I can’t wrap my brain around the computational theory of consciousness.
Who decides how to interpret the computations? If I have a beach, are the lighter grains 0 and darker grains 1? What about the smaller and bigger grains? What if I decide to use the motion of the planets to switch between these 4 interpretations.
Surely under infinite definitions of computation, there are infinite consciousnesses experience infinite states at any given time, just from pure chance.
Suppose that consciousness were not a no-place function, but rather a one-place function. Specifically, whether something is conscious or not is relative to some reality. (A bit like movement relative to reference frames in physics.)
Specifically, whether something is conscious or not is relative to some reality
How does this relate back to the example with the sand? Is there a sand-planet reality that’s just like ours, but in that reality the sand is conscious and we’re not?
I don’t think I quite get what a reality is in the function.
I was thinking of the computational theory of consciousness as basically being the same thing as saying that consciousness could be substrate independent. (E.g. you could have conscious uploads.)
I think this then leads you to ask, “If consciousness is not specific to a substrate, and it’s just a pattern, how can we ever say that something does or does not exhibit the pattern? Can’t I arbitrarily map between objects and parts of the pattern, and say that something is isomorphic to consciousness, and therefore is conscious?”
And my proposal is that maybe it makes sense to talk in terms of something like reference frames. Sure, there’s some reference frame where you could map between grains of sand and neurons, but it’s a crazy reference frame and not one that we care about.
I don’t have a well-developed theory here. But a few related ideas:
simplicity matters
evolution over time matters—maybe you can map all the neurons in my head and their activations at a given moment in time to a bunch of grains of sand, but the mapping is going to fall apart at the next moment (unless you include some crazy updating rule, but that violates the simplicity requirement)
accessibility matters—I’m a bit hesitant on this one. I don’t want to say that someone with locked in syndrome is not conscious. But if some mathematical object that only exists in Tegmark V is conscious (according to the previous definitions), but there’s no way for us to interact with it, then maybe that’s less relevant.
Ahh I see. Yeah, I think that assigning moral weight to different properties of consciousness might be a good way forward here. But it still seems really weird that there are infinite consciousnesses operating at any given time, and makes me a bit suspicious of the computational theory of consciousness.
And my proposal is that maybe it makes sense to talk in terms of something like reference frames. Sure, there’s some reference frame where you could map between grains of sand and neurons, but it’s a crazy reference frame and not one that we care about.
I mean, from that reference frame, does that consciousness feel pain? If so, why do we not care about it? It seems to me like when it comes to morality, the thing that matters is the reference frame of the consciousness, and not our reference frame (I think some similar argument applies to longtermism). Maybe we want to tile the universe in such a way that there more infinitely countable pleasure patterns than pain patterns, or something.
And how does this relate back to realities? Are we saying that the sand operates in separate reality?
It seems to me like when it comes to morality, the thing that matters is the reference frame of the consciousness, and not our reference frame (I think some similar argument applies to longtermism).
For the way I mean reference frame, I only care about my reference frame. (Or maybe I care about other frames in proportion to how much they align with mine.) Note that this is not the same thing as egoism.
My sense is that reference frame for you is something like “how externally similar is this entity to me” whereas for me the thing that matters is “How similar internally is this consciousness to my consciousness.” Which, if the computational theory of consciousness is true, the answer is “many consciousnesses are very similar.”
Obviously this is at the level of “not even a straw man” since you’re gesturing at vague intuitions, but based on our discussion so far this is as close as I can point to a crux.
Hmm, it’s not so much about how similar it is to me as it is like, whether it’s on the same plane of existence.
I mean, I guess that’s a certain kind of similarity. But I’m willing to impute moral worth to very alien kinds of consciousness, as long as it actually “makes sense” to call them a consciousness. The making sense part is the key issue though, and a bit underspecified.
Well, Hamlet doesn’t really exist in our universe, so my plan for now is to not consider him a consciousness worth caring about. But if you start to deal with harder cases, whether it exists in our universe becomes a trickier question.
But if you start to deal with harder cases, whether it exists in our universe becomes a trickier question.
To me this is simply empirical. Is the computational theory of consciousness true without reservation? Then if the computation exists in our universe, the consciousness exists. Perhaps it’s only partially true, and more complex computations, or computations that take longer to run, have less of a sense of consciousness, and therefore it exists, but
He seems to be trying to formalize the intuition about what types of computational consciousness we already intuitively give moral weight to, but the very thing I’m worried about is that our intuitions are wrong (in the same way that our intuitions about physics don’t hold when we think about environments much bigger or smaller than our own).
That is, if the computational consciousness theory is true, and computations with higher complexity feel just as much pain and pleasure and dreams and goals etc as things we normally define as conscious, why should we lower their moral weight?
That makes sense, thanks for clarifying. What I’ve seen most often on LessWrong is to come up with reasons for preferring simple interpretations in the course of trying to solve other philosophical problems such as anthropics, the problem of induction, and infinite ethics. For example, if we try to explain why our world seems to be simple we might end up with something like UDASSA or Scott Garrabrant’s idea of preferring simple worlds (this section is also relevant). Once we have something like UDASSA, we can say that joke interpretations do not have much weight since it takes many more bits to specify how to “extract” the observer moments given a description of our physical world.
That’s why you need to use some sort of complexity-weighting for theories like this, so that minds that are very hard to specify(given some fixed encoding of ‘the world’) are considered ‘less real’ than easy-to-specify ones.
I think that only makes sense to do if those minds are literally “less conscious” than other minds though. Otherwise why would I care less about them because they’re more complex?
It does make sense to me to talk about “speed” and “number of observer moments” as part of moral weight, but “complexity of definition” to me only makes sense if those minds experience things differently than I do.
Description complexity is the natural generalization of “speed” and “number of observer moments” to infinite universes/arbitrary embeddings of minds in those universes. It manages to scale as (the log of) the density of copies of an entity, while avoiding giving all the measure to Boltzmann brains.
Is it an empirical question? It seems more like a philosophical question(what evidence could we see that would change our minds?)
Here’s a (not particularly rigorous) philosophical argument in favour. The substrate on which a mind is running shouldn’t affect its moral status. So we should consider all computable mappings from the world to a mind as being ‘real’. On the other hand, we want the total “number” of observer-moments in a given world to be finite(otherwise we can’t compare the values of different worlds). This suggests that we should assign a ‘weight’ to different experiences, which must be exponentially decreasing in program length for the sum to converge.
But the question then becomes how you sample these minds you are talking to. Do you just go around literally speaking to them? Clearly this will miss a lot of minds. But you can’t use completely arbitrary ways of accessing them either, because then you might end up packing most of the ‘mind’ into your way of interfacing with them. Weighting by complexity is meant to provide a good way of sampling minds, that includes all computable patterns without attributing mind-fulness to noise.
(Just to clarify a bit, ‘complexity’ here is referring to the complexity of selecting a mind given the world, not the complexity of the mind itself. It’s meant to be a generalization of ‘number of copies’ and ‘exists/does not exist’, not a property inherent to the mind)
It seems like you can get quite a bit of data with minds that you can interface with? I think it’s true that you can’t sample the space of all possible minds, but testing this hypothesis on just a few seems like high VoI.
What hypothesis would you be “testing”? What I’m proposing is an idealized version of a sampling procedure that could be used to run tests, namely, sampling mind-like things according to their description complexity.
If you mean that we should check if the minds we usually see in the world have low complexity, I think that already seems to be the case, in that we’re the end-result of a low-complexity process starting from simple conditions, and can be pinpointed in the world relatively simply.
What hypothesis would you be “testing”? What I’m proposing is an idealized version of a sampling procedure that could be used to run tests, namely, sampling mind-like things according to their description complexity.
I mean, I’m saying get minds with many different complexities, figure out a way to communicate with them, and ask them about their experience.
That would help to figure out if complexity is indeed correlated with observer moments.
But how you test this feels different from the question of whether or not it’s true.
I think we’re talking about different things. I’m talking about how you would locate minds in an arbitrary computational structure(and how to count them), you’re talking about determining what’s valuable about a mind once we’ve found it.
I can’t wrap my brain around the computational theory of consciousness.
Who decides how to interpret the computations? If I have a beach, are the lighter grains 0 and darker grains 1? What about the smaller and bigger grains? What if I decide to use the motion of the planets to switch between these 4 interpretations.
Surely under infinite definitions of computation, there are infinite consciousnesses experience infinite states at any given time, just from pure chance.
Suppose that consciousness were not a no-place function, but rather a one-place function. Specifically, whether something is conscious or not is relative to some reality. (A bit like movement relative to reference frames in physics.)
Would that help?
How does this relate back to the example with the sand? Is there a sand-planet reality that’s just like ours, but in that reality the sand is conscious and we’re not?
I don’t think I quite get what a reality is in the function.
I was thinking of the computational theory of consciousness as basically being the same thing as saying that consciousness could be substrate independent. (E.g. you could have conscious uploads.)
I think this then leads you to ask, “If consciousness is not specific to a substrate, and it’s just a pattern, how can we ever say that something does or does not exhibit the pattern? Can’t I arbitrarily map between objects and parts of the pattern, and say that something is isomorphic to consciousness, and therefore is conscious?”
And my proposal is that maybe it makes sense to talk in terms of something like reference frames. Sure, there’s some reference frame where you could map between grains of sand and neurons, but it’s a crazy reference frame and not one that we care about.
I don’t have a well-developed theory here. But a few related ideas:
simplicity matters
evolution over time matters—maybe you can map all the neurons in my head and their activations at a given moment in time to a bunch of grains of sand, but the mapping is going to fall apart at the next moment (unless you include some crazy updating rule, but that violates the simplicity requirement)
accessibility matters—I’m a bit hesitant on this one. I don’t want to say that someone with locked in syndrome is not conscious. But if some mathematical object that only exists in Tegmark V is conscious (according to the previous definitions), but there’s no way for us to interact with it, then maybe that’s less relevant.
Ahh I see. Yeah, I think that assigning moral weight to different properties of consciousness might be a good way forward here. But it still seems really weird that there are infinite consciousnesses operating at any given time, and makes me a bit suspicious of the computational theory of consciousness.
I mean, from that reference frame, does that consciousness feel pain? If so, why do we not care about it? It seems to me like when it comes to morality, the thing that matters is the reference frame of the consciousness, and not our reference frame (I think some similar argument applies to longtermism). Maybe we want to tile the universe in such a way that there more infinitely countable pleasure patterns than pain patterns, or something.
And how does this relate back to realities? Are we saying that the sand operates in separate reality?
For the way I mean reference frame, I only care about my reference frame. (Or maybe I care about other frames in proportion to how much they align with mine.) Note that this is not the same thing as egoism.
How do you define reference frame?
I don’t have a good answer for this. I’m kinda still at the vague intuition stage rather than clear theory stage.
My sense is that reference frame for you is something like “how externally similar is this entity to me” whereas for me the thing that matters is “How similar internally is this consciousness to my consciousness.” Which, if the computational theory of consciousness is true, the answer is “many consciousnesses are very similar.”
Obviously this is at the level of “not even a straw man” since you’re gesturing at vague intuitions, but based on our discussion so far this is as close as I can point to a crux.
Hmm, it’s not so much about how similar it is to me as it is like, whether it’s on the same plane of existence.
I mean, I guess that’s a certain kind of similarity. But I’m willing to impute moral worth to very alien kinds of consciousness, as long as it actually “makes sense” to call them a consciousness. The making sense part is the key issue though, and a bit underspecified.
Here’s an analogy—is Hamlet conscious?
Well, Hamlet doesn’t really exist in our universe, so my plan for now is to not consider him a consciousness worth caring about. But if you start to deal with harder cases, whether it exists in our universe becomes a trickier question.
To me this is simply empirical. Is the computational theory of consciousness true without reservation? Then if the computation exists in our universe, the consciousness exists. Perhaps it’s only partially true, and more complex computations, or computations that take longer to run, have less of a sense of consciousness, and therefore it exists, but
See also https://en.wikipedia.org/wiki/Boltzmann_brain .
Yeah, this has always been my worry as well
Have you seen Brian Tomasik’s page about this? If so what do you find unconvincing, and if not what do you think of it?
He seems to be trying to formalize the intuition about what types of computational consciousness we already intuitively give moral weight to, but the very thing I’m worried about is that our intuitions are wrong (in the same way that our intuitions about physics don’t hold when we think about environments much bigger or smaller than our own).
That is, if the computational consciousness theory is true, and computations with higher complexity feel just as much pain and pleasure and dreams and goals etc as things we normally define as conscious, why should we lower their moral weight?
That makes sense, thanks for clarifying. What I’ve seen most often on LessWrong is to come up with reasons for preferring simple interpretations in the course of trying to solve other philosophical problems such as anthropics, the problem of induction, and infinite ethics. For example, if we try to explain why our world seems to be simple we might end up with something like UDASSA or Scott Garrabrant’s idea of preferring simple worlds (this section is also relevant). Once we have something like UDASSA, we can say that joke interpretations do not have much weight since it takes many more bits to specify how to “extract” the observer moments given a description of our physical world.
That’s why you need to use some sort of complexity-weighting for theories like this, so that minds that are very hard to specify(given some fixed encoding of ‘the world’) are considered ‘less real’ than easy-to-specify ones.
I think that only makes sense to do if those minds are literally “less conscious” than other minds though. Otherwise why would I care less about them because they’re more complex?
It does make sense to me to talk about “speed” and “number of observer moments” as part of moral weight, but “complexity of definition” to me only makes sense if those minds experience things differently than I do.
Description complexity is the natural generalization of “speed” and “number of observer moments” to infinite universes/arbitrary embeddings of minds in those universes. It manages to scale as (the log of) the density of copies of an entity, while avoiding giving all the measure to Boltzmann brains.
Again this seems to be an empirical question that you can’t just assume.
Is it an empirical question? It seems more like a philosophical question(what evidence could we see that would change our minds?)
Here’s a (not particularly rigorous) philosophical argument in favour. The substrate on which a mind is running shouldn’t affect its moral status. So we should consider all computable mappings from the world to a mind as being ‘real’. On the other hand, we want the total “number” of observer-moments in a given world to be finite(otherwise we can’t compare the values of different worlds). This suggests that we should assign a ‘weight’ to different experiences, which must be exponentially decreasing in program length for the sum to converge.
We could talk to different minds and have them describe their experience, and then compare the number of observer moments to their complexity.
But the question then becomes how you sample these minds you are talking to. Do you just go around literally speaking to them? Clearly this will miss a lot of minds. But you can’t use completely arbitrary ways of accessing them either, because then you might end up packing most of the ‘mind’ into your way of interfacing with them. Weighting by complexity is meant to provide a good way of sampling minds, that includes all computable patterns without attributing mind-fulness to noise.
(Just to clarify a bit, ‘complexity’ here is referring to the complexity of selecting a mind given the world, not the complexity of the mind itself. It’s meant to be a generalization of ‘number of copies’ and ‘exists/does not exist’, not a property inherent to the mind)
It seems like you can get quite a bit of data with minds that you can interface with? I think it’s true that you can’t sample the space of all possible minds, but testing this hypothesis on just a few seems like high VoI.
What hypothesis would you be “testing”? What I’m proposing is an idealized version of a sampling procedure that could be used to run tests, namely, sampling mind-like things according to their description complexity.
If you mean that we should check if the minds we usually see in the world have low complexity, I think that already seems to be the case, in that we’re the end-result of a low-complexity process starting from simple conditions, and can be pinpointed in the world relatively simply.
I mean, I’m saying get minds with many different complexities, figure out a way to communicate with them, and ask them about their experience.
That would help to figure out if complexity is indeed correlated with observer moments.
But how you test this feels different from the question of whether or not it’s true.
I think we’re talking about different things. I’m talking about how you would locate minds in an arbitrary computational structure(and how to count them), you’re talking about determining what’s valuable about a mind once we’ve found it.