https://en.m.wikipedia.org/wiki/Chinese_room
Highly recommend trying to grasp the argument, because in a sense it connects complexity to ‘time to compute’. If it takes a room of robots following simple instructions a billion years to act as an AI that reads Chinese, are those robots a collective intelligence? Can it be scaled up to consciousness?
It's not really, IMO, all that valuable.
> because in a sense it connects complexity to ‘time to compute’.
No, it doesn’t. It basically just rejects by definition (and then presents an illustration which is a useful support if you accept its definition or an illustration that its definitions are wrong if you don’t) that a programmed response system that exhibits the behavior of an actor that understands material actually understands material, and from that rejects the idea that brains (which produce minds, which have actual understanding) are equivalent to programmed response systems.
Its been a little while since I've read the paper, but I don't think it's right to say that it rejects this conclusion by definition. Instead, it presents a situation where the reader's intuitions are intended to rebel at the thought that the system in question understands the symbols that it is manipulating. But yes, otherwise I agree with you that the argument then goes that, if this case has caused you to be skeptical about whether a symbol manipulation system could be conscious, you should be skeptical about whether other ones would be as well.
There are, of course, some important limitations/objections. Two that come to mind are that 1) you might not think that the reader's intuitions about a situation have any metaphysical significance--they might just be wrong and 2) the Chinese Room example might just suggest that a particular kind of symbol manipulation system may not be conscious without supporting the much stronger claim that they can't be conscious.
The fact that one can raise objections, though, doesn't mean that it isn't an important contribution to the philosophy of mind. If nothing else, I think Searle was right to point out that the example raises interesting questions about the usefulness of the Turing Test since, by hypothesis, the Chinese Room could pass it.
The other part of the thought experiment, apart from imagining the room, is to imagine how complex the turing test might be.
When people imagine what a Turing Test conversation would look like, they frequently underestimate the conversation. I find Dennet's example of an imaginary Turing Test from Consciousness Explained to be a good counterexample:
Judge: Did you hear about the Irishman who found a magic lamp? When he rubbed it a genie appeared and granted him three wishes. “I’ll have a pint of Guiness!” the Irishman replied and immediately it appeared. The Irishman eagerly set to sipping and then gulping, but the level of Guiness in the glass was always magically restored. After a while the genie became impatient. “Well, what about your second wish?” he asked. Replied the Irishman between gulps, “Oh well, I guess I’ll have another one of these.”
CHINESE ROOM: Very funny. No, I hadn’t heard it– but you know I find ethnic jokes in bad taste. I laughed in spite of myself, but really, I think you should find other topics for us to discuss.
J: Fair enough but I told you the joke because I want you to explain it to me.
CR: Boring! You should never explain jokes.
J: Nevertheless, this is my test question. Can you explain to me how and why the joke “works”?
CR: If you insist. You see, it depends on the assumption that the magically refilling glass will go on refilling forever, so the Irishman has all the stout he can ever drink. So he hardly has a reason for wanting a duplicate but he is so stupid (that’s the part I object to) or so besotted by the alcohol that he doesn’t recognize this, and so, unthinkingly endorsing his delight with his first wish come true, he asks for seconds. These background assumptions aren’t true, of course, but just part of the ambient lore of joke-telling, in which we suspend our disbelief in magic and so forth. By the way we could imagine a somewhat labored continuation in which the Irishman turned out to be “right” in his second wish after all, perhaps he’s planning to throw a big party and one glass won’t refill fast enough to satisfy all his thirsty guests (and it’s no use saving it up in advance– we all know how stale stout loses its taste). We tend not to think of such complications which is part of the explanation of why jokes work. Is that enough?
Dennett: “The fact is that any program that could actually hold up its end in the conversation depicted would have to be an extraordinary supple, sophisticated, and multilayered system, brimming with “world knowledge” and meta-knowledge and meta-meta-knowledge about its own responses, the likely responses of its interlocutor, and much, much more…. Maybe the billions of actions of all those highly structured parts produce genuine understanding in the system after all.”
No, because of the article you linked to. Chinese room argues AGAINST the existence of consciousness/intelligence arising from following symbolic instructions.
Here is a quote I like by John Searle:
> Computational models of consciousness are not sufficient by themselves for consciousness. The computational model for consciousness stands to consciousness in the same way the computational model of anything stands to the domain being modelled. Nobody supposes that the computational model of rainstorms in London will leave us all wet. But they make the mistake of supposing that the computational model of consciousness is somehow conscious. It is the same mistake in both cases.
But first you have to establish this is true.
More intuitively: our words are things which do things in the world. When we acquire them we do so by being embedded in the world. When we use them we use them to change the world. (Largely).
If a child in a cupboard reading a textbook out-loud could fool a chemist that the child understood chemistry, it doesn't mean they do. In the end, you need to turn on that bunsen burner.
The chinese room needs to be taken super-literally: does the person just doing a "hash-table lookup" on "the right answers to pre-prosed questions" actually understand.
To me its incredibly obvious that they do not. This does not establish that, eg. cognition, does not involve computational hastable-like processes. I think it does establish that propontents of a naive computationalism have an incredibly big challenge.
The challenge is to specify the environment computationally, since that is the only means by which the computer itself can be sufficiently embedded and actually understand anything.
The issue here is that if "computer" isnt just a trivial property it has to include, eg., measurability, determinism, etc. Properties that reality lacks. Reality isnt a computer.
Starting with the chinese room one /can/ supplement sufficiently until we reach searle's conclusion.
Or is there such a phrasing?
You "actually understand" if in saying words, w, about a situation, s, the reason, r, you say those words counts as as an understanding of s.
So if you're asked, "what is the weather today?" and a system replies, "its sunny" -- it only understands if the reason it replied was because its saying "sunny" had come about because it understands sunny sitations; and therefore has a justfiied reason on this occasion to say that this occasion is sunny.
If I ask a NN trained on a trillion documents, "do you like what I'm wearing?" it cannot answer. It can say the words, "yes I do!" but it cannot have a reason to say those words. It's just a "weighted average retrievla across a compressed dataset of a trillion documetns".
The question isnt "on average, what -- historically -- would a generic person reply to the question: do you like what i'm wearing?"
The question is if the systems likes what I'm wearing. Its replies are only "actual replies" if it can see what i'm wearing, has experiences/preferences for taste, has some aesthetic judgement, has a disposition to like/dislike -- etc.
No ML system on the planet, in this sense, has any understanding whatsoever. Interpolation over historical data is just a means of compressing history into parameters, ie., its a compressed lookup table. That isn't ever a /reason/, exactly, if a person simply used such a table -- they wouldnt mean what they said
Since my liking you is caused by my immediate environment, it isn't reducible to a weighted average of my history.
Another way of putting it: the historical positions of all the molecules in some water aren't sufficient to determine its present state. It's state depends on its container (ie., the pressure & temp of its environment). And there are a very very large number of states of water, many still being discovered.
In this sense my state in any moment is a point in an infinite space of states -- not determined by my history. But also extremely complexly by my container -- my social, etc. environment. The world hitting my senses is doing more to me than the air on the water. It induces in me a state which cannot be "averaged" from my history.
Thus, no, we are not weighed averages of our histories. We are profoundly chaotic and organic organisms whose growth in our environments enables us to respond to our enviroments by entering a near infinite number of states. These states arent in our history, they are how our biophysical structure -- via history -- responses to the near infinite depth of the here-and-now.
We are more like water than a computer. A computer is a deterministic machine which is a deterministic function of its deterministic inputs. Water is a chatoic system whose state "isnt up to it". Water's state is /in/ its container, and water itself is a non-determinsitic chatoic soup.
The chaos of water is the least of what one nanometer of a cell has; a cell is a trillion times that adaptive and responsive. And we are a trillion of those.
We are a cascade of chaotic state changes provoked by an infinitely rich environment action on our bodies shaped by a long history of organic growth.
We don't have "neural networks", we have cells. That some form "networks" has nothing to do with what we are. A complete misdirection.
It is not clear to me that this cannot be the case of a weighted average very heavily weighted to the immediate past.
> The historical positions of all the molecules in some water aren't sufficient to determine its present state. It's state depends on its container (ie., the pressure & temp of its environment).
AFAIK, given that you could determine the momenta of the molecules from a history of their positions, this would be sufficient to determine its state (maybe you need their angular momenta independently?) The relevant information about the container has been impressed on the motion of the molecules.
Similarly, we can suppose that the history of your environment becomes manifest in your mental states (and a predisposition towards certain state transitions) - though, on account of the complexity of that environment, in a compressed form.
> We are more like water than a computer. A computer is a deterministic machine which is a deterministic function of its deterministic inputs. Water is a chatoic system whose state "isnt up to it". Water's state is /in/ its container, and water itself is a non-determinsitic chatoic soup.
And yet we can usefully model fluid dynamics on a computer, even though the mathematical representation of the problem is analytically intractable. This line of argument does not appear to be leading in the direction you think it does.
> We don't have "neural networks", we have cells. That some form "networks" has nothing to do with what we are. A complete misdirection.
I am generally distrustful of these ontological arguments - quite often, it seems, things that were once thought of as being completely different turned out to be similar in some relevant way.
Without your trained history, you would not be able to say "it is sunny" when the sun hits your skin. We know this for sure, as an elephant or a baby are unable to say it, while a Frenchman would say different words.
Me saying "I like you" is not caused by me liking you. It is caused by me deciding, based on my experience and my perception of the current situation, to utter these words. I may be lying, I may be exaggerating, I may be misdirecting. Human speech is extremely rarey caused directly by interaction with the (external) environment (exceptions would be onomatopoeia exclaimed when in extreme pain, or when surprised, which tend to bypass most computation in the brain).
See my comment to the other reply, an others in this thread about chaos/non-determinism to see why I doubt.
Ie., I dont think our organic growth and adaption to our environment, in being profoundly chaotic (and via QM, thef. non-deterministic) is likely to be describable as a computable function.
Conversely, you can interpret QM to mean that the universe evolves perfectly deterministically, except that no particular observer can predict in which branch they will find themselves.
Also, non-determinism has little to do with computability. All known rules of physics are computable, even if you include the Born rule for non-determinism. A quantum computer is computable by a Turing machine (in exponential time, as far as we know).
You've a bigger issue with talking about the computability of those laws.
I am assuming that somehow the reals can be ditched (which, incidentally, seems unlikely). And in that reality, we still have a problem: the role the reals play is to make reality infinitely precisely measurable. Without this chaotic systems become non-deterministic over realtively short time horizons.
Nearly all actual systems are chaotic in many of their properties. It isnt a matter of just adding an RNG to a CPU. A trivial object has 10^30 atoms, in organic stuff, much of that is behaving chaotically on environmental boundaries. The engery to simulate this, assuming 10^30 RNGs (etc.) is vastly greater than the target system and "cannot outrun" that system.
Not only this but I think simulation of such things quickly surpasses physical bounds on physically possible computers, not least as the RNGs have an energy cost.
When talking about computability in the absolute, theoretical sense, the size of the universe is irrelevant - all we ask is if there exists some arbitrarily large, but finite, Turing machine that performs that computation and then halts after an arbitrarily large but finite amount of time.
If we're taking about practical computability, than my base assertion that the physical world is itself a computer also implies that this computer is exactly fast enough to simulate physics. Since we know that the laws of physics are computable in the theoretical sense above, and since we know that the universe is accurately simulating itself, we have proven my assertion.
Your other points about chaos and non-determinism I have addressed in another thread.
[0] https://stackoverflow.com/questions/4134282/is-pi-a-turing-c...
But this is Searle's sleight-of-hand. The question isn't whether the person in the Chinese room understands Chinese, but whether the system as a whole understands Chinese. The man is analogous to the CPU in a computer, but the CPU is embedded in a larger system and it is the entire system that performs operations of the computer. You can't focus on one component to the exclusion of the rest of the system and expect to derive valid conclusions about the system.
>The challenge is to specify the environment computationally
I agree, but this presents no in principle challenge to a computational system reaching such a detailed description of the environment.
>The issue here is that if "computer" isnt just a trivial property it has to include, eg., measurability, determinism, etc. Properties that reality lacks. Reality isnt a computer.
I don't follow. The issues of measurability and determinism in QM (assuming that's what you are referring to) aren't necessarily problems at the scale of physical systems. Computational systems can certainly make distinctions to within some margin of error, which is sufficient as a substrate for gathering information about the external environment.
> at the scale of physical systems.
Alas, they are. Even classical mechanics isn't deterministic. Most systems are chaotic and require infinite precision in measurement to be deterministic. Since QM precludes this, most chaotic systems literally can only be predicted over short horizons.
Eg., there is moon in our solar system, i believe, with a chaotic orbit. At maximum possible measurement fidelity we can predict 20yrs.
Consider that all the particles of our body are just these little chaotic moons, and their aggregate chaos dramatically dimishes this 20yr horizon.
There is, quite literally, not enough information present in this moment to predict the next moment. Most systems are sufficiently chaotic in a sufficiently large number of parameters to preclude this. This reaches all the way up to classical scales.
Reality isnt a computer.
Put another way, Searle separates a "computer" from the instructions it is running, but there is no such separation. It's like separating a brain from the electrical brain activity. It's true that a brain without electrical activity cannot understand Chinese just like a computer without a program cannot, but that's not a very interesting observation.
If we all agree that the function the man performs can be set to be whatever you wish (ie., any {input sentences} -> {output sentences}) and we agree that the man doesnt understand chinese... then the question arises: what is missing?
I dont think the "systems reply" actually deals with this point -- it rather just "assumes it does" by "assuming some system" to be specified.
If using a language isnt equivalent to a finite computable function -- what is it?
The systems reply needs to include that the reason a reply is given is that the system is caused to give the reply by the relevant causal history / environment / experiences. Ie., that the system says "it's sunny" because it has been acquainted with the sun, and it observes that the day is sunny.
This shows the "systems reply" prima facie fails: it is no reply at all to "presume that the system can be so-implemented so as to make the systems reply correct". You actually need to show it can be so-implemented. No one has done this.
There are lots of reasons to suppose it can be done, not least, that most things arent computable (ie., via non-determinism, chaos, and the like). Given the environment is chaotic, it is a profoundly open question whether computers can be built to "respond to the right causes" and computational systems may be incapable of doing this.
If they cannot, then searle is right. That man, and whatever he may be a part of, will never understand chinese. It is insufficient to "look up the answers", "proper causation" is required.
Edit: He would speak Chinese, but he wouldn't understand it. What is missing in order to understand Chinese is an additional program that translates Chinese to English or pictures or some abstract structure independent of language. Humans have this, but the function in this thought experiment is a black box, so we don't know if it uses an intermediate representation. Thanks to colinmhayes for pointing this out.
Could he give a reasoned answer to "what is the weather"? That depends on whether we include external stimuli as part of the function input. If not, then neither a human nor a computer could give a sensible answer beyond "I don't know".
I see now that your issue is really with the function - could a function ever exist that gives responses based on history/environment/experience. My understanding is that such a function is the premise of the thought experiment; it's a hypothetical we accept in order to have this discussion. Searle claims that even if such a function exists, the computer still doesn't truly understand. But if that's what we're asking, my answer would be yes, as long as history/environment/experience are inputs as well. Of course a computer locked away in a room only receiving questions as input can never give a real answer to "what is the weather", just like a human in that situation couldn't. But if we expand our Chinese Room test to include that type of question and also allow the room or computer to take its environment as input, then it can give answers caused by its environment.
> You actually need to show it can be so-implemented. No one has done this.
I mean, fair enough. It's fine to say "I won't believe it until I see it", but that pretty much ends the discussion. If we want to talk about whether something not yet achieved is possible, then we need to be willing to imagine things that don't exist yet.
I don't necessarily disagree, but it's not so simple. Just because the man speaks chinese doesn't mean he understands it. He could have figured out the proper output for every possible input without knowing what any of it means. If the next person asks in English "what did you talk about with the last person?" what would he answer, assuming he also speaks english? Really the question comes down to if the computer is able to write the book by observing I guess, but even then you could conceive of a different book with instructions on how to write the translation book.
And i just wanna pour us all a cup of tea and sit with that.
It reminds me of the song "anything you can do, i can do better, i can do anything better than you - No you cant! Yes i can!.."
The issue is- tell me what it cant do. Explain to me what the chinese room is not understanding. Whatever model you use to translate the missing functions can then be "encoded".
Id like to think my cpu has a brain and it goes off dreaming with the idle threads- but im also quite convinced that every single thread that runs machine code is a non-concious direct representation of the mental model i encoded into its functions.
No, the function would.
My CPU doesn't speak HTML. My web browser does. Does running my web browser mean the CPU speaks HTML, even if the whole browser is loaded into cache? I don't think it does; the CPU is still running machine code.
If I memorise a Turing machine, does that mean I understand the computation it's performing? No; I'd have to pick it apart and work out what each bit does, then put that meaning back together, in order to try to work out how it works.
Memorising the function would enable the man to teach himself Chinese (whichever language “Chinese” is), just like memorising a good mathematics would allow most people to teach themselves the concepts therein. But memorisation isn't understanding.
To be clear, the point of the systems reply isn't to demonstrate how a computational system can understand language, it is to point out a loophole such that computationalism avoids the main thrust of the Chinese room.
>If using a language isnt equivalent to a finite computable function -- what is it?
In the Chinese room, the man isn't an embodiment of the computable function (algorithm). The man is simply the computational engine of the algorithm. The embodied algorithm includes the state of the "tape", the unbounded stack of paper as "scratch space" used by the man in carrying out the instructions of the algorithm. So it remains an open question whether the finite computable function that implements the language function must "use language" in the relevant sense.
What reason is there to think that it does use language in the relevant sense? For one, a semantic view of the dynamics of the algorithm as it processes the symbols has predictive power for the output of the system. Such predictive power must be explained. If we assume that the room can in principle respond meaningfully and substantively to any meaningful Chinese input, we can go as far as to say the room is computing over the semantic space of the Chinese language embedded in some environment context encoded in its dictionary. This is because, in the limit of substantive Chinese output for all input, there is no way to duplicate the output without duplicating the structure, in this case semantic structure. The algorithm, in a very straightforward sense, is using language to determine appropriate responses.
Kripke's quaddition vs. addition distinction is also a good point in this regard.
As for your first claim, it appears to be begging the question, predefining, without justification, 'real' computation as being something that only humans can do, while machines can only simulate so doing.
"The system" isn't the kind of thing that understands. If I give n people a letter each and a number indicating the position the letter should occupy on some grid, it does not follow that the group knows the sentence because groups are not knowing subjects. If they arrange the letters on the grid, then each individual person in the group now knows the sentence when they read it.
Only individual knowing agents know and understand. Abusing the definition of "knowing" or "understanding" doesn't help clarify anything. All it does is lead to equivocation.
Firstly, it is a circular argument in which an unargued-for intuition - '"The system" isn't the kind of thing that understands' - is used in trying to justify that intuition.
Secondly, no-one is suggesting, and Strong AI certainly does not imply, that the utterly simple 'system' here is capable of understanding anything. The fact that it does not proves nothing.
Thirdly, it is irrelevant (a variant of the homunculus fallacy) to see any significance in what the actors understand and when they do so, as they are being employed to act as unthinking automata. Replace each of them with a simple mechanism for placing a letter in a predetermined spot, and you have an equivalent system, also mindless, that is equally useless for demonstrating anything about the feasibility of Strong AI.
That’s not how it’s phrased. The Chinese Room needs to pass the Turing Test with arbitrary input of Chinese characters. So a convincing Chinese Elissa basically.
> The issue here is that if "computer" isnt just a trivial property it has to include, eg., measurability, determinism, etc. Properties that reality lacks. Reality isnt a computer.
Reality isn’t a human brain either.
Or is it?
Either way I don’t see how that is relevant.
I don't see why a machine can't be said to "know" a language; after all, it can translate languages. I can translate from French to English (I'm a native English speaker); but there are sentences in English that I can't "understand", either because I don't really know what all the words mean, or because the concept expressed by the sentence is beyond me.
The person doesn't, but the (person + hash-table) system does. This is not weird - presumably whoever wrote the hash-table did understand Chinese. The whole point of using a room for this thought experiment is that we're no longer talking about a person, we're talking about a "machine" consisting of a person, a hash-table, and some kind of input/output system.
The hypothetical computer is the same. The CPU does not understand Chinese, but the whole computer as a system does, because part of that system is actual knowledge that came from someone that understands Chinese. When people ask if a computer is intelligent, they are not asking about one particular component, they're asking about the computer as a whole. Just like when you talk about a human's knowledge and abilities, you don't ask about particular sectors of their brain, you ask about the person as a whole.
I agree that the person is frivolous, that's why it's strange that Searle asks whether the person in his thought experiment understands Chinese. That's irrelevant, what matters is whether the room as a whole does, and clearly it does.
Why did I have to add something, if the room already understood Chinese before? Because we're essentially adding a second program now and trying to share understanding between them. The room did understand Chinese, but that understanding will not be accessible to a new component unless we design with that in mind. An analogy would be asking a person "which muscles did you use to digest that food?" Clearly some part of the human-system knows this, because it activated the right muscles and successfully digested the food, but the part of the system that hears and responds to English doesn't have access to that knowledge. We would need to redesign the system and programs in order to share understanding between these different parts of the system.
I realize we were using the term "hash-table" in this thread and what I'm talking about wouldn't be possible with just tables, but such a simple program is not a requirement for the thought experiment. The idea is we just have some black box function that takes inputs and produces outputs and we don't know how it works.
You may recognize that, by distinguishing between the operator and the room, I am prefiguring the so-called "systems reply", which is that the system as a whole is demonstrating whatever level of understanding is manifest here, and there is no assumption, in so supposing, that this understanding is manifest in any single component, even if one of those components has the ability to understand some things on its own. By deliberate construction, Searle is using the room's operator to mechanistically implement an algorithm, a task that does not require knowledge of the language in which the questions are posed (in fact, he is using the language barrier to prevent the operator answering the questions himself.) Searle has not presented an argument for the operator needing to do anything more, and it is obvious that people can be trained to perform tasks they do not have any understanding of beyond the specific actions involved.
Searle's attempt to refute this reply seems to show that he cannot even conceive it fully, as that response is to modify the scenario such that the operator memorizes the rule book, as if where it is stored makes any difference.
Searle's stance here seems to me to be a version of the homunculus fallacy, supposing that our minds are implemented by (as opposed to being) a conscious entity.
A hashtable lookup cannot be what it is to "act in the world with words".
When I say, "do you like what I'm wearing?" -- regardless of what has been written in any book, at any time, I do not want that reply.
I may want those words. But the reply I want is to use your judgement of taste and experiences of the world to tell me your opinion. The words dont answer the question if the reason they are used is incorrect.
A hashtable lookup can, basically, never be a reason to use words.
And hence the Chinese room reveals much much more than simply a circular point. And the systems reply fails to say much much less than it needs to.
The systems reply needs to explain "by what system" the system's use of language counts as a use. "by what system" words are born of understanding, not "mere reterival".
To do this, you need to specify the whole world, experience, etc. as "computational systems". And thus the systems reply simply fails against this example. It isnt sufficient to say "maybe" you have to say, "and this is how!".
And worse, the world is clearly not a computer -- as any non-trivial definition (eg., universal turing machine) cannot simulate properties the world has (non-determinism, chaos, etc.).
Nor is Searle arguing that the Chinese room, if it could be implemented, must use language (beyond the capabilities that he does require of it, which is answering questions from an ill-defined domain.) And as far as I can tell, he does not require it to, as you somewhat vaguely put it, "act in the world with words" beyond this.
The systems reply does not need to do anything more than show that Searle is making some unargued-for assumptions about what Hard AI would imply - and Searle's response to it so completely misses the point of that reply that he is actually helping make it clear! (For more details, see the second paragraph of my previous post; as far as I can tell, no argument has been raised against the points I made there.)
The limitations of Turing machines are similarly beside the point. A computer, such as those involved in the composition and delivery of this reply, is not just a Turing machine (not even a universal one), even though it is capable of implementing one (up to the physical limits of memory). In particular, a computer (as opposed to a Turing machine) can incorporate environmental inputs, including physically-derived entropy, and can use these in simulating physical processes.
You are evidently fond of the phrase "[Reality/the world] is not a computer"; that may be so, but rather more definitively, I think, one can say that the mind is an information processor.
I understand that it seems inconceivable to you that a mind could be produced by a suitably-programmed computer of any power, and I am aware that I can only offer hand-wavy arguments for that proposition. I think you are mistaken, however, to read Searle's Chinese Room argument as being about, let alone supporting, the broader intuitions you have expressed in your reply to me and to others. Searle, who is, after all, a philosopher of considerable stature, is well aware that you need something more than just one's intuitions to make a respectable argument; that alone would not pass peer review.
This is precisely wrong. With the possible exception of consciousness, in our current understanding of physics, the world is a computer. I Bieber some of your intuition fails because you have wormgly assumed this point to be obvious.
You're also mixing up the Chinese room experiment with the larger problem of an interactive AI. The Chinese room system is restricted, by definition, to only interact with the word through language. Over time, as it interacts with a person, it may decide that it likes them or not, and its responses may start changing based on this - that is allowed to be part of the algorithm that the human/machine inside the room is applying.
For a silly example, the algorithm may say that 'people who use the same word three times in a row are extremely nice', and use this state of 'extremely nice' to modify its future responses once the same word has been recorded three times in a row. This is no different, fundamentally, to me liking people whose face is symmetric, or the same gender as myself - it's not a choice I made, it's just an arbitrary rule of my (genetic) algorithm.
QM systems cannot be described by such functions, hence the world isnt a computer. Eg., any quantum randomness isnt a function as it has no input state and a non-determinstic output state.
Worse, clearly at least the input and output states of a computer should be measurable -- but most systems arent. Even including classical ones.
Ie., in chaotic systems it isn't possible, due to QM, to measure x st. f in y = f(x, t) is just a set (x, y, t) -- for large t. f here cannot be a function -- as y is necessarily a distribution of states.
In abitary chaotic systems y spans the space of all possible states of abitary t, and thus we have zero ability to predict anything beyond relatively narrow time-horizons.
If you wish to claim reality is a computer: (1) state the properties of a computer; (2) show that having these properties is non-trivial (ie., it is a substantive claim that something is a computer); and (3) show that these properties are at least consistent with our best theories of physics.
This same process also extends to chaotic systems: a Turing machine can take all possible measurement clause and compute the state of the system after N steps.
The mistake you are making is in the word 'predict'. Of course you can't predict the result of a non-deterministic process. But a computer doesn't predict: it computes, that is, it follows mechanical steps to transform an input value. And non-deterministic computations are nothing special, we do them every day in real computers ( rand() is a non-deterministic computation). The most famous class of algorithms in complexity theory is even called 'Nondeterministic Polynomial Time', NP for short (of P=NP? fame).
To our best knowledge, that would lead to the MWI.
For given input (the seed), the operation of rand() is deterministic and completely predictable.
Either way, extending the TM model with an idealized rand() that is purely random doesn't significantly alter the model, and it makes it capable of non-deterministic computations.
Hilbert space is infinite-dimensional and real-valued.
QM says that there is no `rand(seed)`, rather there is only `rand()` that is what "no hidden variables" means.
If you want to do `rand(s) forall s`, that "forall" requires real numbers -- and then you're out-of-luck on computability.
> Hilbert space is infinite-dimensional and real-valued.
Infinite-dimensional Hilbert state spaces are a useful mathematical model, but they are not physical. Physical systems have finite elements and thus finite state spaces. We could argue that space itself is infinitely divisible, so that we need infinite-dimensional state spaces to represent the infinite possible positions of a particle, but this is not a physical concern, as it is impossible to differentiate in finite time two states that only differ in an infinitesimal position change - so, we are free to choose some minimal unit of length (say, one over Graham's number of the Planck length) and get a finite-dimensional state space. Crucially, for any amount of time and for any sensitivity of instruments, we can always choose some such unit and ensure that our results will not be distinguishable with those measurement instruments within that amount of time.
> QM says that there is no `rand(seed)`, rather there is only `rand()` that is what "no hidden variables" means.
QM says no such thing, though many interpretations do. Still, that is exactly why I chose rand(), not rand(seed) in my example, so I'm not sure why you're bringing this up. rand(seed) is a deterministic computation, rand() is a non-deterministic computation.
> If you want to do `rand(s) forall s`, that "forall" requires real numbers -- and then you're out-of-luck on computability.
This is not what I was proposing. I was proposing a Turing machine + perfectly random rand() as the simulation of a random world.
Even this is not necessary if you believe in the MWI, where the whole universe actually evolves perfectly linearly and deterministically from a fixed initial state, with no randomness of any kind. This interpretation is perfectly compatible with all observations of QM, if we also add the postulate that observers can only observe one state at a time (the one they are entangled with).
Even without going there, the construction I proposed earlier, where we essentially select enough real numbers to satisfy any possible measurement device leads to such that our computable set is indistinguishable in practice from the infinite uncomputable set we started with still applies.
In general, there is no (known) way to introduce infinity into empirical science - it is a priori impossible to distinguish, in finite time, between an arbitrarily large but finite quantity, and a truly infinite quantity.
That's neither here nor there though.
A myriad of other issues exist. The uncertainty principle precludes "in any world", infinitely precise measurement. Thus chaotic systems, at least of a large time horizon, arent deterministic.
A chaotic system is one in which "insignificant digits" in the input determine "significant digits" in the output. You need very very very low digit precession in inputs the longer the time horizon. This gets beyond the uncertainty principal.
You can run as many turing machines as you like (necessarily, uncountably many to be consistent with QM). But you cannot escape the problem.
The best theories of physics are extremely far away from "effective computation". They are phrased actively hostile to it. And you have to completely revise physics to make it even half-plausible. It's a project which is barely stated, and a long way from finished.
It seems, indeed, a little doomed to failure. Properties that physics uses day-in-day-out posses the trivial infinities of geometry; and the trivial randomness of uncountable ensembles etc. -- to claim that physics states that reality is a computer is bizarre.
The laws of physics are known to be computable. Quantum Mechanics is one of the easiest to prove this for, since it is based on simple linear transforms! (Thermodynamics and GR are more complex, but still computable).
I have no idea why you believe differently.
A chaotic system requires arbitrary precision in the input measurements to be able to predict arbitrarily precisely its state after arbitrarily many steps. But, crucially, it does not require infinite precision in the input unless you want to predict its state with infinite precision (non-goal) or after infinite time (also a non-goal).
The uncertainty principle is also irrelevant. The uncertainty principle has to do with measurement of classical properties. Conversely, the wave function of a system is precise, not fuzzy. It's only in measuring properties of the wave function that we reach uncertainty or probability at all. So, a computer can precisely compute the simple linear evolution of the wave function, according to the Schrodinger equation, which is what the universe itself is doing. The result of this computation is a complex value. You can then apply the Born rule to this result to get the probability of your system being in a particular state after following the simulated evolution, just as you would for a real quantum mechanical system.
We do this kind of quantum simulation every day with great success, with a countable number of classical computers (typically 1), though it is only tractable for very very simple systems. The results don't typically differ in any perceivable way from the actual experiment.
> To me its incredibly obvious that they do not.
That is Searle’s approach, but I think the point where Searle and others taking it fall down in the inability to do this: Explain any testable manner in which we can confirm that those who you would describe as “really understanding” a language (or anything else) are actually doing something materially different from what you assert is obviously not understanding.
Until someone can do this, it is clear that the Chinese Room is nothing more than a conclusion that mechanistic computation is insufficient for understanding resting solely on the assumption that...mechanistic computation is insufficient for understanding.
Which is not an argument for any particular intepretation of the Chinese room, it’s only to say that the bare fact that the original author thought of it in a particular way doesn’t mean we should do so.
Is the consciousness given existence because you perform calculations on paper, or does it exist already within the fabric of reality and you're merely exposing it?
Just because we don’t tend to recognize one without the other doesn’t mean they necessarily have a causative relationship.
I don't see any sense in which these aren't different, unless perhaps you're introducing some separate cosmological "Boltzmann brain"-style argument that all possible finite results of all halting algorithms will arise or have arisen due to random fluctuations. But in that case you've got a whole new set of assumptions to address.
[0]: https://en.wikipedia.org/wiki/Mathematical_universe_hypothes...
Did philosophers not watch Blade runner? Sure Descartes didn't - but contemporaries sure as shit do!
To answer your question, I claim that you can't. Radical skepticism ("we live in the universe of the evil demon"), which is what I advocate for, means that you can't even claim that you exist. See the works of Max Stirner for further elaboration of the implications of this for Philosophy.
Paraphrasing from memory: he pins "cogito ergo sum" on the assumption that objective knowledge is possible. That assumption is built atop imagining an alternative: an evil demon has full control over his senses and feeds him whatever it wants to (essentially, the 'Matrix hypothesis'). His solution to this concern? "A loving God wouldn't create a universe that banal and meaningless."
It's only a proof if you accept the postulate that objective reality is correlated with experience.
The Chinese room argument is effectively a counterargument to the Turing test. Turing defined intelligence in terms of external behavior of the system. According to him, a system is intelligent, if its behavior is indistinguishable from that of a human in a task that requires intelligence.
Searle claimed that intelligence is a property of the physical system itself, not of its external behavior. In order to determine whether a system understands something, we have to observe its internal behavior. A state machine could in principle pass the Turing test due to an extensive set of rules, but Searle would not call it intelligent, because it is fundamentally a simple mechanism.
The original Chinese room argument went further. Searle claimed that the same principle that rules out the intelligence of a state machine also rules out the intelligence of any programmed computer. I think that can be attributed to taking models of computation too literally. If we define computation as taking a fixed program and a single input and producing a single output, it feels mechanical rather than intelligent. A system like that can simulate another system that interacts with the environment and updates its internal state, but like Searle himself claimed, a simulation is not the same as the simulated system.
The point about him doing the manual execution of the computer is a stupid parlor trick. Individual proteins in axons have no special awareness of "English" encoded in the firing. RNA expression has no special awareness of "English". It seems like appeal to bias rather than an actual argument.
I find far too many philosophical constructs insufficiently deconstructed.
Plus everyone has a different personal interpretation of language and meaning shaded by different exposures and interpretations of sensory input. Sufficient common experience is likely what gives words some semblance of common meaning between people.
I would argue what makes AI truly hard to do is that a computer can't have the experience of birth, child rearing, etc. It's too foreign. But at some point language veers into a coded representation of physical laws and reality, and there AI can converge with human experience.
Whether the Chinese room responds quickly or slowly doesn't really factor into whether the room understands Chinese.
So, Searle supposes that he is in a closed room and has a book with encoded instructions that supposedly allow him to communicate in Chinese; and he also supposes that being able to convince the Chinese-speaker that they're speaking to a human (i.e. passing a Turing test) is sufficient to demonstrate "understanding" - for whatever definition of "understanding" Searle wants to use.
If we discount the second assumption (which might be the case - IMHO humans could be tricked in a Turing test by something that makes zero pretense at understanding anything), then there's no paradox, then it's just an experiment that's too weak to say anything about "understanding" whatsoever; but since Searle considers the experiment valid, let's assume that this holds, and if someone/something can convince the Chinese person of "proper understanding", then that someone/something would really have "proper understanding".
So it comes down to whether simply blindly following instructions a book can achieve proper understanding. If yes, then Searle must concede the argument. And if not, then Searle's room simply doesn't work - apparently such a book is impossible to make or have, the Chinese room can't exist and the impossible experiment can't have any results.
In essence, it's a tautology - no matter how ridiculous you consider encoding "proper understanding" in a book of instructions, if you start with an assumption that you do have such a book, then you obviously have assumed a world where the combination of instructions+mindless executor can have "proper understanding"; and if you start with an assumption that a machine like that definitely can not be thinking "properly" (as Searle does) and Strong AI is impossible, then you have thrown away the key thing required for the thought experiment - the possibility for such a Strong-AI-enabling book to exist.
If you make two assumptions and find out that they are incompatible - like Searle does here - then you can deduce that at least one of them is wrong, but it does not say anything at all about which one of them is wrong, so I don't see much value in this thought experiment.
This is a fallacy, regardless of the language. Some comms between conscious entities must always contain metaphoric language, including signaling through body language. A shrug is a metaphor for uncertainty. This is why the premise of the Chinese Room is true (we can’t build a mind in a box) but also why conscious machines will necessarily have to have bodies.
Anything that is conscious and can hold a (real) dialog with a human will have emotional output tied to the body and it’s collective experience.
What do you mean? Humans can certainly communicate and maintain meaningful relationships solely through writing.