Many have seen a glaring problem with Searle’s use of the Chinese Room Argument, namely that it shows too much. If we place the necessary conditions for the mental too high, we run the risk of thereby disqualifying human-mentality. It is with this very idea in mind that Searle embraces a seemingly vague position in which biological matter has some undefined and undescovered capacity to cause mental states, whatever those are for Searle. This appeal to our “causal powers” is essentially the only vangaurd which Searle has against the brain-simulator reply.
The brain-simulator reply to the Chinese Room Argument, it will be remembered, suggests that if we simply program a computer which simulates the neural firings of the brain, down to the last neuron, we simply have no basis for claiming that such a computer does not understand language or is not access-conscious in the same way humans do/are. Searle’s counter-reply is that such a position is to confuse simulation with duplication. Sure a computer can simulate a neural processes down to the last neuron, but since such a program does not involve the same causal powers it does not amount to the actual duplication of mental states. Just as computer simulations of storms cannot get us wet, so too computer simulations of brain cannot be conscious.
This response, however, brings to light an inconsistency in Searle’s thought surrounding neural-networks. After all, connectionist networks are not instantiations of the Chinese Room sicne they are not Von Neumann machines; there is no central processor which reads and writes symbols. Since a neural-net has no symbols, Searle’s Chinese Room Argument simply does not apply.
To this, Searle objects. All neural-nets, he argues, can be run on Universal Turing machines, and therefore are subject to his argument. The problem with this response, however, is that Searle is confusing simulation with duplication. A Universal Turing machine can certainly be said to simulate a neural-net, but it is unclear, at best, that it can actually duplicate one in the relevant sense. In other words, Searle needs to clarify why a neural-net which can be simulated on a Universal Turing machine falls prey to the Chinese Room Argument while a biological brain which can also be simulated on a Universal Turing machine does not.
Of course the most obvious counter-reply would be an appeal to causal powers. This appeal seems to wear very thin within the present context, however. It was precisely because a Universal Turing machine merely simulated rather than duplicated the causal processes in the brain that the causal powers seemed important. Why cannot the neural-net escape the Chinese Room argument since a Universal Turing machine merely simulates rather than duplicates its causal powers? What is it that is simulated or duplicated in one case but not the other?
That’s a good argument Jeff. I think Searle could have a response though. But I don’t know how strong.
First of all, lets make sure we agree on Searle’s belief that a machine could be conscious, that the brain is just a machine. But for a machine to be conscious, it would need to have the right causal account. And we don’t have any idea currently how to figure out that causal account.
So let’s propose the machine, x, which is a complex neural net. We begin training the neural net and eventually are able to tune it to provide human-like responses. Let’s call Cx the causal account of x, and Fx the functional account. The problem is, during the training, we’ve tuned to Fx but Cx is an utter mystery. I don’t know for sure what Searle has in mind with his response, but I’d wager he’s thinking that since the only thing AI research can do is tune machines with unknown Cx to a desired Fx, then simulating x by a von neumann machine is a matter of siumulating Fx which I think we could say can be an exact duplication.
In Searle’s defense, I’d say that if there is no reason believe Cx is constrained by Fx (since it’s an utter mystery), then the probability we’d create a machine with human-like Fx while being luck enough to hit upon the right Cx is almost nil. So Searle is probably just assuming in all this that that Cx is something AI research has nothing to say about. But similar to the neural net, the von neumann machine’s Cx is also a mystery and it’s possible that computers are conscious. The chinese room just tries to make that possibility look stupid. Searle doesn’t know, since Cx is a mystery, that computers aren’t conscious, that silicon and serial processors don’t have the right causal powers for consciousness. All he can say is that syntax isn’t a sufficient condition for semantics, or consciousness. Which is what AI deals in.
Comment by A.G. — February 20, 2007 @ 11:21 am
It is with this very idea in mind that Searle embraces a seemingly vague position in which biological matter has some undefined and undescovered capacity to cause mental states, whatever those are for Searle.
And yet simultaneously deny panpsychism or even limited panpsychism. This always seemed the problem with Searle. He adopts a robust view of consciousness, denies any kind of ontological emergence, and doesn’t want to accord anything like a proto-consciousness to matter. To me there seems nothing else left. It seems to me that Searle’s causal requirements require one of the options he’s already discarded.
I also agree with AG’s point about the mystery which seems to be at the heart of the odd zombie arguments as well. Just because we can conceive of zombies doesn’t mean they are possible. Searle’s approach might be a sophisticate polemic against zombies but it doesn’t really address the underlying issues which is ultimately just the problem of other minds. We assume other humans have a mind because they are close enough. But we don’t assume this about computers because they “seem” different. If AI ever gets discovered (and I doubt it will – but who knows) then our intuitions will simply change as we get used to them.
Comment by Clark — February 22, 2007 @ 1:05 pm
While I thought that my argument was a pretty good one against Searle, but I didn’t think it was all that new. After all, it seems pretty obvious in his exchange with the Churchlands that he focuses on the importance of duplication/simulation when it comes to brains and Von Neumann machines, but when it comes to neural-nets he simply talks about computational equivalence. Personally, I think that this inconsistency shows that Searle is really just trying to protect the exclusive “specialness” of humans. His reaction should have been, it seems to me, a cautious endorsement of neural-nets.
As for A.G.’s counter-reply in Searle’s defense, I simply have no clue what Searle means by “mental” or “causal.” I don’t think anybody does and I think Searle likes this just fine. The secret to consciousness is always somewhere else. The Chinese Room was about the problem of syntax/semantics and how causation gets brought into the discussion is anybodies guess as far as I can tell. In case of neural-nets, however, there is no syntax, only causation. To say that it isn’t the “right” kind of causation is simply blind assertion as far as I can tell.
Comment by Jeff G — February 23, 2007 @ 11:07 pm
Anyways, apparently my professor thought my objection was pretty new and started to look around to see if anybody had made it yet and what, if anything, Searle had said in response. Unfortunately (for me) Copeland beat me to the punch:
http://www.phil.canterbury.ac.nz/personal_pages/jack_copeland/pub/chincym.pdf
Comment by Jeff G — February 23, 2007 @ 11:09 pm
“As for A.G.’s counter-reply in Searle’s defense, I simply have no clue what Searle means by “mental” or “causal.” I don’t think anybody does and I think Searle likes this just fine.”
By causal he means, at least superficially, the medium of implementation. And he does think this has something to do with the chinese room argument. From “…Mystery p.210″
“Any system-from men sitting on high stools….to vacuum tubes…, that is rich enough and stable enough to carry the program can be the implementing medium. All this was shown by the C. room.”
I’m just saying, I don’t think he’d be convinced because no one has any knowledge of the right causal account of the material, so it doesn’t matter if it’s a neural net because AI research revolves around the functional states. In his own words, up in that paragraph:
“All causal powers are in the implementing medium..But the operation of the program has to be defined totally independently of any implementing medium” My Cx and Fx.
Comment by AG — February 24, 2007 @ 9:14 pm