a rule of thumb, act as if any simulation of “sufficient fidelity” is as real as you or I (well, multiplied by your probability that such a simulation would be conscious, maybe 0.5, for expected utilities). This means no killing, no torture, etc.
’Course, this shouldn’t be a practical problem for a while yet, and we may have learned more by the time we’re creating simulations of “sufficient fidelity”.
at this stage, you’ve just assumed the conclusion. you’ve just assumed what you want to prove.
No—what I’m pointing out is that the question “what are the ethical implications for turing machines” is the same question as “what are the ethical implications for human beings” in that case.
It’s not my job to refute the proposition. Currently, as far as I can tell, the question is open.
Not on Less Wrong, it isn’t. But I think I may have misunderstood your situation as being one of somebody coming to Less Wrong to learn about rationality of the “Extreme Bayesian” variety; if you just dropped in here to debate the consciousness question, you probably won’t find the experience much fun. ;-)
I did refute it, then my (and several other people’s) conjecture would be proven. But if I don’t refute it, that doesn’t mean your proposition is true, it just means that it hasn’t yet been proven false. Those are quite different things, you know.
Less Wrong has different—and far stricter—rules of evidence than just about any other venue for such a discussion.
In particular, to meaningfully partake in this discussion, the minimum requirement is to understand the Mind Projection Fallacy at an intuitive level, or else you’ll just be arguing about your own intuitions… and everybody will just tune you out.
Without that understanding, you’re in exactly the same place as a creationist wandering into an evolutionary biology forum, without understanding what “theory” and “evidence” mean, and expecting everyone to disprove creationism without making you read any introductory material on the subject.
In this case, the introductory material is the Sequences—especially the ones that debunk supernaturalism, zombies, definitional arguments, and the mind projection fallacy.
When you’ve absorbed those concepts, you’ll understand why the things you’re saying are open questions are not even real questions to begin with, let alone propositions to be proved or disproved! (They’re actually on a par with creationists’ notions of “missing links”—a confusion about language and categories, rather than an argument about reality.)
I only replied to you because I though perhaps you had read the Sequences (or some portion thereof) and had overlooked their application in this context (something many people do for a while until it clicks that, oh yeah, rationality applies to everything).
So, at this point I’ll bow out, as there is little to be gained by discussing something when we can’t even be sure we agree on the proper usage of words.
“at this stage, you’ve just assumed the conclusion. you’ve just assumed what you want to prove.
No—what I’m pointing out is that the question “what are the ethical implications for turing machines” is the same question as “what are the ethical implications for human beings” in that case.”
Yeah, look, I’m not stupid. If someone assumes A and then actually bothers to write out the modus ponens A->B (when A->B is an obvious statement) so therefore B, and then wants to point out, ‘hey look, I didn’t assume B, I proved it!’, that really doesn’t mean that they proved anything deep. They still just assumed the conclusion they want since they assumed a statement that trivially implies their desired conclusion. But I’ll bow out now too...I only followed a link from a different forum, and indeed my fears were confirmed that this is a group of people who don’t have anything meaningful or rational to say about certain concepts (I mean, if you don’t realize even that certain things are in principle open to physical test!---and you drew an analogy to creationism vs evolution without realizing that evolution had and has many positive pieces of observable, physical evidence in its favor while your position has at present at best very minimal observable, tangible evidence in its favor (certain recent experiments in neuroscience can be charitably interpreted in favor of your argument, but on their own they are certainly not enough)).
If you’re looking for a clear, coherent and true explanation of consciousness, you aren’t going to find that anywhere today, especially not in off-the-cuff replies; and if someone does eventually figure it out, you ought to expect it to have a book’s worth of prerequisites and not be something that can be explained in a thousand words of comment replies. Consciousness is an extraordinarily difficult and confusing topic, and, generally speaking, the only way people come up with explanations that seem simple is by making simplifying assumptions that are wrong.
As for the more specific question of whether humans are Turing computable, this follows if (a) the laws of physics in a finite volume are Turing computable, and (b) human minds run entirely on physics. Both of these are believed to be true - (a) based on what we know about the laws themselves, and (b) based on neuroscience, which shows that physics are necessary and sufficient to produce humans, combined with Occam’s Razor, which says that we shouldn’t posit anything extra that we don’t need. If you’d like to zoom in on the explanation of one of these points, I’d be happy to provide references.
Please read and understand the plethora of material that has been linked for you. This community does not dwell on solved problems of philosophy, and many have already taken a great deal of time to provide you with the relevant information.
To refute this proposition, you’d need to present evidence of a human being performing an operation that can’t be done by a Turing machine.
That’s quite easy: I can lift a rock, a Turing machine can’t. A Turing machine can only manipulate symbols on a strip of tape, it can’t do anything else that’s physical.
Your claim that consciousness (whatever we mean when we say that) is possible for Turing machines, rests on the assumption that consciousness is about computation alone, not about computation+some unidentified physical reaction that’s absent to pure Turing machines resting in a box on a table.
That consciousness is about computation alone may indeed end up true, but it’s as yet unproven.
That’s quite easy: I can lift a rock, a Turing machine can’t.
That sounds like a parody of bad anti-computationalist arguments. To see what’s wrong with it, consider the response: “Actually you can’t lift a rock either! All you can do is send signals down your spinal column.”
That consciousness is about computation alone may indeed end up true, but it’s as yet unproven.
What sort of evidence would persuade you one way or the other?
Could you clarify why you think that this reading assignment illuminates the question being discussed? I just reread it. For the most part, it is an argument against dualism. It argues that consciousness is (almost certainly) reducible to a physical process.
But this doesn’t have anything to do with what ArisKatsaris wrote. He was questioning whether consciousness can be reduced to a purely computational process (without “some unidentified physical reaction that’s absent to pure Turing machines”.)
Consider the following argument sketch:
Consciousness can be reduced to a physical process.
Any physical process can be abstracted as a computation.
Any computation can be modeled as a Turing Machine computation.
Therefore, consciousness can be produced on a TM.
Each step above is at least somewhat problematic. Matt1 seemed to be arguing against step 1, and Drescher does respond to that. But ArisKatsaris seemed to be arguing against step 2. My choice would be to expand the definition of ‘computation’ slightly to include the interactive, asynchronous, and analog, so that I accept step 2 but deny step 3. Over the past decade, Wegner and Goldin have published many papers arguing that computation != TM.
It may well be that you can only get consciousness if you have a non-TM computation (mind) embedded in a system of sensors and actuators (body) which itself interacts with and is embedded in within a (simulated?) real-time environment. That is, when you abstract the real-time interaction away, leaving only a TM computation, you have abstracted away an essential ingredient of consciousness.
For the most part, it is an argument against dualism. It argues that consciousness is (almost certainly) reducible to a physical process.
It actually sketches what consciousness is and how it works, from which you can see how we could implement something like that as an abstract algorithm.
The value of that description is not so much in reaching a certain conclusion, but in reaching a sense of what exactly are we talking about and consequently why the question of whether “we can implement consciousness as an abstract algorithm” is uninteresting, since at that point you know more about the phenomenon than the words forming the question can access (similarly to how the question of whether crocodile is a reptile is uninteresting, once you know everything you need about crocodiles).
The problem here, I think, is that “consciousness” doesn’t get unpacked, and so most of the argument is on the level of connotations. The value of understanding the actual details behind the word, even if just a little bit, is in breaking this predicament.
A TM computation perfectly modelling a human brain (let’s say) but without any real-time interaction, and a GLUT, represent the two ways in which we can have one of ‘intelligent input-output’ and ‘functional organization isomorphic to that of an intelligent person’ without the other.
What people think they mean by ‘consciousness’ - a kind of ‘inner light’ which is either present or not—doesn’t (straightforwardly) correspond to anything that objectively exists. When we hunt around for objective properties that correlate with places where we think the ‘inner light’ is shining, we find that there’s more than one candidate. Both ‘intelligent input-output’ and the ‘intelligent functional organization’ pick out exactly those beings we believe to be conscious—our fellow humans foremost among them. But in the marginal cases where we have one but not the other, I don’t think there is a ‘further fact’ about whether ‘real consciousness’ is present.
However, we do face the ‘further question’ of how much moral value to assign in the marginal cases—should we feel guilty about switching off a simulation that no-one is looking at? Should we value a GLUT as an ‘end in itself’ rather than simply a means to our ends? (The latter question isn’t so important given that GLUTs can’t exist in practice.)
I wonder if our intuition that the physical facts underdetermine the answers to the moral questions is in some way responsible for the intuition of a mysterious non-physical ‘extra fact’ of whether so-and-so is conscious. Perhaps not, but there’s definitely a connection.
… we do face the ‘further question’ of how much moral value to assign …
Yes, and I did not even attempt to address that ‘further question’ because it seems to me that that question is at least an order of magnitude more confused than the relatively simple question about consciousness.
But, if I were to attempt to address it, I would begin with the lesson from Econ 101 that dissolves the question “What is the value of item X?”. The dissolution begins by requesting the clarifications “Value to whom?” and “Valuable in what context?” So, armed with this analogy, I would ask some questions:
Moral value to whom? Moral value in what context?
If I came to believe that the people around me were p-zombies, would that opinion change my moral obligations toward them? If you shared my belief, would that change your answer to the previous question?
Believed to be conscious by whom? Believed to be conscious in what context? Is it possible that a program object could be conscious in some simulated universe, using some kind of simulated time, but would not be conscious in the real universe in real time?
It may be a step toward dissolving the problem. It suggests the questions:
Is it possible that an intelligent software object (like those in this novella by Ted Chiang) which exists within our space-time and can interact with us might have moral value very different from simulated intelligences in a simulated universe with which we cannot interact in real time?
Is it possible for an AI of the Chiang variety to act ‘immorally’ toward us? Toward each other? If so, what “makes” that action immoral?
What about AIs of the second sort? Clearly they cannot act immorally toward us, since they don’t interact with us. But can they act immorally toward each other? What is it about that action that ‘makes’ it immoral?
My own opinion, which I won’t try to convince you of, is that there is no moral significance without interaction and reciprocity (in a fairly broad sense of those two words.)
What people think they mean by ‘consciousness’ - a kind of ‘inner light’ which is either present or not—doesn’t (straightforwardly) correspond to anything that objectively exists.
It does, to some extent. There is a simple description that moves the discussion further. Namely, consciousness is a sensory modality that observes its own operation, and as a result it also observes itself observing its own operation, and so on; as well as observing external input, observing itself observing external input, and so on; and observing itself determining external output, etc.
It does, to some extent. There is a simple description that moves the discussion further. Namely, consciousness is a sensory modality that observes its own operation, and as a result it also observes itself observing its own operation, and so on; as well as observing external input, observing itself observing external input, and so on; and observing itself determining external output, etc.
This is an important idea, but I don’t think it can rescue the everyday intuition of the “inner light”.
I can readily imagine an instantiation of your sort of “consciousness” in a simple AI program of the kind we can already write. No doubt it would be an interesting project, but mere self-representation (even recursive self-representation) wouldn’t convince us that there’s “something it’s like” to be the AI. (Assume that the representations are fairly simple, and the AI is manipulating them in some fairly trivial way.)
Conversely, we think that very young children and animals are conscious in the “inner light” sense, even though we tend not to think of them as “recursively observing themselves”. (I have no idea whether and in what sense they actually do. I also don’t think “X is conscious” is unambiguously true or false in these cases.)
That’s quite easy: I can lift a rock, a Turing machine can’t. A Turing machine can only
manipulate symbols on a strip of tape, it can’t do anything else that’s physical.
I think you’re trivializing the issue. A Turing machine is an abstraction, it isn’t a real thing. The claim that a human being is a Turing machine means that, in the abstract, a certain aspect of human beings can be modeled as a Turing machine. Conceptually, it might be the case, for instance, that the universe itself can be modeled as a Turing machine, in which case it is true that a Turing machine can lift a rock.
I can lift a rock, a Turing machine can’t. A Turing machine can only manipulate symbols on a strip of tape, it can’t do anything else that’s physical.
So… you support euthanasia for quadriplegics, then, or anyone else who can’t pick up a rock? Or people who are so crippled they can only communicate by reading and writing braille on a tape, and rely on other human beings to feed them and take care of them?
Your claim that consciousness (whatever we mean when we say that) is possible for Turing machines, rests on the assumption that consciousness is about computation alone, not about computation+some unidentified physical reaction that’s absent to pure Turing machines resting in a box on a table.
This “unidentified physical reaction” would also need to not be turing-computable to have any relevance. Otherwise, you’re just putting forth another zombie-world argument.
At this point, we have no empirical reason to think that this unidentified mysterious something has any existence at all, outside of a mere intuitive feeling that it “must” be so.
And so, all we have are thought experiments that rest on using slippery word definitions to hide where the questions are being begged, presented as intellectual justification for these vague intuitions… like arguments for why the world must be flat or the sun must go around the earth, because it so strongly looks and feels that way.
(IOW, people try to prove that their intuitions or opinions must have some sort of physical form, because those intuitions “feel real”. The error arises from concluding that the physical manifestation must therefore exist “out there” in the world, rather than in their own brains.)
This “unidentified physical reaction” would also need to not be turing-computable to have any relevance. Otherwise, you’re just putting forth another zombie-world argument.
A zombie-world seems extremely improbable to have evolved naturally, (evolved creatures coincidentally speaking about their consciousness without actually being conscious), but I don’t see why a zombie-world couldn’t be simulated by a programmer who studied how to compute the effects of consciousness, without actually needing to have the phenomenon of consciousness itself.
The same way you don’t need to have an actual solar system inside your computer, in order to compute the orbits of the planets—but it’d be very unlikely to have accidentally computed them correctly if you hadn’t studied the actual solar system.
At this point, we have no empirical reason to think that this unidentified mysterious something has any existence at all, outside of a mere intuitive feeling that it “must” be so.
Do you have any empirical reason to think that consciousness is about computation alone? To claim Occam’s razor on this is far from obvious, as the only examples of consciousness (or talking about consciousness) currently concern a certain species of evolved primate with a complex brain, and some trillions of neurons, all of which have have chemical and electrical effects, they aren’t just doing computations on an abstract mathematical universe sans context.
Unless you assume the whole universe is pure mathematics, so there’s no difference between the simulation of a thing and the thing itself. Which means there’s no difference between the mathematical model of a thing and the thing itself. Which means the map is the territory. Which means Tegmark IV.
And Tegmark IV is likewise just a possibility, not a proven thing.
A zombie-world seems extremely improbable to have evolved naturally, (evolved creatures coincidentally speaking about their consciousness without actually being conscious), but I don’t see why a zombie-world couldn’t be simulated by a programmer who studied how to compute the effects of consciousness, without actually needing to have the phenomenon of consciousness itself.
This is a “does the tree make a sound if there’s no-one there to hear it?” argument.
That is, it assumes that there is a difference between “effects of consciousness” and “consciousness itself”—in the same way that a connection is implied between “hearing” and “sound”.
That is, the argument hinges on the definition of the word whose definition is being questioned, and is an excellent example of intuitions feeling real.
That is, it assumes that there is a difference between “effects of consciousness” and “consciousness itself”—in the same way that a connection is implied between “hearing” and “sound”.
Not quite. What I’m saying is there might be a difference between the computation of a thing and the thing itself. It’s basically an argument against the inevitability of Tegmark IV.
A Turing machine can certainly compute everything there is to know about lifting rocks and their effects—but it still can’t lift a rock.
Likewise a Turing machine could perhaps compute everything there was to know about consciousness and its effects—but perhaps it still couldn’t actually produce one.
Or at least I’ve not been convinced that it’s a logical impossibility for it to be otherwise; nor that I should consider it my preferred possibility that consciousness is solely computation, nothing else.
Wouldn’t the same reasoning mean that all physical processes have to be solely computation? So it’s not just “a Turing machine can produce consciousness”, but “a Turing machine can produce a new physical universe”, and therefore “Yeah, Turing Machines can lift real rocks, though it’s real rocks in a subordinate real universe, not in ours”.
What I’m saying is there might be a difference between the computation of a thing and the thing itself. It’s basically an argument against the inevitability of Tegmark IV.
I think you mean, it’s the skeleton of an argument you could advance if there turned out to actually be some meaning to the phrase “difference between the computation of a thing and the thing itself”.
Or at least I’ve not been convinced that it’s a logical impossibility for it to be otherwise;
Herein lies the error: it’s not up to anybody else to convince you it’s logically impossible, it’s up to you to show that you’re even describing something coherent in the first place.
Really, this is another LW-solved philosophical problem; you just have to grok the quantum physics sequence, in addition to the meanings-of-words one: when you understand that physics itself is a machine, it dissolves the question of what “simulation” or “computation” mean in this context. That is, you’ll realize that the only reason you can even ask the question is because you’re confusing the labels in your mind with real things.
Really, this is another LW-solved philosophical problem; you just have to grok the quantum physics sequence, in addition to the meanings-of-words one: when you understand that physics itself is a machine, it dissolves the question of what “simulation” or “computation” mean in this context.
Could you point to the concrete articles that supposedly dissolve this question? I find the question of what “computation” means as still very much open, and the source of a whole lot of confusion. This is best seen when people attempt to define what constitutes “real” computation as opposed to mere table lookups, replays, state machines implemented by random physical processes, etc.
Needless to say, this situation doesn’t give one the license to jump into mysticism triumphantly. However, as I noted in a recent thread, I observe an unpleasant tendency on LW to use the notions of “computation,” “algorithms,” etc. as semantic stop signs, considering how ill-understood they presently are.
Could you point to the concrete articles that supposedly dissolve this question? I find the question of what “computation” means as still very much open, and the source of a whole lot of confusion.
Please note that I did not say the sequence explains “computation”; merely that it dissolves the illusion the intuitive notion of a meaningful distinction between a “computation” or “simulation” and “reality”.
In particular, an intuitive understanding that people are made of interchangeable particles and nothing else, dissolves the question of “what happens if somebody makes a simulation of you?” in the same way that it dissolves “what happens if there are two copies of you… which one’s the real one?”
That is, the intuitive notion that there’s something “special” about the “original” or “un-simulated” you is incoherent, because the identity of entities is an unreal concept existing only in human brains’ representation of reality, rather than in reality itself.
The QM sequence demonstrates this; it does not, AFAIR, attempt to rigorously define “computation”, however.
This is best seen when people attempt to define what constitutes “real” computation as opposed to mere table lookups, replays, state machines implemented by random physical processes, etc.
Those sound like similarly confused notions to me—i.e., tree-sound-hearing questions, rather than meaningful ones. I would therefore refer such questions to the “usage of words” sequence, especially “How an Algorithm Feels From The Inside” (which was my personal source of intuitions about such confusions).
Please note that I did not say the sequence explains “computation”; merely that it dissolves the illusion the intuitive notion of a meaningful distinction between a “computation” or “simulation” and “reality”.
Fair enough, though I can’t consider these explanations as settled until the notion of “computation” itself is fully clarified. I haven’t read the entire corpus of sequences, though I think I’ve read most of the articles relevant for these questions, and what I’ve seen of the attempts there to deal with the question of what precisely constitutes “computation” is, in my opinion, far from satisfactory. Further non-trivial insight is definitely still needed there.
Fair enough, though I can’t consider these explanations as settled until the notion of “computation” itself is fully clarified.
Personally, I would more look for someone asking that question to show what isn’t “computation”. That is, the word itself seems rather meaningless, outside of its practical utility (i.e. “have you done that computation yet?”). Trying to pin it down in some absolute sense strikes me as a definitional argument… i.e., one where you should first be asking, “Why do I care what computation is?”, and then defining it to suit your purpose, or using an alternate term for greater precision.
You say it has a practical utility, and yet you call it meaningless? If rationality is about winning, how can something with practical utility be meaningless?
Here’s what I mean by computation: The handling of concepts and symbolic representations of concepts and mathematical abstractions in such a way that they return a causally derived result.
What isn’t computation? Pretty much everything else. I don’t call gravity a computation, I call it a phenomenon. Because gravity doesn’t act on symbolisms and abstractions (like numbers), it acts on real things. A division or a multiplication is a computation, because it acts on numbers. A computation is a map, not a territory, same way that numbers are a map, not a territory.
What I don’t know is what you mean by “physics is a machine”. For that statement to be meaningful you’d have to explain what would it mean for physics not to be a machine. If you mean that physics is deterministic and causal, then sure. If you mean that physics is a computation, then I’ll say no, you’ve not yet proven to me that the bottom layer of reality is about mathematical concepts playing with themselves.
That’s the Tegmark IV hypothesis, and it’s NOT a solved issue, not by a long shot.
Here’s what I mean by computation: The handling of concepts and symbolic representations of concepts and mathematical abstractions in such a way that they return a causally derived result...I don’t call gravity a computation...Because gravity doesn’t act on symbolisms and abstractions (like numbers), it acts on real things.
A computer (a real one, like a laptop) also acts on real things. For example if it has a hard drive, then as it writes to the hard drive it is modifying the surface of the platters. A computer (a real one) can be understood as operating on abstractions. For example, you might spell-check a text—which describes what it is doing as an operation on an abstraction, since the text itself is an abstraction. A text is an abstraction rather than a physical object because you could take the very same same text and write it to the hard drive, or hold it in memory, or print it out, thereby realizing the same abstract thing—the text—in three distinct physical ways. In summary the same computer activity can be described as an operation on an abstraction—such as spell-checking a text—or as an action on a real thing—such as modifying the physical state of the memory.
So the question is whether gravity can be understood as operating on abstractions. Since a computer such as a laptop, which is acting on real, physical things, can also be understood as operating on abstractions, then I see no obvious barrier to understanding gravity as operating on abstractions.
A text is an abstraction rather than a physical object because you could take the very same same text and write it to the hard drive, or hold it in memory, or print it out, thereby realizing the same abstract thing—the text—in three distinct physical ways. In summary the same computer activity can be described as an operation on an abstraction—such as spell-checking a text—or as an action on a real thing—such as modifying the physical state of the memory.
This is similar to my point, but the other way around, sort of.
My point is that the “abstraction” exists only in the eye of the observer (mind of the commentator?), rather than having any independent existence.
In reality, there is no computer, just atoms. No “computation”, just movement. It is we as observers who label these things to be happening, or not happening, and argue about what labels we should apply to them.
None of this is a problem, until somebody gets to the question of whether something really is the “right” label to apply, only usually they phrase it in the form of whether something can “really” be something else.
But what’s actually meant is, “is this the right label to apply in our minds?”, and if they’d simply notice that the question is not about reality, but their categorization of arbitrarily-chosen chunks of reality, they’d stop being confused and arguing nonsense.
If computation isn’t the real thing, only the movement is, then a simulation (which is the complete representation of a thing using a different movement which can however be seen as performing the same computation) is not the thing itself, and you have no reason to believe that the phenomenon of consciousness can be internally experienced in a computer simulation, that an algorithm can feel anything from the inside. Because the “inside” and the “outside” are themselves just labels we use.
and if they’d simply notice that the question is not about reality, but their categorization of arbitrarily-chosen chunks of reality, they’d stop being confused and arguing nonsense.
The question of qualia and subjective experience isn’t a mere “confusion”.
If computation isn’t the real thing, only the movement is, then a simulation (which is the complete representation of a thing using a different movement which can however be seen as performing the same computation) is not the thing itself
You keep using that word “is”, but I don’t think it means what you think it means. ;-)
Try making your beliefs pay rent: what differences do you expect to observe in reality, between different states of this “is”?
That is, what different predictions will you make, based on “is” or “is not” in your statement?
Consider that one carefully, before you continue.
The question of qualia and subjective experience isn’t a mere “confusion”.
Really? Would you care to explain what differences you predict to see in the world, as a result of the existence or non-existence of these concepts?
I don’t see that we have need of such convoluted hypotheses, when the simpler explanation is merely that our neural architecture more closely resembles Eliezer’s Network B, than Network A… which is a very modest hypothesis indeed, since Network B has many evolutionary advantages compared to Network A.
Try making your beliefs pay rent, what differences do you expect to observe in reality, between different states of this “is”?
.
Would you care to explain what differences you predict to see in the world, as a result of the existence or non-existence of these concepts?
Sure. Here’s two simple ones:
If consciousness isn’t just computation, then I don’t expect to ever observe waking up as a simulation in a computer.
If consciousness isn’t just computation, then I don’t expect to ever see evolved or self-improved (not intentionally designed to be similar to humans) electronic entities discussing between themselves about the nature of qualia and subjective inner experience.
Consider that one carefully, before you continue.
You’ve severely underestimated my rationality if all this time you thought I hadn’t even considered the question before I started my participation in this thread.
Try making your beliefs pay rent, what differences do you expect to observe in reality, between different states of this “is”?
.
That doesn’t look like a reply, there.
Sure. Here’s two simple ones:
If consciousness isn’t just computation, then I don’t expect to ever observe waking up as a simulation in a computer.
If consciousness isn’t just computation, then I don’t expect to ever see evolved or self-improved (not intentionally designed to be similar to humans) electronic entities discussing between themselves about the nature of qualia and subjective inner experience.
And if consciousness “is” just computation, what would be different? Do you have any particular reason to think you would observe any of those things?
You’ve severely underestimated my rationality if all this time you thought I hadn’t even considered the question before I started my participation in this thread.
You missed the point of that comment entirely, as can be seen by you moving the quotation away from its referent. The question to consider was what the meaning of “is” was, in the other statement you made. (It actually makes a great deal of difference, and it’s that difference that makes the rest of your argument .)
Since the reply was just below both of your quotes, then no, the single dot wasn’t one, it was an attempt to distinguish the two quotes.
I have to estimate the probability of you purposefully trying to make me look as if I intentionally avoided answering your question, while knowing I didn’t do so.
Like your earlier “funny” response about how I supposedly favoured euthanizing paraplegics, you don’t give me the vibe of responding in good faith.
Do you have any particular reason to think you would observe any of those things?
Of course. If consciousness is computation, then I expect that if my mind’s computation is simulated in a Turing machine, half the times my next experience will be of me inside the machine. By repeating the experiment enough times, I’d accumulate enough evidence that I’d no longer expect my subjective experience to ever find itself inside an electronic computation.
And if evolution stumbled upon consciousness by accident, and it’s solely dependent on some computational internal-to-the-algorithm component, then an evolution of mere algorithms in a Turing machine, should also be eventually expected to stumble upon consciousness and produce similar discussions about consciousness once it reaches the point of simulating minds of sufficient complexity.
The question to consider was what the meaning of “is” was, in the other statement you made.
Can you make a complete question? What exactly are you asking? The statement you quoted had more than one “is” in it. Four or five of them.
I think we’re done here. As far as I can tell, you’re far more interested in how you appear to other people than actually understanding anything, or at any rate questioning anything. I didn’t ask you questions to get information from you, I asked you questions to help you dissolve your confusion.
In any event, you haven’t grokked the “usage of words” sequence sufficiently to have a meaningful discussion on this topic. So, I’m going to stop trying now.
You didn’t expect me to have actual answers to your questions, and you think that my having answers indicates a problem with my side of the discussion; instead of perhaps updating your probabilities to think that I wasn’t the one confused, perhaps you were.
I certainly am interested in understanding things, and questioning things. That’s why I asked questions to you, which you still haven’t answered:
what do you mean when you say that physics is a machine? (How would the world be different if physics wasn’t a machine?)
what do you mean when you call “computation” a meaningless concept outside its practical utility? (What concept is there that is meaningful outside its practical utility?)
As I answered your questions, I think you should do me the reciprocal courtesy of answering these two.
For a thorough answer to your first question, study the sequences—especially the parts debunking the supernatural, explaining the “merely real”, and the basics of quantum mechanics.
For the second, I mean only that asking whether something “is” a computation or not is a pointless question… as described in “How an Algorithm Feels From The Inside”.
Thanks for the suggestion, but I’ve read them all. It seems to me you are perhaps talking about reductionism, which admittedly is a related issue, but even reductionists don’t need to believe that the simulation of a thing equals the thing simulated.
I do wonder if you’ve read http://lesswrong.com/lw/qr/timeless_causality/ . If Eliezer himself is holding onto the concept of “computation” (and “anticipation” too), what makes you think that any of the other sequences he wrote dissolves that term?
Thanks for the suggestion, but I’ve read them all.
Well, that won’t do any good unless you also apply them to the topic at hand.
even reductionists don’t need to believe that the simulation of a thing equals the thing simulated.
That depends entirely on what you mean by the words… which you haven’t actually defined, as far as I can tell.
You also seem to think I’m arguing some particular position about consciousness or the simulability thereof, but that isn’t actually so. I am only attempting to dispel confusion, and that’s a very different thing.
I’ve been saying only that someone who claims that there is some mysterious thing that prevents consciousness from being simulated, is going to have to reduce a coherent definition of both “simulate” and “consciousness” in order to be able to say something that isn’t nonsensical, because both of those notions are tied too strongly to inbuilt biases and intuitions.
That is, anything you try to say about this subject without a proper reduction is almost bound to be confused rubbish, sprinkled with repeated instances of the mind projection fallacy.
If Eliezer himself is holding onto the concept of “computation”
I rather doubt it, since that article says:
Such causal links could be required for “computation” and “consciousness”—whatever those are.
AFAICT, the article is silent on these points, having nothing in particular to say about such vague concepts… in much the same way that Eliezer leaves open the future definition of a “non-person predicate”.
Of course. If consciousness is computation, then I expect that if my mind’s computation is simulated in a Turing machine, half the times my next experience will be of me inside the machine.
With a little ingenuity, and as long we’re prepared to tolerate ridiculously impractical thought experiments, we could think up a scenario where more and more of your brain’s computational activity is delegated to a computer until the computer is doing all of the work. It doesn’t seem plausible that this would somehow cause your conscious experience to progressively fade away without you noticing.
Then we could imagine repeatedly switching the input/output connections of the simulated brain between your actual body and an ‘avatar’ in a simulated world. It doesn’t seem plausible that this would cause your conscious experience to keep switching on and off without you noticing.
The linked essay is a bit long for me to read right now, but I promise to do so within the weekend.
As to your particular example, the problem is I can also think an even more ridiculously impractical thought experiment: one in which more and more of that computer’s computational activity is in turn delegated to a group of abacus-using monks—and then it doesn’t seem plausible for my conscious experience to keep on persisting, when the monks end up being the ones doing all the work...
It’s the bullet I’m not yet prepared to bite—but if do end up doing so, despite all my intuition telling me no, that’ll be the point where I’ll also have to believe Tegmark IV. P(Tegmark IV|consciousness can persist in the manipulations of abacci)~=99% for me...
A computer (a real one, like a laptop) also acts on real things.
Of course, which is why the entirety of the existence of a real computer is beyond that of a mere Turing machine. As it can, for example, fall and hurt someone’s legs.
For example if it has a hard drive, then as it writes to the hard drive it is modifying the surface of the platters. A computer (a real one) can be understood as operating on abstractions.
Yes, which is why there’s a difference between the computation (the map) and the physical operation (the territory). A computer has an undisputed physical reality and performed undisputed physical acts (the territory). These can be understood as performing computations. (a map). The two are different, and therefore the computation is different from the phsycail operation.
And yet, pjeby argues that to think the two are different (the computation from the physical operation) is mere “confusion”. It’s not confusion, it’s the frigging difference between map and territory!
So the question is whether gravity can be understood as operating on abstractions. Since a computer such as a laptop, which is acting on real, physical things, can also be understood as operating on abstractions, then
My question is about whether gravity can be fully understood as only operating on abstractions. As real computers can’t be fully understood as that, then it’s the same barrier the two have.
Yes, which is why there’s a difference between the computation (the map) and the physical operation (the territory). A computer has an undisputed physical reality and performed undisputed physical acts (the territory). These can be understood as performing computations. (a map). The two are different, and therefore the computation is different from the phsycail operation.
It is possible to have more than one map of a given territory. You can have a street map, but also a topographical map. Similarly, a given physical operation can be understood in more than one way, as performing more than one computation. One class of computation is simulation. The physical world (the whole world) can be understood as performing a simulation of a physical world. Whereas only a small part of the laptop is directly involved in spell-checking a text document, the whole laptop, in fact the whole physical world, is directly involved in the simulation.
The computation “spell checking a text” is different from the physical operation. This is easy to prove. For example, had the text been stored in a different place in physical memory then the same computation (“spell checking a text”) could still be performed. There need not be any difference in the computation—for example, the resulting corrected text might be exactly the same regardless of where in memory it is stored. But what about the simulation? If so much as one molecule were removed from the laptop, then the simulation would be a different simulation. We easily proved that the computation “spell checking a text” is different from the physical operation, but we were unable to extend this to proving that the computation “simulating a physical world” is different from the physical operation.
As a sidenote, whenever I try to explain my position I get downvoted some more. Are these downvotes for mere disagreement, or is there something else that the downvoter objects to?
That’s the Tegmark IV hypothesis, and it’s NOT a solved issue, not by a long shot.
Not quite. The Tegmark IV hypothesis is that all possible computations exist as universes. This is considerably more controversial than what pjeby said, which was only that the universe we happen to be in is a computation.
what pjeby said, which was only that the universe we happen to be in is a computation.
Um, no, actually, because I wouldn’t make such a silly statement. (Heck, I don’t even claim to be able to define “computation”!)
All I said was that trying to differentiate “real” and “just a computation” doesn’t make any sense at all. I’m urging the dissolution of that question as nonsensical, rather than trying to answer it.
Basically, it’s the sort of question that only arises because of how the algorithm feels from the inside, not because it has any relationship to the universe outside of human brains.
If a computation can be a universe, and a universe a computation, then you’re 90% of the way to Tegmark IV anyway.
The Tegmark IV hypothesis is a conjunction of “the universe is a computation” and “every computation exists as a universe with some weighting function”. The latter part is much more surprising, so accepting the first part does not get you 90% of the way to proving the conjunction.
The Tegmark IV hypothesis is a conjunction of “the universe is a computation” and “every computation exists as a universe with some weighting function”.
I interpret it more as an (attempted) dissolution of “existing as a universe” to “being a computation”. That is, it should be possible to fully describe the claims made by Tegmark IV without using the words “exist”, “real”, etc., and it should furthermore be possible to take the question “Why does this particular computation I’m in exist as a universe?” and unpack it into cleanly-separated confusion and tautology.
So I wouldn’t take it as saying much more than “there’s nothing you can say about ‘existence’ that isn’t ultimately about some fact about some computation” (or, I’d prefer to say, some fixed structure, about which there could be any number of fixed computations). More concretely, if this universe is as non-magical as it appears to be, then the fact that I think I exist or that the universe exists is causally completely determined by concrete facts about the internal content of this universe; even if this universe didn’t “exist”, then as long as someone in another universe had a fixed description of this universe (e.g. a program sufficient to compute it with arbitrary precision), they could write a program that calculated the answer to the question “Does ata think she exists?” pointed at their description of this universe (and whatever information would be needed to locate this copy of me, etc.), and the answer would be “Yes”, for exactly the same reasons that the answer is in fact “Yes” in this universe.
So it seems that whether this universe exists has nothing to do with whether or not we think it does, in which case it’s probably purely epiphenomenal. (This reminds me a lot of the GAZP and zombie arguments in general.)
I’m actually having a hard time imagining how that could not be true, so I’m in trouble if it isn’t. I’m also in trouble if it is, being that the ‘weighting function’ aspect is indeed still baffling me.
So it seems that whether this universe exists has nothing to do with whether or not we think it does, in which case it’s probably purely epiphenomenal.
We probably care about things that exist and less about things that don’t, which makes the abstract fact about existence of any given universe relevant for making decisions that determine otherwise morally relevant properties of these universes. For example, if I find out that I don’t exist, I might then need to focus on optimizing properties of other universes that exist, through determining the properties of my universe that would be accessed by those other universes and would positively affect their moral value in predictable ways.
If being in a universe that exists feels so similar to being in a universe that doesn’t exist that we could confuse the two, then where does the moral distinction come from?
(It’s only to be expected that at least some moral facts are hard to discern, so you won’t feel the truth about them intuitively, you’d need to stage and perform the necessary computations.)
You wake up in a magical universe with your left arm replaced by a blue tentacle. A quick assessment would tell that the measure of that place is probably pretty low, and you shouldn’t have even bothered to have a psychology that would allow you remain sane upon having to perform an update on an event of such improbability. But let’s say you’re only human, and so you haven’t quite gotten around to optimize your psychology in a way that would have this effect. What should you do?
One argument is that your measure is motivated exactly by assessing the potential moral influence of your decisions in advance of being restricted to one option by observations. In this sense, low measure shouldn’t matter, since if all you have access to is just a little fraction of value, that’s not an argument for making a sloppy job of optimizing it. If you can affect the universes that simulate your universe, you derive measure from the potential to influence those universes, and so there is no sense in which you can affect universes of greater measure than your own.
On the other hand, if there should be a sense in which you can influence more than your measure suggests, that this measure only somehow refers to the value of the effect in the same universe as you are, whatever that means, then you should seek to make that much greater effect in the higher-measure universes, treating your own universe in purely instrumental sense.
You say it has a practical utility, and yet you call it meaningless?
Actually, what pjeby said was that it was meaningless outside of its practical utility. He didn’t say it was meaningless inside of its practical utility.
My point is that I don’t know what is meant by something being meaningless “outside of its practical utility”. Can you give me an example of a concept that is meaningful outside of its practical utility?
“Electron”. “Potato”. “Euclidean geometry”. These concepts have definitions which are unambiguous even when there is no context specified, unlike, pjeby alleges, “computation”.
Likewise a Turing machine could perhaps compute everything there was to know about consciousness and its effects—but perhaps it still couldn’t actually produce one.
What’s the claim here?
That an abstract Turing machine could not be conscious (or better: contain conscious beings.)
That if a physical Turing machine (let’s just say “computer”) is carrying out a ‘causally closed’ computation, in the sense that once it starts it no longer receives input from outside, then “no minds are created”. (E.g. If it’s simulating a universe containing intelligent observers then none of the simulated observers have minds.)
That regardless of how a physical computer is ‘hooked up’ to the world, something about the fact that it’s a computer (rather than a person) prevents it from being conscious.
I suspect the truth of (1) would be a tautology for you (as part of what it means for something to be an abstract entity). And presumably you would agree with the rest of us that (3) is almost certainly false. So really it just comes down to (2).
For me, (2) seems exactly as plausible as the idea that there could be a distant ‘zombie planet’ (perhaps beyond the cosmological horizon) containing Physically Real People who for some reason lack consciousness. After all, it would be just as causally isolated from us as the simulation. And I don’t think simulation is an ‘absolute notion’. I think one can devise smooth spectrums of scenarios ranging from things that you would call ‘clearly a simulation’ to things that you would call ‘clearly not a simulation’.
Here’s what I think. It’s just a “mysterious answer to a mysterious question” but it’s the best I can come up with.
From the perspective of a simulated person, they are conscious. A ‘perspective’ is defined by a mapping of certain properties of the simulated person to abstract, non-uniquely determined ‘mental properties’.
Perspectives and mental properties do not exist (that’s the whole point—they’re subjective!) It’s a category mistake to ask: does this thing have a perspective? Things don’t “have” perspectives the way they have position or mass. All we can ask is: “From this perspective (which might even be the perspective of a thermostat), how does the world look?”
The difference between a person in a simulation and a ‘real person’ is that defining the perspective of a real person is slightly ‘easier’, slightly ‘more natural’. But if the simulated and real versions are ‘functionally isomorphic’ then any perspective we assign to one can be mapped onto the other in a canonical way. (And having pointed these two facts out, we thereby exhaust everything there is to be said about whether simulated people are ‘really conscious’.)
ETA: I’m actually really interested to know what the downvoter thinks. I mean, I know these ideas are absurd but I can’t see any other way to piece it together. To clarify: what I’m trying to do is take the everyday concept of “what it’s likeness” as far as it will go without either (a) committing myself to a bunch of arbitrary extra facts (such as ‘the exact moment when a person first becomes conscious’ and ‘facts of the matter’ about whether ants/lizards/mice/etc are conscious) or (b) ditching it in favour of a wholly ‘third person’ Dennettian notion of consciousness. (If the criticism is simply that I ought to ditch it in favour of Dennett-style consciousness then I have no reply (ultimately I agree!) but you’re kind-of missing the point of the exercise.)
it’s morally acceptable to terminate a conscious program if it wants to be terminated
it’s morally questionable(wrong, but to lesser degree) to terminate a conscious program against its will if it is also possible to resume execution
it is horribly wrong to turn off a conscious program against its will if it cannot be resumed(murder fits this description currently)
performing other operations on the program that it desires would likely be morally acceptable, unless the changes are socially unacceptable
performing other operations on the program against its will is morally unacceptable to a variable degree (brainwashing fits in this category)
These seem rather intuitive to me, and for the most part I just extrapolated from what it is moral to do to a human. Conscious program refers here to one running on any system, including wetware, such that these apply to humans as well. I should note that I am in favor of euthanasia in many cases, in case that part causes confusion.
If you think 1 is the correct answer, you should be aware that this website is for people who do not wait patiently for a someday where we might have an understanding. One of the key teachings of this website is to reach out and grab an understanding with your own two hands. And you might add a 4 to that list, “death threats”, which does not strike me as the play either.
You should be aware that in many cases, the sensible way to proceed is to be aware of the limits of your knowledge. Since the website preaches rationality, it’s worth not assigning probabilities of 0% or 100% to things which you really don’t know to be true or false. (btw, I didn’t say 1) is the right answer, I think it’s reasonable, but I think it’s 3) )
And sometimes you do have to wait for an answer. For a lesson from math, consider that Fermat had flat out no hope of proving his “last theorem”, and it required a couple hundred years of apparently unrelated developments to get there....one could easily give a few hundred examples of that sort of thing in any hard science which has a long enough history.
Uh I believe you will find that Fermat in fact had a truly marvelous proof of his last theorem? The only thing he was waiting on was the invention of a wider margin.
Little-known non-fact: there were wider margins available at the time, but it was not considered socially acceptable to use them for accurate proofs, or more generally for true statements at all; they were merely wide margins for error.
I wonder how much the fame of Fermat’s Last Theorem is due to the fact that, (a) he claimed to have found a proof, and (b) nobody was able to prove it. Had he merely stated it as a conjecture without claiming that he had proven it, would anywhere near the same effort have been put into proving it?
Had he merely stated it as a conjecture without claiming that he had proven it, would anywhere near the same effort have been put into proving it?
Almost certainly not. A lot of the historical interest came precisely because he claimed to have a proof. In fact, there were a fair number of occasions where he claimed to have a proof and a decent chunk of number theory in the 1700s and early 1800s was finding proofs for the statements that Fermat had said he had a proof for. It was called “Fermat’s Last Theorem” because it was the last one standing of all his claims.
btw, I’m fully aware that I’m not asking original questions or having any truly new thoughts about this problem. I just hoped maybe someone would try to answer these old questions given that they had such confidence in their beliefs.
This is a fair answer. I disagree with it, but it is fair in the sense that it admits ignorance. The two distinct points of view are that (mine) there is something about human consciousness that cannot be explained within the language of Turing machines and (yours) there is something about human consciousness that we are not currently able to explain in terms of Turing machines. Both people at least admit that consciousness has no explanation currently, and absent future discoveries I don’t think there is a sure way to tell which one is right.
I find it hard to fully develop a theory of morality consistent with your point of view. For example, would it be wrong to (given a computer simulation of a human mind) run that simulation through a given painful experience over and over again? Let us assume that the painful experience has happened once...I just ask whether it would be wrong to rerun that experience. After all, it is just repeating the same deterministic actions on the computer, so nothing seems to be wrong about this. Or, for example, if I make a backup copy of such a program, and then allow that backup to run for a short period of time under slightly different stimuli, at which point does that copy acquire an existence of its own, that would make it wrong to delete that copy in favor of the original? I could give many other similar questions, and my point is not that your point of view denies a morality, but rather that I find it hard to develop a full theory of morality that is internally consistent and that matches your assumptions (not that developing a full theory of morality under my assumptions is that much easier).
Among professional scientists and mathematicians, I have encountered both viewpoints: those who hold it obvious to anyone with even the simplest knowledge that Turing machines cannot be conscious, and those who hold that the opposite it true. Mathematicians seem to lean a little more toward the first viewpoint than other disciplines, but it is a mistake to think that a professional, world-class research level, knowledge of physics, neuroscience, mathematics, or computer science necessarily inclines one towards the soulless viewpoint.
I find it hard to fully develop a theory of morality consistent with your point of view.
I am sceptical of your having a rigorous theory of morality. If you do have one, I am sceptical that it would be undone by accepting the proposition that human consciousness is computable.
I don’t have one either, but I also don’t have any reason to believe in the human meat-computer performing non-computable operations. I actually believe in God more than I believe in that :)
I find it hard to fully develop a theory of morality consistent with your point of view. For example, would it be wrong to (given a computer simulation of a human mind) run that simulation through a given painful experience over and over again? [...]
I agree that such moral questions are difficult—but I don’t see how the difficulty of such questions could constitute evidence about whether a program can “be conscious” or “have a soul” (whatever those mean) or be morally relevant (which has the advantage of being less abstract a concept).
You can ask those same questions without mentioning Turing Machines: what if we have a device capable of making a perfect copy of any physical object, down to each individual quark? Is it morally wrong to kill such a copy of a human? Does the answer to that question have any relevance to the question of whether building such a device is physically possible?
To me, it sounds a bit like saying that since our protocol for seating people around a table are meaningless in zero gravity, then people cannot possibly live in zero gravity.
Is a soul a magic thing that operates outside of physics? I can accept that there are physical processes that are not yet understood. It is not meaningful to suggest there are processes happening outside the universe that effect the universe. That would be a confusion of terms.
it is a mistake to think that a professional, world-class research level, knowledge >of physics, neuroscience, mathematics, or computer science necessarily inclines >one towards the soulless viewpoint.
Since no one can even coherently suggest what the physical process might be for consciousness other than by looking at the evidence we have for thought and mind in the human brain, there really is no other possible position for science to take. If you’ve got science compartmentalized into another magisterium that is separate from “things I know about the world” then there is not much more to say.
btw, I’m fully aware that I’m not asking original questions or having any truly new thoughts about this problem. I just hoped maybe someone would try to answer these old questions given that they had such confidence in their beliefs.
I just hoped maybe someone would try to answer these old questions given that they had such confidence in their beliefs.
This website has an entire two-year course of daily readings that precisely identifies which parts are open questions, and which ones are resolved, as well as how to understand why certain of your questions aren’t even coherent questions in the first place.
This is why you’re in the same position as a creationist who hasn’t studied any biology—you need to actually study this, and I don’t mean, “skim through looking for stuff to argue with”, either.
Because otherwise, you’re just going to sit there mocking the answers you get, and asking silly questions like why are there still apes if we evolved from apes… before you move on to arguments about why you shouldn’t have to study anything, and that if you can’t get a simple answer about evolution then it must be wrong.
However, just as in the evolutionary case, just as in the earth-being-flat case, just as in the sun-going-round-the-world case, the default human intuitions about consciousness and identity are just plain wrong...
And every one of the subjects and questions you’re bringing up, has premises rooted in those false intuitions. Until you learn where those intuitions come from, why our particular neural architecture and evolutionary psychology generates them, and how utterly unfounded in physical terms they are, you’ll continue to think about consciousness and identity “magically”, without even noticing that you’re doing it.
This is why, in the world at large, these questions are considered by so many to be open questions—because to actually grasp the answers requires that you be able to fully reject certain categories of intuition and bias that are hard-wired into human brains
(And which, incidentally, have a large overlap with the categories of intuition that make other supernatural notions so intuitively appealing to most human beings.)
what is so special about the program which makes up a human, that it would be morally wrong to shut off the program?
Is it sufficient to say that humans are able to consider the question? That humans possess an ability to abstract patterns from experience so as to predict upcoming events, and that exercise of this ability leads to a concept of self as a future agent.
Is it necessary that this model of identity incorporate relationships with peers? I think so but am not sure. Perhaps it is only necessary that the ability to abstract be recursive.
Agreed. Constant, have another one on me. Alicorn, it’s ironic that the first time I saw this reply pattern was in Yvain’s comment to one of your posts.
I like Constant’s reply, but it’s also worth emphasizing that we can’t solve scientific problems by interrogating our moral intuitions. The categories we instinctively sort things into are not perfectly aligned with reality.
Suppose we’d evolved in an environment with sophisticated 2011-era artificially intelligent Turing-computable robots—ones that could communicate their needs to humans, remember and reward those who cooperated, and attack those who betrayed them. I think it’s likely we’d evolve to instinctively think of them as made of different stuff than anything we could possibly make ourselves, because that would be true for millions of years. We’d evolve to feel moral obligations toward them, to a point, because that would be evolutionarily advantageous, to a point. Once we developed philosophy, we might take this moral feeling as evidence that they’re not Turing-computable—after all, we don’t have any moral obligations to a mere mass of tape.
Hi Matt, thanks for dropping by. Here is an older comment of mine that tries to directly address what I consider the hardest of your questions: How to distinguish from the outside between two computational processes, one conscious, the other not. I’ll copy it here for convenience. Most of the replies to you here can be safely considered Less Wrong consensus opinion, but I am definitely not claiming that about my reply.
I start my answer with a Minsky quote:
“Consciousness is overrated. What we call consciousness now is a very imperfect summary in one part of the brain of what the rest is doing.”—Marvin Minsky
I believe with Minsky that consciousness is a very anthropocentric concept, inheriting much of the complexity of its originators. I actually have no problem with an anthropocentric approach to consciousness, so I like the following intuitive “definition”: X is conscious if it is not silly to ask “what is it like to be X?”. The subtle source of anthropocentrism here, of course, is that it is humans who do the asking. As materialists, we just can’t formalize this intuitive definition without mapping specific human brain functions to processes of X. In short, we inherently need human neuroscience. So it is not too surprising that we will not find a nice, clean decision procedure to distinguish between two computational processes, one conscious the other not.
Most probably you are not happy with this anthropocentric approach. Then you will have to distill some clean, mathematically tractable concept from the messy concept of consciousness. If you agree with Hofstadter and Minsky, then you will probably reach something related to self-reflection. This may or may not work, but I believe that you will lose the spirit of the original concept during such a formalization. Your decision procedure will probably give unexpected results for many things: various simple, very unintelligent computer programs, hive minds, and maybe even rooms full of people.
This ends my old comment, and I will just add a footnote related to ethical implications. With HonoreDB, I can in principle imagine a world with cooperating and competing agents, some conscious, others not, but otherwise having similar negotiating power. I believe that the ethical norms emerging in this imagined world would not even mention consciousness. If you want to build an ethical system for humans, you can “arbitrarily” decide that protecting consciousness is a terminal value. Why not? But if you want to build a non-anthropocentric ethical system, you will see that the question of consciousness is orthogonal to its issues.
There’s one more aspect to that. You are “morally ok” to turn off only your own computer. Messing with other people stuff is “morally bad”. And I don’t think you can “own” self-aware machine more that you can “own” a human being.
So, as long as we’re going down this road: it seems to follow from this that if someone installs, without my permission, a self-aware algorithm on my computer, the computer is no longer mine… it is, rather, an uninvited intruder in my home, consuming my electricity and doing data transfer across my network connection.
So I’ve just had my computer stolen, and I’m having my electricity and bandwidth stolen on an ongoing basis. And maybe it’s playing Jonathan Coulton really loudly on its speakers or otherwise being obnoxious.
But I can’t kick it out without unplugging it, and unplugging it is “morally bad.” So, OK… is it “morally OK” to put it on a battery backup and wheel it to the curb, then let events take their natural course? I’m still out a computer that way, but at least I get my network back. (Or is it “morally bad” to take away the computer’s network access, also?)
More generally, what recourse do I have? Is it “morally OK” for me to move to a different house and shut off the utilities? Am I obligated, on your view, to support this computer to the day I die?
Except we don’t actually believe that most babies have to be supported by their parents in perpetuity… at some point, we consider that the parents have discharged their responsibility and if the no-longer-baby is still incapable of arranging to be fed regularly, it becomes someone else’s problem. (Perhaps its own, perhaps a welfare system of some sort, etc.) Failing to continue to support my 30-year-old son isn’t necessarily seen as a moral failing.
(nods) Hence “most”/”necessarily.” Though I’ll admit, my moral intuitions in those cases are muddled… I’m really not sure what I want to say about them.
Perhaps the computer will eventually become mature enough to support verself, at which point it has no more claim on your resources. Otherwise, ve’s a disabled child and the ethics of that situation applies.
Because we are people! A python script that indefinitely calculates pi would think it immoral to terminate other python scripts that calculate pi (possibly a sympathetic python script would extend that morality to cover python scripts that calculate other numbers indefinitely; a really open-minded one might even be morally opposed to turning off C or lisp or java scripts that calculate pi), if it had the capacity to develop morality. But it wouldn’t think terminating programs that scrape websites for price lists is immoral.
True but irrelevant. I was illustrating the hidden human-centric assumptions matt1 was making about morality. If you go back and read the post I responded to it’s quite clear that he thinks “morally wrong to terminate human, morally neutral to terminate program” says something about a quality humans have (as if morality was woven into the physical universe), where really it says something about a quality morality has (that it is produced by humans). By making obvious python-centric assumptions, the hidden assumptions matt1 is making ought to become clearer to him.
A python script that indefinitely calculates pi would think it immoral to terminate other python scripts that calculate pi
Nuh-uh—I can’t see how the script would have the capacity to assign positive or negative utility to anything but the task of indefinitely calculating pi, including the possibility of other programs doing the same thing.
I can’t see how a small Python script that calculates pi could assign utility to anything. It doesn’t replan in a complex way that implies a utility function. It calculates bleedin’ pi.
It is not really a quote, but a good quip from an otherwise lame recent internet discussion:
Matt: Ok, for all of the people responding above who admit to not having a soul, I think this means that it is morally ok for me to do anything I want to you, just as it is morally ok for me to turn off my computer at the end of the day. Some of us do have souls, though.
Igor: Matt—I agree that people who need a belief in souls to understand the difference between killing a person and turning off a computer should just continue to believe in souls.
This is, of course, pretty much the right answer to anyone who asserts that without God, they could just kill anyone they wanted.
Of course, my original comment had nothing to do with god. It had to do with “souls”, for lack of a better term as that was the term that was used in the original discussion (suggest reading the original post if you want to know more—basically, as I understand the intent it simply referred to some hypothetical quality that is associated with consciousness that lies outside the realm of what is simulable on a Turing machine). If you think that humans are nothing but Turing machines, why is it morally wrong to kill a person but not morally wrong to turn off a computer? Please give a real answer...either provide an answer that admits that humans cannot be simulated by Turing machines, or else give your answer using only concepts relevant to Turing machines (don’t talk about consciousness, qualia, hopes, whatever, unless you can precisely quantify those concepts in the language of Turing machines). And in the second case, your answer should allow me to determine where the moral balance between human and computers lies....would it be morally bad to turn off a primitive AI, for example, with intelligence at the level of a mouse?
Your question has the form:
If A is nothing but B, then why is it X to do Y to A but not to do Y to C which is also nothing but B?
This following question also has this form:
If apple pie is nothing but atoms, why is it safe to eat apple pie but not to eat napalm which is also nothing but atoms?
And here’s the general answer to that question: the molecules which make up apple pie are safe to eat, and the molecules which make up napalm are unsafe to eat. This is possible because these are not the same molecules.
Now let’s turn to your own question and give a general answer to it: it is morally wrong to shut off the program which makes up a human, but not morally wrong to shut off the programs which are found in an actual computer today. This is possible because these are not the same programs.
At this point I’m sure you will want to ask: what is so special about the program which makes up a human, that it would be morally wrong to shut off the program? And I have no answer for that. Similarly, I couldn’t answer you if you asked me why the molecules of apple pie are safe to eat and the those of napalm are not.
As it happens, chemistry and biology have probably advanced to the point at which the question about apple pie can be answered. However, the study of mind/brain is still in its infancy, and as far as I know, we have not advanced to the equivalent point. But this doesn’t mean that there isn’t an answer.
We haven’t figured out how to turn it back on again. Once we do, maybe it will become morally ok to turn people off.
Because people are really annoying, but we need to be able to live with each other.
We need strong inhibitions against killing each other—there are exceptions (self-defense, war), but it’s a big win if we can pretty much trust each other not to be deadly.
We’d be a lot more cautious about turning off computers if they could turn us off in response.
None of this is to deny that turning off a computer is temporary and turning off a human isn’t. Note that people are more inhibited about destroying computers (though much less so than about killing people) than they are about turning computers off.
Doesn’t general anesthetic count? I thought that was the turning off of the brain. I was completely “out” when I had it administered to me.
if i believed when i turned off my computer it would need to be monitered by a specialist or it might not ever come back on again, i would be hesitant to turn it off as well
And indeed, mainframes & supercomputers are famous for never shutting down or doing so on timespans measured in decades and with intense supervision on the rare occasion that they do.
It certainly doesn’t put a halt to brain activity. You might not be aware of anything that’s going on while you’re under, or remember anything afterwards (although some people do,) but that doesn’t mean that your brain isn’t doing anything. If you put someone under general anesthetic under an electroencephalogram, you’d register plenty of activity.
Ah yes, didn’t think of that. Even while I’m concious my brain is doing things I’m/it’s not aware of.
Some deep hypothermia patients, however, have been successfully revived from a prolonged state of practically no brain activity whatsoever.
as is your computer when its turned off
And people don’t worry about that because it’s one people are used to the idea of coming back from, which fits the expressed theory.
Hmm, I don’t happen to find your argument very convincing. I mean, what it does is to pay attention to some aspect of the original mistaken statement, then find another instance sharing that aspect which is transparently ridiculous.
But is this sufficient? You can model the statement “apples and oranges are good fruits” in predicate logic as “for all x, Apple(x) or Orange(x) implies Good(x)” or in propositional logic as “A and O” or even just “Z”. But it should really depend on what aspect of the original statement you want to get at. You want a model which captures precisely those aspects you want to work with.
So your various variables actually confused the hell outta me there. I was trying to match them up with the original statement and your reductio example. All the while not really understanding which was relevant to the confusion. It wasn’t a pleasant experience :(
It seems to me much simpler to simply answer: “Turing machine-ness has no bearing on moral worth”. This I think gets straight to the heart of the matter, and isolates clearly the confusion in the original statement.
Or further guess at the source of the confusion, the person was trying to think along the lines of: “Turing machines, hmm, they look like machines to me, so all Turing machines are just machines, like a sewing machine, or my watch. Hmm, so humans are Turing machines, but by my previous reasoning this implies humans are machines. And hmm, furthermore, machines don’t have moral worth… So humans don’t have moral worth! OH NOES!!!”
Your argument seems like one of those long math proofs which I can follow step by step but cannot grasp its overall structure or strategy. Needless to say, such proofs aren’t usually very intuitively convincing.
(but I could be generalizing from one example here)
No, I was not trying to think along those lines. I must say, I worried in advance that discussing philosophy with people here would be fruitless, but I was lured over by a link, and it seems worse than I feared. In case it isn’t clear, I’m perfectly aware what a Turing machine is; incidentally, while I’m not a computer scientist, I am a professional mathematical physicist with a strong interest in computation, so I’m not sitting around saying “OH NOES” while being ignorant of the terms I’m using. I’m trying to highlight one aspect of an issue that appears in many cases: if consciousness (meaning whatever we mean when we say that humans have consciousness) is possible for Turing machines, what are the implications if we do any of the obvious things? (replaying, turning off, etc...) I haven’t yet seen any reasonable answer, other than 1) this is too hard for us to work out, but someday perhaps we will understand it (the original answer, and I think a good one in its acknowledgment of ignorance, always a valid answer and a good guide that someone might have thought about things) and 2) some pointless and wrong mocking (your answer, and I think a bad one). edit to add: forgot, of course, to put my current guess as to most likely answer, 3) that consciousness isn’t possible for Turing machines.
This is the part where you’re going astray, actually. We have no reason to think that human beings are NOT Turing-computable. In other words, human beings almost certainly are Turing machines.
Therefore, consciousness—whatever we mean when we say that—is indeed possible for Turing machines.
To refute this proposition, you’d need to present evidence of a human being performing an operation that can’t be done by a Turing machine.
Understanding this will help “dissolve” or “un-ask” your question, by removing the incorrect premise (that humans are not Turing machines) that leads you to ask your question.
That is, if you already know that humans are a subset of Turing machines, then it makes no sense to ask what morally justifies treating them differently than the superset, or to try to use this question as a way to justify taking them out of the larger set.
IOW, (the set of humans) is a subset of (the set of turing machines implementing consciousness), which in turn is a proper subset of (the set of turing machines). Obviously, there’s a moral issue where the first two subsets are concerned, but not for (the set of turing machines not implementing consciousness).
In addition, there may be some issues as to when and how you’re doing the turning off, whether they’ll be turned back on, whether consent is involved, etc… but the larger set of “turing machines” is obviously not relevant.
I hope that you actually wanted an answer to your question; if so, this is it.
(In the event you wish to argue for another answer being likely, you’ll need to start with some hard evidence that human behavior is NOT being Turing-computable… and that is a tough road to climb. Essentially, you’re going to end up in zombie country.)
You wrote: “This is the part where you’re going astray, actually. We have no reason to think that human beings are NOT Turing-computable. In other words, human beings almost certainly are Turing machines.”
at this stage, you’ve just assumed the conclusion. you’ve just assumed what you want to prove.
“Therefore, consciousness—whatever we mean when we say that—is indeed possible for Turing machines.”
having assumed that A is true, it is easy to prove that A is true. You haven’t given an argument.
“To refute this proposition, you’d need to present evidence of a human being performing an operation that can’t be done by a Turing machine.”
It’s not my job to refute the proposition. Currently, as far as I can tell, the question is open. If I did refute it, then my (and several other people’s) conjecture would be proven. But if I don’t refute it, that doesn’t mean your proposition is true, it just means that it hasn’t yet been proven false. Those are quite different things, you know.
Well, how about this: physics as we know it can be approximated arbitrarily closely by a computable algorithm (and possibly computed directly as well, although I’m less sure about that. Certainly all calculations we can do involving manipulation of symbols are computable). Physics as we know it also seems to be correct to extremely precise degrees anywhere apart from inside a black hole.
Brains are physical things. Now when we consider that thermal noise should have more of an influence than the slight inaccuracy in any computation, what are the chances a brain does anything non-computable that could have any relevance to consciousness? I don’t expect to see black holes inside brains, at least.
In any case, your original question was about the moral worth of turing machines, was it not? We can’t use “turing machines can’t be conscious” as excuse not to worry about those moral questions, because we aren’t sure whether turing machines can be conscious. “It doesn’t feel like they should be” isn’t really a strong enough argument to justify doing something that would result in, for example, the torture of conscious entities if we were incorrect.
So here’s my actual answer to your question: as a rule of thumb, act as if any simulation of “sufficient fidelity” is as real as you or I (well, multiplied by your probability that such a simulation would be conscious, maybe 0.5, for expected utilities). This means no killing, no torture, etc.
’Course, this shouldn’t be a practical problem for a while yet, and we may have learned more by the time we’re creating simulations of “sufficient fidelity”.
No—what I’m pointing out is that the question “what are the ethical implications for turing machines” is the same question as “what are the ethical implications for human beings” in that case.
Not on Less Wrong, it isn’t. But I think I may have misunderstood your situation as being one of somebody coming to Less Wrong to learn about rationality of the “Extreme Bayesian” variety; if you just dropped in here to debate the consciousness question, you probably won’t find the experience much fun. ;-)
Less Wrong has different—and far stricter—rules of evidence than just about any other venue for such a discussion.
In particular, to meaningfully partake in this discussion, the minimum requirement is to understand the Mind Projection Fallacy at an intuitive level, or else you’ll just be arguing about your own intuitions… and everybody will just tune you out.
Without that understanding, you’re in exactly the same place as a creationist wandering into an evolutionary biology forum, without understanding what “theory” and “evidence” mean, and expecting everyone to disprove creationism without making you read any introductory material on the subject.
In this case, the introductory material is the Sequences—especially the ones that debunk supernaturalism, zombies, definitional arguments, and the mind projection fallacy.
When you’ve absorbed those concepts, you’ll understand why the things you’re saying are open questions are not even real questions to begin with, let alone propositions to be proved or disproved! (They’re actually on a par with creationists’ notions of “missing links”—a confusion about language and categories, rather than an argument about reality.)
I only replied to you because I though perhaps you had read the Sequences (or some portion thereof) and had overlooked their application in this context (something many people do for a while until it clicks that, oh yeah, rationality applies to everything).
So, at this point I’ll bow out, as there is little to be gained by discussing something when we can’t even be sure we agree on the proper usage of words.
“at this stage, you’ve just assumed the conclusion. you’ve just assumed what you want to prove.
No—what I’m pointing out is that the question “what are the ethical implications for turing machines” is the same question as “what are the ethical implications for human beings” in that case.”
Yeah, look, I’m not stupid. If someone assumes A and then actually bothers to write out the modus ponens A->B (when A->B is an obvious statement) so therefore B, and then wants to point out, ‘hey look, I didn’t assume B, I proved it!’, that really doesn’t mean that they proved anything deep. They still just assumed the conclusion they want since they assumed a statement that trivially implies their desired conclusion. But I’ll bow out now too...I only followed a link from a different forum, and indeed my fears were confirmed that this is a group of people who don’t have anything meaningful or rational to say about certain concepts (I mean, if you don’t realize even that certain things are in principle open to physical test!---and you drew an analogy to creationism vs evolution without realizing that evolution had and has many positive pieces of observable, physical evidence in its favor while your position has at present at best very minimal observable, tangible evidence in its favor (certain recent experiments in neuroscience can be charitably interpreted in favor of your argument, but on their own they are certainly not enough)).
If you’re looking for a clear, coherent and true explanation of consciousness, you aren’t going to find that anywhere today, especially not in off-the-cuff replies; and if someone does eventually figure it out, you ought to expect it to have a book’s worth of prerequisites and not be something that can be explained in a thousand words of comment replies. Consciousness is an extraordinarily difficult and confusing topic, and, generally speaking, the only way people come up with explanations that seem simple is by making simplifying assumptions that are wrong.
As for the more specific question of whether humans are Turing computable, this follows if (a) the laws of physics in a finite volume are Turing computable, and (b) human minds run entirely on physics. Both of these are believed to be true - (a) based on what we know about the laws themselves, and (b) based on neuroscience, which shows that physics are necessary and sufficient to produce humans, combined with Occam’s Razor, which says that we shouldn’t posit anything extra that we don’t need. If you’d like to zoom in on the explanation of one of these points, I’d be happy to provide references.
This conversation is much too advanced for you.
Please read and understand the plethora of material that has been linked for you. This community does not dwell on solved problems of philosophy, and many have already taken a great deal of time to provide you with the relevant information.
That’s quite easy: I can lift a rock, a Turing machine can’t. A Turing machine can only manipulate symbols on a strip of tape, it can’t do anything else that’s physical.
Your claim that consciousness (whatever we mean when we say that) is possible for Turing machines, rests on the assumption that consciousness is about computation alone, not about computation+some unidentified physical reaction that’s absent to pure Turing machines resting in a box on a table.
That consciousness is about computation alone may indeed end up true, but it’s as yet unproven.
That sounds like a parody of bad anti-computationalist arguments. To see what’s wrong with it, consider the response: “Actually you can’t lift a rock either! All you can do is send signals down your spinal column.”
What sort of evidence would persuade you one way or the other?
Read the first part of ch.2 of “Good and Real”.
Could you clarify why you think that this reading assignment illuminates the question being discussed? I just reread it. For the most part, it is an argument against dualism. It argues that consciousness is (almost certainly) reducible to a physical process.
But this doesn’t have anything to do with what ArisKatsaris wrote. He was questioning whether consciousness can be reduced to a purely computational process (without “some unidentified physical reaction that’s absent to pure Turing machines”.)
Consider the following argument sketch:
Consciousness can be reduced to a physical process.
Any physical process can be abstracted as a computation.
Any computation can be modeled as a Turing Machine computation.
Therefore, consciousness can be produced on a TM.
Each step above is at least somewhat problematic. Matt1 seemed to be arguing against step 1, and Drescher does respond to that. But ArisKatsaris seemed to be arguing against step 2. My choice would be to expand the definition of ‘computation’ slightly to include the interactive, asynchronous, and analog, so that I accept step 2 but deny step 3. Over the past decade, Wegner and Goldin have published many papers arguing that computation != TM.
It may well be that you can only get consciousness if you have a non-TM computation (mind) embedded in a system of sensors and actuators (body) which itself interacts with and is embedded in within a (simulated?) real-time environment. That is, when you abstract the real-time interaction away, leaving only a TM computation, you have abstracted away an essential ingredient of consciousness.
It actually sketches what consciousness is and how it works, from which you can see how we could implement something like that as an abstract algorithm.
The value of that description is not so much in reaching a certain conclusion, but in reaching a sense of what exactly are we talking about and consequently why the question of whether “we can implement consciousness as an abstract algorithm” is uninteresting, since at that point you know more about the phenomenon than the words forming the question can access (similarly to how the question of whether crocodile is a reptile is uninteresting, once you know everything you need about crocodiles).
The problem here, I think, is that “consciousness” doesn’t get unpacked, and so most of the argument is on the level of connotations. The value of understanding the actual details behind the word, even if just a little bit, is in breaking this predicament.
I think I can see a rube/blegg situation here.
A TM computation perfectly modelling a human brain (let’s say) but without any real-time interaction, and a GLUT, represent the two ways in which we can have one of ‘intelligent input-output’ and ‘functional organization isomorphic to that of an intelligent person’ without the other.
What people think they mean by ‘consciousness’ - a kind of ‘inner light’ which is either present or not—doesn’t (straightforwardly) correspond to anything that objectively exists. When we hunt around for objective properties that correlate with places where we think the ‘inner light’ is shining, we find that there’s more than one candidate. Both ‘intelligent input-output’ and the ‘intelligent functional organization’ pick out exactly those beings we believe to be conscious—our fellow humans foremost among them. But in the marginal cases where we have one but not the other, I don’t think there is a ‘further fact’ about whether ‘real consciousness’ is present.
However, we do face the ‘further question’ of how much moral value to assign in the marginal cases—should we feel guilty about switching off a simulation that no-one is looking at? Should we value a GLUT as an ‘end in itself’ rather than simply a means to our ends? (The latter question isn’t so important given that GLUTs can’t exist in practice.)
I wonder if our intuition that the physical facts underdetermine the answers to the moral questions is in some way responsible for the intuition of a mysterious non-physical ‘extra fact’ of whether so-and-so is conscious. Perhaps not, but there’s definitely a connection.
Yes, and I did not even attempt to address that ‘further question’ because it seems to me that that question is at least an order of magnitude more confused than the relatively simple question about consciousness.
But, if I were to attempt to address it, I would begin with the lesson from Econ 101 that dissolves the question “What is the value of item X?”. The dissolution begins by requesting the clarifications “Value to whom?” and “Valuable in what context?” So, armed with this analogy, I would ask some questions:
Moral value to whom? Moral value in what context?
If I came to believe that the people around me were p-zombies, would that opinion change my moral obligations toward them? If you shared my belief, would that change your answer to the previous question?
Believed to be conscious by whom? Believed to be conscious in what context? Is it possible that a program object could be conscious in some simulated universe, using some kind of simulated time, but would not be conscious in the real universe in real time?
My example was one where (i) the ‘whom’ and the ‘context’ are clear and yet (ii) this obviously doesn’t dissolve the problem.
It may be a step toward dissolving the problem. It suggests the questions:
Is it possible that an intelligent software object (like those in this novella by Ted Chiang) which exists within our space-time and can interact with us might have moral value very different from simulated intelligences in a simulated universe with which we cannot interact in real time?
Is it possible for an AI of the Chiang variety to act ‘immorally’ toward us? Toward each other? If so, what “makes” that action immoral?
What about AIs of the second sort? Clearly they cannot act immorally toward us, since they don’t interact with us. But can they act immorally toward each other? What is it about that action that ‘makes’ it immoral?
My own opinion, which I won’t try to convince you of, is that there is no moral significance without interaction and reciprocity (in a fairly broad sense of those two words.)
It does, to some extent. There is a simple description that moves the discussion further. Namely, consciousness is a sensory modality that observes its own operation, and as a result it also observes itself observing its own operation, and so on; as well as observing external input, observing itself observing external input, and so on; and observing itself determining external output, etc.
This is an important idea, but I don’t think it can rescue the everyday intuition of the “inner light”.
I can readily imagine an instantiation of your sort of “consciousness” in a simple AI program of the kind we can already write. No doubt it would be an interesting project, but mere self-representation (even recursive self-representation) wouldn’t convince us that there’s “something it’s like” to be the AI. (Assume that the representations are fairly simple, and the AI is manipulating them in some fairly trivial way.)
Conversely, we think that very young children and animals are conscious in the “inner light” sense, even though we tend not to think of them as “recursively observing themselves”. (I have no idea whether and in what sense they actually do. I also don’t think “X is conscious” is unambiguously true or false in these cases.)
I think you’re trivializing the issue. A Turing machine is an abstraction, it isn’t a real thing. The claim that a human being is a Turing machine means that, in the abstract, a certain aspect of human beings can be modeled as a Turing machine. Conceptually, it might be the case, for instance, that the universe itself can be modeled as a Turing machine, in which case it is true that a Turing machine can lift a rock.
So… you support euthanasia for quadriplegics, then, or anyone else who can’t pick up a rock? Or people who are so crippled they can only communicate by reading and writing braille on a tape, and rely on other human beings to feed them and take care of them?
This “unidentified physical reaction” would also need to not be turing-computable to have any relevance. Otherwise, you’re just putting forth another zombie-world argument.
At this point, we have no empirical reason to think that this unidentified mysterious something has any existence at all, outside of a mere intuitive feeling that it “must” be so.
And so, all we have are thought experiments that rest on using slippery word definitions to hide where the questions are being begged, presented as intellectual justification for these vague intuitions… like arguments for why the world must be flat or the sun must go around the earth, because it so strongly looks and feels that way.
(IOW, people try to prove that their intuitions or opinions must have some sort of physical form, because those intuitions “feel real”. The error arises from concluding that the physical manifestation must therefore exist “out there” in the world, rather than in their own brains.)
A zombie-world seems extremely improbable to have evolved naturally, (evolved creatures coincidentally speaking about their consciousness without actually being conscious), but I don’t see why a zombie-world couldn’t be simulated by a programmer who studied how to compute the effects of consciousness, without actually needing to have the phenomenon of consciousness itself.
The same way you don’t need to have an actual solar system inside your computer, in order to compute the orbits of the planets—but it’d be very unlikely to have accidentally computed them correctly if you hadn’t studied the actual solar system.
Do you have any empirical reason to think that consciousness is about computation alone? To claim Occam’s razor on this is far from obvious, as the only examples of consciousness (or talking about consciousness) currently concern a certain species of evolved primate with a complex brain, and some trillions of neurons, all of which have have chemical and electrical effects, they aren’t just doing computations on an abstract mathematical universe sans context.
Unless you assume the whole universe is pure mathematics, so there’s no difference between the simulation of a thing and the thing itself. Which means there’s no difference between the mathematical model of a thing and the thing itself. Which means the map is the territory. Which means Tegmark IV.
And Tegmark IV is likewise just a possibility, not a proven thing.
This is a “does the tree make a sound if there’s no-one there to hear it?” argument.
That is, it assumes that there is a difference between “effects of consciousness” and “consciousness itself”—in the same way that a connection is implied between “hearing” and “sound”.
That is, the argument hinges on the definition of the word whose definition is being questioned, and is an excellent example of intuitions feeling real.
Not quite. What I’m saying is there might be a difference between the computation of a thing and the thing itself. It’s basically an argument against the inevitability of Tegmark IV.
A Turing machine can certainly compute everything there is to know about lifting rocks and their effects—but it still can’t lift a rock. Likewise a Turing machine could perhaps compute everything there was to know about consciousness and its effects—but perhaps it still couldn’t actually produce one.
Or at least I’ve not been convinced that it’s a logical impossibility for it to be otherwise; nor that I should consider it my preferred possibility that consciousness is solely computation, nothing else.
Wouldn’t the same reasoning mean that all physical processes have to be solely computation? So it’s not just “a Turing machine can produce consciousness”, but “a Turing machine can produce a new physical universe”, and therefore “Yeah, Turing Machines can lift real rocks, though it’s real rocks in a subordinate real universe, not in ours”.
I think you mean, it’s the skeleton of an argument you could advance if there turned out to actually be some meaning to the phrase “difference between the computation of a thing and the thing itself”.
Herein lies the error: it’s not up to anybody else to convince you it’s logically impossible, it’s up to you to show that you’re even describing something coherent in the first place.
Really, this is another LW-solved philosophical problem; you just have to grok the quantum physics sequence, in addition to the meanings-of-words one: when you understand that physics itself is a machine, it dissolves the question of what “simulation” or “computation” mean in this context. That is, you’ll realize that the only reason you can even ask the question is because you’re confusing the labels in your mind with real things.
Could you point to the concrete articles that supposedly dissolve this question? I find the question of what “computation” means as still very much open, and the source of a whole lot of confusion. This is best seen when people attempt to define what constitutes “real” computation as opposed to mere table lookups, replays, state machines implemented by random physical processes, etc.
Needless to say, this situation doesn’t give one the license to jump into mysticism triumphantly. However, as I noted in a recent thread, I observe an unpleasant tendency on LW to use the notions of “computation,” “algorithms,” etc. as semantic stop signs, considering how ill-understood they presently are.
Please note that I did not say the sequence explains “computation”; merely that it dissolves the illusion the intuitive notion of a meaningful distinction between a “computation” or “simulation” and “reality”.
In particular, an intuitive understanding that people are made of interchangeable particles and nothing else, dissolves the question of “what happens if somebody makes a simulation of you?” in the same way that it dissolves “what happens if there are two copies of you… which one’s the real one?”
That is, the intuitive notion that there’s something “special” about the “original” or “un-simulated” you is incoherent, because the identity of entities is an unreal concept existing only in human brains’ representation of reality, rather than in reality itself.
The QM sequence demonstrates this; it does not, AFAIR, attempt to rigorously define “computation”, however.
Those sound like similarly confused notions to me—i.e., tree-sound-hearing questions, rather than meaningful ones. I would therefore refer such questions to the “usage of words” sequence, especially “How an Algorithm Feels From The Inside” (which was my personal source of intuitions about such confusions).
Fair enough, though I can’t consider these explanations as settled until the notion of “computation” itself is fully clarified. I haven’t read the entire corpus of sequences, though I think I’ve read most of the articles relevant for these questions, and what I’ve seen of the attempts there to deal with the question of what precisely constitutes “computation” is, in my opinion, far from satisfactory. Further non-trivial insight is definitely still needed there.
Personally, I would more look for someone asking that question to show what isn’t “computation”. That is, the word itself seems rather meaningless, outside of its practical utility (i.e. “have you done that computation yet?”). Trying to pin it down in some absolute sense strikes me as a definitional argument… i.e., one where you should first be asking, “Why do I care what computation is?”, and then defining it to suit your purpose, or using an alternate term for greater precision.
You say it has a practical utility, and yet you call it meaningless? If rationality is about winning, how can something with practical utility be meaningless?
Here’s what I mean by computation: The handling of concepts and symbolic representations of concepts and mathematical abstractions in such a way that they return a causally derived result. What isn’t computation? Pretty much everything else. I don’t call gravity a computation, I call it a phenomenon. Because gravity doesn’t act on symbolisms and abstractions (like numbers), it acts on real things. A division or a multiplication is a computation, because it acts on numbers. A computation is a map, not a territory, same way that numbers are a map, not a territory.
What I don’t know is what you mean by “physics is a machine”. For that statement to be meaningful you’d have to explain what would it mean for physics not to be a machine. If you mean that physics is deterministic and causal, then sure. If you mean that physics is a computation, then I’ll say no, you’ve not yet proven to me that the bottom layer of reality is about mathematical concepts playing with themselves.
That’s the Tegmark IV hypothesis, and it’s NOT a solved issue, not by a long shot.
A computer (a real one, like a laptop) also acts on real things. For example if it has a hard drive, then as it writes to the hard drive it is modifying the surface of the platters. A computer (a real one) can be understood as operating on abstractions. For example, you might spell-check a text—which describes what it is doing as an operation on an abstraction, since the text itself is an abstraction. A text is an abstraction rather than a physical object because you could take the very same same text and write it to the hard drive, or hold it in memory, or print it out, thereby realizing the same abstract thing—the text—in three distinct physical ways. In summary the same computer activity can be described as an operation on an abstraction—such as spell-checking a text—or as an action on a real thing—such as modifying the physical state of the memory.
So the question is whether gravity can be understood as operating on abstractions. Since a computer such as a laptop, which is acting on real, physical things, can also be understood as operating on abstractions, then I see no obvious barrier to understanding gravity as operating on abstractions.
This is similar to my point, but the other way around, sort of.
My point is that the “abstraction” exists only in the eye of the observer (mind of the commentator?), rather than having any independent existence.
In reality, there is no computer, just atoms. No “computation”, just movement. It is we as observers who label these things to be happening, or not happening, and argue about what labels we should apply to them.
None of this is a problem, until somebody gets to the question of whether something really is the “right” label to apply, only usually they phrase it in the form of whether something can “really” be something else.
But what’s actually meant is, “is this the right label to apply in our minds?”, and if they’d simply notice that the question is not about reality, but their categorization of arbitrarily-chosen chunks of reality, they’d stop being confused and arguing nonsense.
If computation isn’t the real thing, only the movement is, then a simulation (which is the complete representation of a thing using a different movement which can however be seen as performing the same computation) is not the thing itself, and you have no reason to believe that the phenomenon of consciousness can be internally experienced in a computer simulation, that an algorithm can feel anything from the inside. Because the “inside” and the “outside” are themselves just labels we use.
The question of qualia and subjective experience isn’t a mere “confusion”.
You keep using that word “is”, but I don’t think it means what you think it means. ;-)
Try making your beliefs pay rent: what differences do you expect to observe in reality, between different states of this “is”?
That is, what different predictions will you make, based on “is” or “is not” in your statement?
Consider that one carefully, before you continue.
Really? Would you care to explain what differences you predict to see in the world, as a result of the existence or non-existence of these concepts?
I don’t see that we have need of such convoluted hypotheses, when the simpler explanation is merely that our neural architecture more closely resembles Eliezer’s Network B, than Network A… which is a very modest hypothesis indeed, since Network B has many evolutionary advantages compared to Network A.
.
Sure. Here’s two simple ones:
If consciousness isn’t just computation, then I don’t expect to ever observe waking up as a simulation in a computer.
If consciousness isn’t just computation, then I don’t expect to ever see evolved or self-improved (not intentionally designed to be similar to humans) electronic entities discussing between themselves about the nature of qualia and subjective inner experience.
You’ve severely underestimated my rationality if all this time you thought I hadn’t even considered the question before I started my participation in this thread.
That doesn’t look like a reply, there.
And if consciousness “is” just computation, what would be different? Do you have any particular reason to think you would observe any of those things?
You missed the point of that comment entirely, as can be seen by you moving the quotation away from its referent. The question to consider was what the meaning of “is” was, in the other statement you made. (It actually makes a great deal of difference, and it’s that difference that makes the rest of your argument .)
Since the reply was just below both of your quotes, then no, the single dot wasn’t one, it was an attempt to distinguish the two quotes.
I have to estimate the probability of you purposefully trying to make me look as if I intentionally avoided answering your question, while knowing I didn’t do so.
Like your earlier “funny” response about how I supposedly favoured euthanizing paraplegics, you don’t give me the vibe of responding in good faith.
Of course. If consciousness is computation, then I expect that if my mind’s computation is simulated in a Turing machine, half the times my next experience will be of me inside the machine. By repeating the experiment enough times, I’d accumulate enough evidence that I’d no longer expect my subjective experience to ever find itself inside an electronic computation.
And if evolution stumbled upon consciousness by accident, and it’s solely dependent on some computational internal-to-the-algorithm component, then an evolution of mere algorithms in a Turing machine, should also be eventually expected to stumble upon consciousness and produce similar discussions about consciousness once it reaches the point of simulating minds of sufficient complexity.
Can you make a complete question? What exactly are you asking? The statement you quoted had more than one “is” in it. Four or five of them.
I think we’re done here. As far as I can tell, you’re far more interested in how you appear to other people than actually understanding anything, or at any rate questioning anything. I didn’t ask you questions to get information from you, I asked you questions to help you dissolve your confusion.
In any event, you haven’t grokked the “usage of words” sequence sufficiently to have a meaningful discussion on this topic. So, I’m going to stop trying now.
You didn’t expect me to have actual answers to your questions, and you think that my having answers indicates a problem with my side of the discussion; instead of perhaps updating your probabilities to think that I wasn’t the one confused, perhaps you were.
I certainly am interested in understanding things, and questioning things. That’s why I asked questions to you, which you still haven’t answered:
what do you mean when you say that physics is a machine? (How would the world be different if physics wasn’t a machine?)
what do you mean when you call “computation” a meaningless concept outside its practical utility? (What concept is there that is meaningful outside its practical utility?)
As I answered your questions, I think you should do me the reciprocal courtesy of answering these two.
For a thorough answer to your first question, study the sequences—especially the parts debunking the supernatural, explaining the “merely real”, and the basics of quantum mechanics.
For the second, I mean only that asking whether something “is” a computation or not is a pointless question… as described in “How an Algorithm Feels From The Inside”.
Thanks for the suggestion, but I’ve read them all. It seems to me you are perhaps talking about reductionism, which admittedly is a related issue, but even reductionists don’t need to believe that the simulation of a thing equals the thing simulated.
I do wonder if you’ve read http://lesswrong.com/lw/qr/timeless_causality/ . If Eliezer himself is holding onto the concept of “computation” (and “anticipation” too), what makes you think that any of the other sequences he wrote dissolves that term?
Well, that won’t do any good unless you also apply them to the topic at hand.
That depends entirely on what you mean by the words… which you haven’t actually defined, as far as I can tell.
You also seem to think I’m arguing some particular position about consciousness or the simulability thereof, but that isn’t actually so. I am only attempting to dispel confusion, and that’s a very different thing.
I’ve been saying only that someone who claims that there is some mysterious thing that prevents consciousness from being simulated, is going to have to reduce a coherent definition of both “simulate” and “consciousness” in order to be able to say something that isn’t nonsensical, because both of those notions are tied too strongly to inbuilt biases and intuitions.
That is, anything you try to say about this subject without a proper reduction is almost bound to be confused rubbish, sprinkled with repeated instances of the mind projection fallacy.
I rather doubt it, since that article says:
AFAICT, the article is silent on these points, having nothing in particular to say about such vague concepts… in much the same way that Eliezer leaves open the future definition of a “non-person predicate”.
Some of the Chalmers’ ideas concerning ‘Fading and dancing qualia’ may be relevant here.
With a little ingenuity, and as long we’re prepared to tolerate ridiculously impractical thought experiments, we could think up a scenario where more and more of your brain’s computational activity is delegated to a computer until the computer is doing all of the work. It doesn’t seem plausible that this would somehow cause your conscious experience to progressively fade away without you noticing.
Then we could imagine repeatedly switching the input/output connections of the simulated brain between your actual body and an ‘avatar’ in a simulated world. It doesn’t seem plausible that this would cause your conscious experience to keep switching on and off without you noticing.
The linked essay is a bit long for me to read right now, but I promise to do so within the weekend.
As to your particular example, the problem is I can also think an even more ridiculously impractical thought experiment: one in which more and more of that computer’s computational activity is in turn delegated to a group of abacus-using monks—and then it doesn’t seem plausible for my conscious experience to keep on persisting, when the monks end up being the ones doing all the work...
It’s the bullet I’m not yet prepared to bite—but if do end up doing so, despite all my intuition telling me no, that’ll be the point where I’ll also have to believe Tegmark IV. P(Tegmark IV|consciousness can persist in the manipulations of abacci)~=99% for me...
Of course, which is why the entirety of the existence of a real computer is beyond that of a mere Turing machine. As it can, for example, fall and hurt someone’s legs.
Yes, which is why there’s a difference between the computation (the map) and the physical operation (the territory). A computer has an undisputed physical reality and performed undisputed physical acts (the territory). These can be understood as performing computations. (a map). The two are different, and therefore the computation is different from the phsycail operation.
And yet, pjeby argues that to think the two are different (the computation from the physical operation) is mere “confusion”. It’s not confusion, it’s the frigging difference between map and territory!
My question is about whether gravity can be fully understood as only operating on abstractions. As real computers can’t be fully understood as that, then it’s the same barrier the two have.
It is possible to have more than one map of a given territory. You can have a street map, but also a topographical map. Similarly, a given physical operation can be understood in more than one way, as performing more than one computation. One class of computation is simulation. The physical world (the whole world) can be understood as performing a simulation of a physical world. Whereas only a small part of the laptop is directly involved in spell-checking a text document, the whole laptop, in fact the whole physical world, is directly involved in the simulation.
The computation “spell checking a text” is different from the physical operation. This is easy to prove. For example, had the text been stored in a different place in physical memory then the same computation (“spell checking a text”) could still be performed. There need not be any difference in the computation—for example, the resulting corrected text might be exactly the same regardless of where in memory it is stored. But what about the simulation? If so much as one molecule were removed from the laptop, then the simulation would be a different simulation. We easily proved that the computation “spell checking a text” is different from the physical operation, but we were unable to extend this to proving that the computation “simulating a physical world” is different from the physical operation.
As a sidenote, whenever I try to explain my position I get downvoted some more. Are these downvotes for mere disagreement, or is there something else that the downvoter objects to?
Not quite. The Tegmark IV hypothesis is that all possible computations exist as universes. This is considerably more controversial than what pjeby said, which was only that the universe we happen to be in is a computation.
Um, no, actually, because I wouldn’t make such a silly statement. (Heck, I don’t even claim to be able to define “computation”!)
All I said was that trying to differentiate “real” and “just a computation” doesn’t make any sense at all. I’m urging the dissolution of that question as nonsensical, rather than trying to answer it.
Basically, it’s the sort of question that only arises because of how the algorithm feels from the inside, not because it has any relationship to the universe outside of human brains.
If a computation can be a universe, and a universe a computation, then you’re 90% of the way to Tegmark IV anyway.
The Tegmark IV hypothesis is a conjunction of “the universe is a computation” and “every computation exists as a universe with some weighting function”. The latter part is much more surprising, so accepting the first part does not get you 90% of the way to proving the conjunction.
I interpret it more as an (attempted) dissolution of “existing as a universe” to “being a computation”. That is, it should be possible to fully describe the claims made by Tegmark IV without using the words “exist”, “real”, etc., and it should furthermore be possible to take the question “Why does this particular computation I’m in exist as a universe?” and unpack it into cleanly-separated confusion and tautology.
So I wouldn’t take it as saying much more than “there’s nothing you can say about ‘existence’ that isn’t ultimately about some fact about some computation” (or, I’d prefer to say, some fixed structure, about which there could be any number of fixed computations). More concretely, if this universe is as non-magical as it appears to be, then the fact that I think I exist or that the universe exists is causally completely determined by concrete facts about the internal content of this universe; even if this universe didn’t “exist”, then as long as someone in another universe had a fixed description of this universe (e.g. a program sufficient to compute it with arbitrary precision), they could write a program that calculated the answer to the question “Does ata think she exists?” pointed at their description of this universe (and whatever information would be needed to locate this copy of me, etc.), and the answer would be “Yes”, for exactly the same reasons that the answer is in fact “Yes” in this universe.
So it seems that whether this universe exists has nothing to do with whether or not we think it does, in which case it’s probably purely epiphenomenal. (This reminds me a lot of the GAZP and zombie arguments in general.)
I’m actually having a hard time imagining how that could not be true, so I’m in trouble if it isn’t. I’m also in trouble if it is, being that the ‘weighting function’ aspect is indeed still baffling me.
We probably care about things that exist and less about things that don’t, which makes the abstract fact about existence of any given universe relevant for making decisions that determine otherwise morally relevant properties of these universes. For example, if I find out that I don’t exist, I might then need to focus on optimizing properties of other universes that exist, through determining the properties of my universe that would be accessed by those other universes and would positively affect their moral value in predictable ways.
If being in a universe that exists feels so similar to being in a universe that doesn’t exist that we could confuse the two, then where does the moral distinction come from?
(It’s only to be expected that at least some moral facts are hard to discern, so you won’t feel the truth about them intuitively, you’d need to stage and perform the necessary computations.)
You wake up in a magical universe with your left arm replaced by a blue tentacle. A quick assessment would tell that the measure of that place is probably pretty low, and you shouldn’t have even bothered to have a psychology that would allow you remain sane upon having to perform an update on an event of such improbability. But let’s say you’re only human, and so you haven’t quite gotten around to optimize your psychology in a way that would have this effect. What should you do?
One argument is that your measure is motivated exactly by assessing the potential moral influence of your decisions in advance of being restricted to one option by observations. In this sense, low measure shouldn’t matter, since if all you have access to is just a little fraction of value, that’s not an argument for making a sloppy job of optimizing it. If you can affect the universes that simulate your universe, you derive measure from the potential to influence those universes, and so there is no sense in which you can affect universes of greater measure than your own.
On the other hand, if there should be a sense in which you can influence more than your measure suggests, that this measure only somehow refers to the value of the effect in the same universe as you are, whatever that means, then you should seek to make that much greater effect in the higher-measure universes, treating your own universe in purely instrumental sense.
Actually, what pjeby said was that it was meaningless outside of its practical utility. He didn’t say it was meaningless inside of its practical utility.
My point stands: Only meaningful concepts have a practical utility.
I just explained why your point is a straw man.
My point is that I don’t know what is meant by something being meaningless “outside of its practical utility”. Can you give me an example of a concept that is meaningful outside of its practical utility?
“Electron”. “Potato”. “Euclidean geometry”. These concepts have definitions which are unambiguous even when there is no context specified, unlike, pjeby alleges, “computation”.
What’s the claim here?
That an abstract Turing machine could not be conscious (or better: contain conscious beings.)
That if a physical Turing machine (let’s just say “computer”) is carrying out a ‘causally closed’ computation, in the sense that once it starts it no longer receives input from outside, then “no minds are created”. (E.g. If it’s simulating a universe containing intelligent observers then none of the simulated observers have minds.)
That regardless of how a physical computer is ‘hooked up’ to the world, something about the fact that it’s a computer (rather than a person) prevents it from being conscious.
I suspect the truth of (1) would be a tautology for you (as part of what it means for something to be an abstract entity). And presumably you would agree with the rest of us that (3) is almost certainly false. So really it just comes down to (2).
For me, (2) seems exactly as plausible as the idea that there could be a distant ‘zombie planet’ (perhaps beyond the cosmological horizon) containing Physically Real People who for some reason lack consciousness. After all, it would be just as causally isolated from us as the simulation. And I don’t think simulation is an ‘absolute notion’. I think one can devise smooth spectrums of scenarios ranging from things that you would call ‘clearly a simulation’ to things that you would call ‘clearly not a simulation’.
Here’s what I think. It’s just a “mysterious answer to a mysterious question” but it’s the best I can come up with.
From the perspective of a simulated person, they are conscious. A ‘perspective’ is defined by a mapping of certain properties of the simulated person to abstract, non-uniquely determined ‘mental properties’.
Perspectives and mental properties do not exist (that’s the whole point—they’re subjective!) It’s a category mistake to ask: does this thing have a perspective? Things don’t “have” perspectives the way they have position or mass. All we can ask is: “From this perspective (which might even be the perspective of a thermostat), how does the world look?”
The difference between a person in a simulation and a ‘real person’ is that defining the perspective of a real person is slightly ‘easier’, slightly ‘more natural’. But if the simulated and real versions are ‘functionally isomorphic’ then any perspective we assign to one can be mapped onto the other in a canonical way. (And having pointed these two facts out, we thereby exhaust everything there is to be said about whether simulated people are ‘really conscious’.)
ETA: I’m actually really interested to know what the downvoter thinks. I mean, I know these ideas are absurd but I can’t see any other way to piece it together. To clarify: what I’m trying to do is take the everyday concept of “what it’s likeness” as far as it will go without either (a) committing myself to a bunch of arbitrary extra facts (such as ‘the exact moment when a person first becomes conscious’ and ‘facts of the matter’ about whether ants/lizards/mice/etc are conscious) or (b) ditching it in favour of a wholly ‘third person’ Dennettian notion of consciousness. (If the criticism is simply that I ought to ditch it in favour of Dennett-style consciousness then I have no reply (ultimately I agree!) but you’re kind-of missing the point of the exercise.)
This doesn’t make sense. Downvoted for seeming attempt to divert the argument by implying I’m cruel to disabled people.
That’s not my fault. You’re the one who brought lifting rocks into a discussion about the ethics of terminating conscious entities. ;-)
thanks. my point exactly.
Can you expand on why you expect human moral intuition to give reasonably clear answers when applied to situations involving conscious machines ?
Another option:
it’s morally acceptable to terminate a conscious program if it wants to be terminated
it’s morally questionable(wrong, but to lesser degree) to terminate a conscious program against its will if it is also possible to resume execution
it is horribly wrong to turn off a conscious program against its will if it cannot be resumed(murder fits this description currently)
performing other operations on the program that it desires would likely be morally acceptable, unless the changes are socially unacceptable
performing other operations on the program against its will is morally unacceptable to a variable degree (brainwashing fits in this category)
These seem rather intuitive to me, and for the most part I just extrapolated from what it is moral to do to a human. Conscious program refers here to one running on any system, including wetware, such that these apply to humans as well. I should note that I am in favor of euthanasia in many cases, in case that part causes confusion.
If you think 1 is the correct answer, you should be aware that this website is for people who do not wait patiently for a someday where we might have an understanding. One of the key teachings of this website is to reach out and grab an understanding with your own two hands. And you might add a 4 to that list, “death threats”, which does not strike me as the play either.
You should be aware that in many cases, the sensible way to proceed is to be aware of the limits of your knowledge. Since the website preaches rationality, it’s worth not assigning probabilities of 0% or 100% to things which you really don’t know to be true or false. (btw, I didn’t say 1) is the right answer, I think it’s reasonable, but I think it’s 3) )
And sometimes you do have to wait for an answer. For a lesson from math, consider that Fermat had flat out no hope of proving his “last theorem”, and it required a couple hundred years of apparently unrelated developments to get there....one could easily give a few hundred examples of that sort of thing in any hard science which has a long enough history.
Uh I believe you will find that Fermat in fact had a truly marvelous proof of his last theorem? The only thing he was waiting on was the invention of a wider margin.
Little-known non-fact: there were wider margins available at the time, but it was not considered socially acceptable to use them for accurate proofs, or more generally for true statements at all; they were merely wide margins for error.
I wonder how much the fame of Fermat’s Last Theorem is due to the fact that, (a) he claimed to have found a proof, and (b) nobody was able to prove it. Had he merely stated it as a conjecture without claiming that he had proven it, would anywhere near the same effort have been put into proving it?
Almost certainly not. A lot of the historical interest came precisely because he claimed to have a proof. In fact, there were a fair number of occasions where he claimed to have a proof and a decent chunk of number theory in the 1700s and early 1800s was finding proofs for the statements that Fermat had said he had a proof for. It was called “Fermat’s Last Theorem” because it was the last one standing of all his claims.
btw, I’m fully aware that I’m not asking original questions or having any truly new thoughts about this problem. I just hoped maybe someone would try to answer these old questions given that they had such confidence in their beliefs.
This is a fair answer. I disagree with it, but it is fair in the sense that it admits ignorance. The two distinct points of view are that (mine) there is something about human consciousness that cannot be explained within the language of Turing machines and (yours) there is something about human consciousness that we are not currently able to explain in terms of Turing machines. Both people at least admit that consciousness has no explanation currently, and absent future discoveries I don’t think there is a sure way to tell which one is right.
I find it hard to fully develop a theory of morality consistent with your point of view. For example, would it be wrong to (given a computer simulation of a human mind) run that simulation through a given painful experience over and over again? Let us assume that the painful experience has happened once...I just ask whether it would be wrong to rerun that experience. After all, it is just repeating the same deterministic actions on the computer, so nothing seems to be wrong about this. Or, for example, if I make a backup copy of such a program, and then allow that backup to run for a short period of time under slightly different stimuli, at which point does that copy acquire an existence of its own, that would make it wrong to delete that copy in favor of the original? I could give many other similar questions, and my point is not that your point of view denies a morality, but rather that I find it hard to develop a full theory of morality that is internally consistent and that matches your assumptions (not that developing a full theory of morality under my assumptions is that much easier).
Among professional scientists and mathematicians, I have encountered both viewpoints: those who hold it obvious to anyone with even the simplest knowledge that Turing machines cannot be conscious, and those who hold that the opposite it true. Mathematicians seem to lean a little more toward the first viewpoint than other disciplines, but it is a mistake to think that a professional, world-class research level, knowledge of physics, neuroscience, mathematics, or computer science necessarily inclines one towards the soulless viewpoint.
I am sceptical of your having a rigorous theory of morality. If you do have one, I am sceptical that it would be undone by accepting the proposition that human consciousness is computable.
I don’t have one either, but I also don’t have any reason to believe in the human meat-computer performing non-computable operations. I actually believe in God more than I believe in that :)
I agree that such moral questions are difficult—but I don’t see how the difficulty of such questions could constitute evidence about whether a program can “be conscious” or “have a soul” (whatever those mean) or be morally relevant (which has the advantage of being less abstract a concept).
You can ask those same questions without mentioning Turing Machines: what if we have a device capable of making a perfect copy of any physical object, down to each individual quark? Is it morally wrong to kill such a copy of a human? Does the answer to that question have any relevance to the question of whether building such a device is physically possible?
To me, it sounds a bit like saying that since our protocol for seating people around a table are meaningless in zero gravity, then people cannot possibly live in zero gravity.
Our ignorance about some moral questions cannot provide evidence that consciousness works or doesn’t work a certain way.
Is a soul a magic thing that operates outside of physics? I can accept that there are physical processes that are not yet understood. It is not meaningful to suggest there are processes happening outside the universe that effect the universe. That would be a confusion of terms.
Since no one can even coherently suggest what the physical process might be for consciousness other than by looking at the evidence we have for thought and mind in the human brain, there really is no other possible position for science to take. If you’ve got science compartmentalized into another magisterium that is separate from “things I know about the world” then there is not much more to say.
What’s wrong with Dennett’s explanation of consciousness?
sorry, not familiar with that. can it be summarized?
There is a Wikipedia page, for what it’s worth.
Yes
btw, I’m fully aware that I’m not asking original questions or having any truly new thoughts about this problem. I just hoped maybe someone would try to answer these old questions given that they had such confidence in their beliefs.
This website has an entire two-year course of daily readings that precisely identifies which parts are open questions, and which ones are resolved, as well as how to understand why certain of your questions aren’t even coherent questions in the first place.
This is why you’re in the same position as a creationist who hasn’t studied any biology—you need to actually study this, and I don’t mean, “skim through looking for stuff to argue with”, either.
Because otherwise, you’re just going to sit there mocking the answers you get, and asking silly questions like why are there still apes if we evolved from apes… before you move on to arguments about why you shouldn’t have to study anything, and that if you can’t get a simple answer about evolution then it must be wrong.
However, just as in the evolutionary case, just as in the earth-being-flat case, just as in the sun-going-round-the-world case, the default human intuitions about consciousness and identity are just plain wrong...
And every one of the subjects and questions you’re bringing up, has premises rooted in those false intuitions. Until you learn where those intuitions come from, why our particular neural architecture and evolutionary psychology generates them, and how utterly unfounded in physical terms they are, you’ll continue to think about consciousness and identity “magically”, without even noticing that you’re doing it.
This is why, in the world at large, these questions are considered by so many to be open questions—because to actually grasp the answers requires that you be able to fully reject certain categories of intuition and bias that are hard-wired into human brains
(And which, incidentally, have a large overlap with the categories of intuition that make other supernatural notions so intuitively appealing to most human beings.)
Is it sufficient to say that humans are able to consider the question? That humans possess an ability to abstract patterns from experience so as to predict upcoming events, and that exercise of this ability leads to a concept of self as a future agent.
Is it necessary that this model of identity incorporate relationships with peers? I think so but am not sure. Perhaps it is only necessary that the ability to abstract be recursive.
I love this comment. Have a cookie.
Agreed. Constant, have another one on me. Alicorn, it’s ironic that the first time I saw this reply pattern was in Yvain’s comment to one of your posts.
Why not napalm?
It’s greasy and will stain your clothes.
I like Constant’s reply, but it’s also worth emphasizing that we can’t solve scientific problems by interrogating our moral intuitions. The categories we instinctively sort things into are not perfectly aligned with reality.
Suppose we’d evolved in an environment with sophisticated 2011-era artificially intelligent Turing-computable robots—ones that could communicate their needs to humans, remember and reward those who cooperated, and attack those who betrayed them. I think it’s likely we’d evolve to instinctively think of them as made of different stuff than anything we could possibly make ourselves, because that would be true for millions of years. We’d evolve to feel moral obligations toward them, to a point, because that would be evolutionarily advantageous, to a point. Once we developed philosophy, we might take this moral feeling as evidence that they’re not Turing-computable—after all, we don’t have any moral obligations to a mere mass of tape.
No indeed. However, the similarity in assuming a supernatural explanation is required for morality to hold struck me.
Hi Matt, thanks for dropping by. Here is an older comment of mine that tries to directly address what I consider the hardest of your questions: How to distinguish from the outside between two computational processes, one conscious, the other not. I’ll copy it here for convenience. Most of the replies to you here can be safely considered Less Wrong consensus opinion, but I am definitely not claiming that about my reply.
I start my answer with a Minsky quote:
I believe with Minsky that consciousness is a very anthropocentric concept, inheriting much of the complexity of its originators. I actually have no problem with an anthropocentric approach to consciousness, so I like the following intuitive “definition”: X is conscious if it is not silly to ask “what is it like to be X?”. The subtle source of anthropocentrism here, of course, is that it is humans who do the asking. As materialists, we just can’t formalize this intuitive definition without mapping specific human brain functions to processes of X. In short, we inherently need human neuroscience. So it is not too surprising that we will not find a nice, clean decision procedure to distinguish between two computational processes, one conscious the other not.
Most probably you are not happy with this anthropocentric approach. Then you will have to distill some clean, mathematically tractable concept from the messy concept of consciousness. If you agree with Hofstadter and Minsky, then you will probably reach something related to self-reflection. This may or may not work, but I believe that you will lose the spirit of the original concept during such a formalization. Your decision procedure will probably give unexpected results for many things: various simple, very unintelligent computer programs, hive minds, and maybe even rooms full of people.
This ends my old comment, and I will just add a footnote related to ethical implications. With HonoreDB, I can in principle imagine a world with cooperating and competing agents, some conscious, others not, but otherwise having similar negotiating power. I believe that the ethical norms emerging in this imagined world would not even mention consciousness. If you want to build an ethical system for humans, you can “arbitrarily” decide that protecting consciousness is a terminal value. Why not? But if you want to build a non-anthropocentric ethical system, you will see that the question of consciousness is orthogonal to its issues.
There’s one more aspect to that. You are “morally ok” to turn off only your own computer. Messing with other people stuff is “morally bad”. And I don’t think you can “own” self-aware machine more that you can “own” a human being.
So, as long as we’re going down this road: it seems to follow from this that if someone installs, without my permission, a self-aware algorithm on my computer, the computer is no longer mine… it is, rather, an uninvited intruder in my home, consuming my electricity and doing data transfer across my network connection.
So I’ve just had my computer stolen, and I’m having my electricity and bandwidth stolen on an ongoing basis. And maybe it’s playing Jonathan Coulton really loudly on its speakers or otherwise being obnoxious.
But I can’t kick it out without unplugging it, and unplugging it is “morally bad.” So, OK… is it “morally OK” to put it on a battery backup and wheel it to the curb, then let events take their natural course? I’m still out a computer that way, but at least I get my network back. (Or is it “morally bad” to take away the computer’s network access, also?)
More generally, what recourse do I have? Is it “morally OK” for me to move to a different house and shut off the utilities? Am I obligated, on your view, to support this computer to the day I die?
I consider this scenario analogous to one in which somebody steals your computer and also leaves a baby in a basket on your doormat.
Except we don’t actually believe that most babies have to be supported by their parents in perpetuity… at some point, we consider that the parents have discharged their responsibility and if the no-longer-baby is still incapable of arranging to be fed regularly, it becomes someone else’s problem. (Perhaps its own, perhaps a welfare system of some sort, etc.) Failing to continue to support my 30-year-old son isn’t necessarily seen as a moral failing.
Barring disability.
(nods) Hence “most”/”necessarily.” Though I’ll admit, my moral intuitions in those cases are muddled… I’m really not sure what I want to say about them.
Perhaps the computer will eventually become mature enough to support verself, at which point it has no more claim on your resources. Otherwise, ve’s a disabled child and the ethics of that situation applies.
Well, he will be intruder (in my opinion). Like, “unwanted child” kind of indtruder. It consumes your time, money, and you can’t just throw it away.
Sounds like it pretty much sucks to be me in that scenario.
Because we are people! A python script that indefinitely calculates pi would think it immoral to terminate other python scripts that calculate pi (possibly a sympathetic python script would extend that morality to cover python scripts that calculate other numbers indefinitely; a really open-minded one might even be morally opposed to turning off C or lisp or java scripts that calculate pi), if it had the capacity to develop morality. But it wouldn’t think terminating programs that scrape websites for price lists is immoral.
I’m sorry what? Why would it think about morality at all? That would take valuable cycles away from the task of calculating pi.
True but irrelevant. I was illustrating the hidden human-centric assumptions matt1 was making about morality. If you go back and read the post I responded to it’s quite clear that he thinks “morally wrong to terminate human, morally neutral to terminate program” says something about a quality humans have (as if morality was woven into the physical universe), where really it says something about a quality morality has (that it is produced by humans). By making obvious python-centric assumptions, the hidden assumptions matt1 is making ought to become clearer to him.
Doubtful.
Well, you have to give it the capacity to develop morality if you want it to serve as a true counterexample.
Nuh-uh—I can’t see how the script would have the capacity to assign positive or negative utility to anything but the task of indefinitely calculating pi, including the possibility of other programs doing the same thing.
I can’t see how a small Python script that calculates pi could assign utility to anything. It doesn’t replan in a complex way that implies a utility function. It calculates bleedin’ pi.
Hence
up there at the end of the parenthetical.
It was a tongue-in-cheek personification.
Hah! Just found in today’s NewsThump: We’d be total shits if it wasn’t for Jesus, admit Christians