+1 But there is an asymmetry that may make the comparison invalid. Yes, neurons cannot understand English and Searle can. However, the man in the room, unlike neurons in the analogy, can understand Chinese but does not. Moreover, Searle is conscious. Are we comfortable asserting the room-system is conscious (beyond the man) and can understand Chinese, while the man cannot understand Chinese? The analogy neuron is to brain as man is to room is fraught with additional relevant details that may make it misleading.

+1 Once again, philosophy is fishing out of its depth. We have current science that has highly sophisticated descriptions and models of machine cognition. There is no need to ruminate in an armchair about what is and isn't possible - people are out there doing it now. Philosophy likes to pretend it lives in a vacuum where no other knowledge exists outside of it. The world has moved on. It's not the 18th century any more - if philosophers want to offer an opinion about AI and be taken seriously they need to do their homework first. Nondeterministic algorithms exist.

@JustSomeOldMan But if the point Searle is arguing is that no program can ever have "semantic understanding" like a human brain, doesn't using such a simple model in his argument undermine that? His argument should use a system that is much closer in complexity to the brain, eg a room with a million men in it, who together can understand chinese, or do some even more complex that. Using one man reduces the complexity of the program to such a point that we wouldn't expect understanding to arise at that level, thus making it a bad analogy

-1, begging the question. The whole point of Searle's argument is that his ability to understand the problem cannot be explained as the physical function of his neurons. Your supposed "problem" with his argument, simply assumes the opposite.

@J...: Watson can win at Jeopardy but clearly not understand the questions thenewatlantis.com/futurisms/… The global workspace picture of consciousness points to need for ongoing integration rather than a series of ad hoc rules, which will always sometimes fail without reference to the whole situation. Contrary to what you say, philosophy has never been more important, in figuring out what understanding is - I think of Tegmark's AI physicist, or Deutsch's universal constructors, which are both philosophical models also being actively tested

@CriglCragl Because Watson cannot, therefore any AI thereafter also cannot? What kind of argument is that? As for Max Tegmark and David Deutsch - those are real, actual scientists. High powered experts in physics and machine learning. They are not philosophers. John Searle is a philosopher. He knows absolutely nothing about machine learning.

@J...:The Chinese Room is a classic thought experiment for a reason though, it exactly brings into focus the issues. That's what good philosophy does. We have reason to think no amount of database management will lead to understanding, on it's own. That's not saying semantic understanding is impossible for AI, but there are structural organisational issues to semantic understanding which don't appear to be addressed by simply increasing transistors per chip.

@DavidGudeman Searle is begging the question to begin with, so I'd say it's fair play. The whole argument rests on the assumption that the room's occupant does not understand Chinese and, further, that it is possible for said person to learn to generate natural language responses to questions without concurrently developing an understanding of the Chinese language. The "argument" does not offer any proof that this hypothetical outcome is even possible, nor any logical reasons why it should be. It's a farce of an argument and its conclusion is both pointless and without consequence.

@CriglCragl While it's obvious that what's true of a member isn't necessarily true of the class, I'm more interested in understanding how you rule out that Watson understands. I'm not disputing the claim, but seeking clarification by what justification bolsters the claim.

@DavidGudeman Don't all thought experiments beg the question though? Philosopher A assumes a certain outcome that illustrates his view, while philosopher B assumes a different outcome that illustrates his own view. The reason experiments are so useful in science is because we don't know what the outcome will be ahead of time. That's why we run the experiment. An experiment that isn't actually performed can't really tell you anything, other than what the people arguing about them already think - and we don't need a thought experiment for that.

@gardenhead Consider the utility, though. Searle has certainly not proven anything with his argument - he's offered an opinion loaded with tenuous assumptions. Of what use is his argument? We can either choose to believe that it is true or we can choose to believe that it false. What action should we take in either case? Stop working on AI and give up? Because Searle can't conceive of how semantic understanding can be an emergent property of a complex system? Really? Of what utility is this argument? What problem can we solve with its insights that we could not before? Nothing.

@J... I think you misunderstood my comment. I was basically agreeing with you. This thought experiment doesn't demonstrate anything, except what Searle's personally thinks. And I was saying that's true of most thought experiments

@gardenhead Searle's argument, however, does not generate testable theories. He assumes that meaningful communication can happen between a true intelligence and an artificial one which has no true semantic understanding. What value is there in that assumption? What does it let us test? Searle himself only uses it to argue that the artificial intelligence cannot therefore have true semantic understanding - but this was the very assumption we started out with, so it's not useful at all. It's begging the question.

@J... Not to mention assuming that he is a true intelligence, and that his intelligence comes from something other than the neurons of his brain (and whatever other physical and chemical processes are happening in his body). It's something philosophers have been debating to death for as long as philosophers have existed, and it clearly shows the biggest flaw of how philosophers work - they can never resolve any argument, ever. It's telling that while a lot of good has come from philosophy historically, it always started with dropping philosophical approaches and doing something.

@J... "The whole argument rests on the assumption that the room's occupant does not understand Chinese and, further, that it is possible for said person to learn to generate natural language responses to questions without concurrently developing an understanding of the Chinese language." Are you familiar with how Turing machines work? A universal Turing machine can have pretty simple internal rules and yet, given the right input tape, can execute any arbitrarily complex program. A long-lived person could hand-execute that program without ever needing to know more than, say, 19 rules.

@Hypnosifl Whether the person operating it or the turing machine itself, either way the argument is the same. It rests on the assumption that the Turing machine is capable of generating correct natural language responses without having a semantic understanding of what it was doing. By what measure does he propose we examine such a Turing machine to uncover the absence or presence of semantic understanding? He offers no test - just proclaims that it cannot be with a circular argument. The man in the room is really just a strawman - this is about the machine. The man is just a keyboard cable.

@J... "Whether the person operating it or the turing machine itself, either way the argument is the same." Searle's argument specifically depends on it being a person operating it, see the quotes I gave in my answer. As I said in that answer, I don't think Searle has any convincing rebuttal to the "systems reply" to his argument, but you mischaracterize what he's claiming if you say the fact that it's a human running rules isn't critical to what his argument is.

Also, in my last comment I was specifically criticizing you for trying to cast doubt on the part of Searle's argument that says "it is possible for said person to learn to generate natural language responses to questions without concurrently developing an understanding of the Chinese language". While I think there are other parts of his argument that are bad, this premise of the thought-experiment seems perfectly sound to me, given that a person only needs to learn a few rules to execute any arbitrary program, with no need to understand anything about what the program is doing.

@Hypnosifl The man is irrelevant. Executing simple rules is something that can be replaced with the most trivial of automatons, so there's no reason for the man to be in the room at all. The room is a machine, and the man is just playing a ceremonial role in what would otherwise be an automatic system. Searle again assumes that simply by giving this human a ceremonial role in the operation of the machine that, if the machine had semantic understanding, it should somehow magically appear in the man's brain. Why? It's not even involved. There is no reason to expect this outcome.

@J... Irrelevant to what? The truth about whether the execution of the program would lead to semantic understanding, or the internal logic of Searle's argument? I don't agree with Searle, but I was only talking about the latter question--the fact that it's a human is relevant to his argument, even if his argument is wrong. And I agree with you that Searle is wrong to suggest semantic understanding would magically appear in the human's brain just because they were executing the program, that was the point I made in my discussion about the "systems reply" in my own answer to the OP's question.