Comment by jerf | Hacker News Reader

Comment by jerf | Hacker News Reader

In his paper "Why Philosophers Should Care About Computational Complexity" [1], HN-favorite Scott Aaronson resolves to my satisfaction the question of whether a waterfall can be said to be computing the solution to some problem as it cascades over the rocks, because there exists a mapping in which the initial water state is the initial state of the problem, the gyrations of the water "compute", and the final state of the water can be mapped to a solution. He points out that we can with some actual mathematical rigor observe that the mapping itself can be said to be doing all the work.

Similarly, I feel trying to parse conscious intentionality by any known meaning of the term "conscious" out of something like "all the individuals of the United States" is a situation where the mapping is doing all the work.

Clearly, there is something to the "United States" as well as other groupings of human beings. But I daresay in a lot of ways these things are less mysterious than they seem... we deal with these groupings all the time, and if we thereby fail to ascribe consciousness to them, I'd say our experience should probably be listened to. Sure, groups of humans routinely do great things no individual could do, but at the same time, and with no contradiction, groups of humans fail miserably and stupidly too. We've all seen committees that fail to successfully accomplish a goal than any individual on the committee could have, or where the committee remains fuzzy on its purpose (pardon English's anthropomorphization, there) even as the individual members are all quite clear (but divergent). Rather than trying to throw this in the "consciousness" bucket I think it's better understood as, well, the way we all already collective conceive of these things, as human organizations, with their own foibles, characteristics, and properties.

It's not consciousness. It's something else. "Greater" in some ways, and yet, profoundly lesser in others. Trying to view it through the lens of "consciousness" is just anthropormorphism striking again, and I'd say actively harmful in terms of trying to understand the phenomena.

[1]: http://www.scottaaronson.com/papers/philos.pdf

>HN-favorite Scott Aaronson resolves to my satisfaction the question of whether a waterfall can be said to be computing the solution to some problem as it cascades over the rocks, because there exists a mapping in which the initial water state is the initial state of the problem, the gyrations of the water "compute", and the final state of the water can be mapped to a solution. He points out that we can with some actual mathematical rigor observe that the mapping itself can be said to be doing all the work.

This is why panpsychism appeals to me. On some level it makes more sense for me to believe that sentients is an intrinsic property of physical reality then to believe that sentients can emerge from someone moving sticks and stones around according to some algorithm.

The possibility of sticks and stones, their motions, and algorithms are already an intrinsic property of reality, so the things you want already exist in the model you reject.

The thing I "want" is sentience. It's the Qualia problem.

That exists regardless of what you think about it.

You're not making much sense to me.

Maybe sentience is intrinsic to change (someone moving sticks and stones around according to some algorithm). Have you read Giulio Tononi's "Integrated information theory of consciousness"? One of its conclusions is that panpsychism must be true.

Maybe, I think change is a very abstract entity to apply properties to though. No I haven't seen it before.

> He points out that we can with some actual mathematical rigor observe that the mapping itself can be said to be doing all the work.

There is no mathematical rigor whatsoever in that portion of the paper. He just conjectures that there would be no efficient algorithm for encoding (e.g.) chess positions as states of a waterfall. This actually seems fairly unlikely. You only need to be able to specify a set of physical states that compose in such a way that you can, e.g., incrementally push items onto a stack, and you're basically done. How can we possibly be confident that there is no state description of a waterfall that makes that possible? Note that we have to allow the state descriptions to be very complex, since the physical states corresponding to the computational states of a microprocessor are also extremely complex. This complexity happens to be easier for us to deal with because microprocessors have been designed to make it easy for us to put them into particular computational states. But the relevant states of a waterfall, while much more difficult for us to manipulate, will not obviously be any more complex. And to make the key point, we need only find one suitable inanimate physical system. It really doesn't seem so unlikely that there are a few of them out there.

On top of all this, why should it matter if there's no efficient encoding algorithm? It would obviously lead to an infinite regress if we say that a physical state has to have been "encoded" in order to count as a bona fide computational state (since then the input to the encoding algorithm would itself need to have been encoded).

Right - there's also no efficient algorithm for encoding chess positions into the electrical and chemical states of a couple of pounds of active neurons, and yet living humans seem to be able to play chess.

"He just conjectures that there would be no efficient algorithm for encoding (e.g.) chess positions as states of a waterfall. This actually seems fairly unlikely."

I think you're answering a different question than Scott. You seem to be answering "Can I construct a water-based computer that looks like a waterfall that could solve a chess problem, set up the initial parameters, and read the answer off the bottom?" This is conceivable. It is likely it would involve a highly implausible object of highly implausible accuracy and reliability, but if such an object existed it would still be a polynomial calculation to produce it. (Or, at least, a polynomial calculation could produce an object, if not the optimal one. The lack of solution to the Navier-Stokes equations might prevent you from finding an optimal one. And you might find yourself having some trouble with quantum effects. But such are the problems in the real world.) And our polynomial approximation is still not going to look like a waterfall as it will inevitably need looping constructs where state flows "back" uphill... if you want it to be one-way like a waterfall that is going to be exponentially large.

Scott's question is a different one. Given a real world waterfall that you have not constructed, but simply come across, can it be said to be computing the solution to a chess problem? Is a random rock that is sitting on the ground, jiggling away with atoms in constant motion that all the computers in the world could not possibly provide enough computational power to fully simulate, able to be said to be computing something with all that power? Could a boulder be sitting there simulating a human mind? I don't mean in the pan-psychic sense, I mean, literally, is it simulating a human mind? It has the computational capacity, when considered in the raw.

This is where Scott's argument kicks in, where the mapping is doing all the work.

Further evidence of this is that even if you want to sit here and argue that this boulder that is sitting in front of us (metaphorically) really and truly is calculating the state of your mind as well, well, with another equally sensible (and exponential and impossible) mapping, it's also calculating mine, it's also calculating Attila the Hun's, and it's also calculating the brain states of the final human being to ever live, and also deer's brains and the solutions to world hunger and pretty every other interesting thing ever, really. The contortions required to create the mapping of boulder state to human brain state are such that you can fit literally almost anything into it, and therefore, it is reasonable to point out that it is meaningless. It's an interesting argument that provides a surprisingly rigorous line that allows us to say that, no, that waterfall is not solving a chess problem or anything else... it really is, well, a waterfall.

> Given a real world waterfall that you have not constructed, but simply come across, can it be said to be computing the solution to a chess problem?

Yes, that's the question I was addressing. I'm really can't figure out why you thought otherwise. There's nothing in my original comment about constructing artificial water-based computers.

>This is where Scott's argument kicks in, where the mapping is doing all the work.

It is his conjecture that there is no computationally simple mapping, but I see no reason to believe that this conjecture is correct. He certainly doesn't give any reason to think so in the paper. And to echo jameshart's point, he gives no reason to think that the mapping from chess positions to brain states would be any simpler than the mapping from chess positions to waterfalls.

If you find his conjecture plausible for some reason, that's fine, but you're quite wrong if you think that the paper presents any rigorous mathematical argument. The discussion is much less rigorous and sophisticated than the existing philosophical literature on the topic. (After all, Putnam's original paper contained a proof.)

>Further evidence of this is that even if you want to sit here and argue that this boulder that is sitting in front of us (metaphorically) really and truly is calculating the state of your mind as well, well, with another equally sensible (and exponential and impossible) mapping, it's also calculating mine, it's also calculating Attila the Hun's, and it's also calculating the brain states of the final human being to ever live, and also deer's brains and the solutions to world hunger and pretty every other interesting thing ever, really.

Yes, that's the problem. That's why the computational theory of mind seems to be a bit of a non-starter. Since pretty much every physical system computes pretty much every function, it's difficult to see how thought can arise merely as a consequence of the brain computing a particular function.

Agreed. The human tendency to anthropomorphize what it should not has enabled priests and witch doctors of countless varieties, in many places and for thousands of years to make a living without producing except for the propagation of misunderstanding. This ignoble tradition is apparently being carried on in the philosophy department at the University of California.

A bunch of me will arrive at consensus faster than a bunch of a whole lot of different people. I'm all the same, after all, so it's easier to come to some common agreement about what this is. Ask a million people what that is, they'll tell you a thousand different answers.

But if you see this as anthropomorphizing then you're not a materialist, or you'll have to show some qualitative, essential difference between the US and the neurons in your brain. There is nothing to anthropomorphize because there's nothing special about the anthropos: the two are either examples of the same phenomenon or they're not.

You sort of ironically demonstrate the exact sort of antropomorphization I am decrying, which is the casual (as in, unexamined) assumption that the only complex state which can exist is human consciousness (or, if you prefer, "consciousness, which is the same thing that humans have"). If in fact there are other complex states that can exist, then "anthromorphization" is the act of forcing the human state onto those other possibilities. The space of complicated possibilities is much richer than the space of simple ones, and there is no a priori reason to believe that just because something is complicated, and that it has some things vaguely human-consciousness about it, that it is therefore human consciouness. Just because a 58-dimensional object contains some portion of itself that is a spherical section does not mean you have a 58-dimensional sphere.

I feel justified in referring to "the" human state because in the grand possibility space of complex systems, human brains are but a point at scale, in much the same way that our little planet is just a dot.

Any definition of "consciousness" that covers all possible complex systems is just another word for "complex system", and therefore useless.

It isn't for me to have to prove that the quadrillion-dimensional-system of the United States is different than the billion-dimensional-system of the human consciousness [1], with wildly differing characteristics between the relationships of those dimensions... I'd say it's the other way around, that those who think it's all the same ought to give a compelling reason as to how it could possibly be the same thing, because on the fact of it it's actually even more mathematically absurd than common sense would suggest. Common sense still has that anthropomorphization built into it, deceiving you, and leading to the ideas of "pan-psychicism" where the only thing we can conceive is consciousness-as-we-know-it so the only question is how much of it something can have.

[1]: My selection of the dimensionality of the human is arbitrary; my selection of the United States is obtained by simply multiplying people x human states. This is a brutal underestimate because the humans are interacting too and that creates further dimensionality, but, meh. It's really just "big numbers".

Scott Aaronson directly addresses the integrated information theory of consciousness in his this blog post: http://www.scottaaronson.com/blog/?p=1799

Reply on news.ycombinator.com