Maybe I should look up some of my other heroes and heretics while I have the chance. I mean, you don't need to cold e-mail them a challenge. Sometimes they're already known to be at events and such, after all!
I mean, I guess all arguments eventually boil down to something which is "obvious" to one person to mean A, and "obvious" to me to mean B.
Two systems, one feels intuitively like it understands, one doesn’t. But the two systems are functionally identical.
Therefore either my concept of “understanding” is broken, my intuition is wrong, or the concept as a whole is not useful at the edges.
I think it’s the last one. If a bunch of valves can’t understand but a bunch of chemicals and electrical signals can if it’s in someone’s head then I am simply applying “does it seem like biology” as part of the definition and can therefore ignore it entirely when considering machines or programs.
Searle seems to just go the other way and I don’t under Why.
Personally, I'd say that there is a Chinese speaking mind in the room (albeit implemented on a most unusual substrate).
First, it is tempting to assume that a bunch of chemicals is the territory, that it somehow gives rise to consciousness, yet that claim is neither substantiated nor even scientific. It is a philosophical view called “monistic materialism” (or sometimes “naive materialism”), and perhaps the main reason this view is popular currently is that people uncritically adopt it following learning natural scientific fields, as if they made some sort of ground truth statements about the underlying reality.
The key to remember is that this is not a valid claim in the scope of natural sciences; this claim belongs to the larger philosophy (the branch often called metaphysics). It is not a useless claim, but within the framework of natural sciences it’s unfalsifiable and not even wrong. Logically, from scientific method’s standpoint, even if it was the other way around—something like in monistic idealism, where perception of time-space and material world is the interface to (map of) conscious landscape, which was the territory and the cause—you would have no way of proving or disproving this, just like you cannot prove or disprove the claim that consciousness arises from chemical processes. (E.g., if somebody incapacitates some part of you involved in cognition, and your feelings or ability to understand would change as a result, it’s pretty transparently an interaction between your mind and theirs, just with some extra steps, etc.)
The common alternatives to monistic materialism include Cartesian dualism (some of us know it from church) and monistic idealism (cf. Kant). The latter strikes me as the more elegant of the bunch, as it grants objective existence to the least amount of arbitrary entities compared to the other two.
It’s not to say that there’s one truly correct map, but just to warn against mistakenly trying to make a statement about objective truth, actual nature of reality, with scientific method as cover. Natural sciences do not make claims of truth or objective reality, they make experimentally falsifiable predictions and build flawed models that aid in creating more experimentally falsifiable predictions.
Second, what scientific method tries to build is a complete, formally correct and provable model of reality, there are some arguments that such model is impossible to create in principle. I.e., there will be some parts of the territory that are not covered by the map, and we might not know what those parts are, because this territory is not directly accessible to us: unlike a landmass we can explore in person, in this case all we have is maps, the perception of reality supplied by our mind, and said mind is, self-referentially, part of the very territory we are trying to model.
Therefore, it doesn’t strike me as a contradiction that a bunch of valves don’t understand yet we do. A bunch of valves, like an LLM, could mostly successfully mimic human responses, but the fact that this system mimics human responses is not an indication of it feeling and understanding like a human does, it’s simply evidence that it works as designed. There can be a very different territory that causes similar measurable human responses to arise in an actual human. That territory, unlike the valves, may not be fully measurable, and it can cause other effects that are not measurable (like feeling or understanding). Depending on the philosophical view you take, manipulating valves may not even be a viable way of achieving a system that understands; it has not been shown that biological equivalent of valves is what causes understanding, all we have shown is that those entities measurably change at the same time with some measurable behavior, which isn’t a causative relationship.
> A bunch of valves, like an LLM, could mostly successfully mimic human responses,
The argument is not "mostly successfully", it's identically responding. The entire point of the chinese room is that from the outside the two things are impossible to distinguish between.
> The argument is not "mostly successfully", it's identically responding.
This is a thought experiment. Thought experiments can involve things that may be impossible. For example, the Star Trek Transporter thought experiment involves an existence of a thing that instantly moves a living being: the point of the experiment is to give rise to a discussion about the nature of consciousness and identity.
Thing not possibly existing is one possible resolution of the paradox. There may be a limitation we are not aware of.
Similarly, in Searle’s experiment, the system that identically responds might never exist, just like the transporter in all likelihood cannot exist.
> The entire point of the chinese room is that from the outside the two things are impossible to distinguish between.
To a blind person, an orange and a dead mouse are impossible to distinguish between from 10 meters away. If you can’t distinguish between two things, it doesn’t mean the things are the same. Ability to understand, self-awareness and consciousness are things we currently cannot measure. You can either say “these things don’t exist” (we will disagree) or you have to say “the systems can be different”.
The Chinese room is setup so that you cannot tell the difference from the outside. That’s the point of it.
> If you can’t distinguish between two things, it doesn’t mean the things are the same.
But it does mean that the differences between them are irrelevant to you by definition.
> Ability to understand, self-awareness and consciousness are things we currently cannot measure. You can either say “these things don’t exist”
Unless you have a way they could be measured but we just lack the technology or skill then your definitions are of things that may as well not exist because you cannot define them. They are vague words you use and are fine if you accept you have three major categories “yes and here’s why, no and here’s why and no idea” that’s fine. I am happy saying I’m conscious and the pillow next to me is not. I don’t have a definition clear enough to say yes/no if the pillow was arguing with me.
I feel like I could make the same arguments about the chinese room except my definition of "understanding" hinges on whether there's a tin of beans in the room or not. You can't tell from the outside, but that's the difference. Both cases with a person inside answering questions act identically and you can never design a test to tell which room has the tin of beans in.
Now you might then say "I don't care if there's a tin of beans in there, it doesn't matter or make any sort of difference for anything I want to do", in which case I'd totally agree with you.
> just like you cannot prove or disprove the claim that consciousness arises from chemical processes.
Like understanding, I haven't seen a particularly useful definition of consciousness that works around the edges. Without that, talking of a claim like this is pointless.
Not at all. The confusion you expressed in your original comment stems from that claim. If you want to overcome that confusion, we have to talk about that claim.
Your statement was that it’s unclear how a bunch of valves doesn’t understand, but chemical processes do, and maybe you have a wrong intuition. Well, it appears that your intuition is to make this claim of causality, that some sort of object (e.g., valves or neurons), which you believe is part of objective reality, is what would have to cause understanding to exist.
So, I pointed out that assumption of such causality is not a provable claim, it is part of monistic materialism, which is a philosophical view, not scientific fact.
Further hinting at your tendency to assume monistic materialism is calling the systems “functionally identical”. It’s fairly evident that they are not functionally identical if one of them understands and the other doesn’t; it’s easy to make this mistake if you subconsciously already decide that understanding isn’t really a thing that exists (as many monistic materialists do).
> Like understanding, I haven't seen a particularly useful definition of consciousness that works around the edges.
Inability to define consciousness is fine, because logically circular definitions are difficult. However, lack of definition for the phenomenon is not the same thing as denying its objective existence.
You can escape the necessity to admit its existence by waving it away as an illusion or “not really” existing. Which is absolutely fine, as long as you recognize that it’s simply a workaround to not have to define things (if it’s an illusion, whom does it act on?), that conscious illusionism is just as unfalsifiable and unprovable as any other philosophical view about the nature of reality or consciousness, and that logically it’s quite ridiculous to dismiss as illusion literally the only thing that we empirically have direct unmediated access to.
> It's not mostly mimicking, it's exactly identical.
> Both cases with a person inside answering questions act identically and you can never design a test to tell which room has the tin of beans in.
If you constructed a system A that produces some output, and there is a system B, which you did not construct and which you don't have an full understanding of how it works, which produces identical output but is also believed to produce other output that cannot be measured with current technology (a.k.a. feelings and understanding), you have two options: 1) say that if we cannot measure something today then it certainly doesn’t matter, doesn’t exist, etc., or 2) admit that system A could be a p-zombie.
Then you could tell the difference and the thought experiment is broken. The whole point is that outside observers can’t tell. Not that they’re too stupid, that there isn’t a way they could tell, no question they could ask.
> but is also believed to produce other output that cannot be measured with current technology
Are you suggesting that Searle was saying that there was a difference between the rooms and that we just needed more advanced technology to see inside them? Come on.
I tried to explain that outside observers may not observe the entirety of what matters, whether due to current technical limitations or fundamental impossibility. In fact, to assume externally observed behaviour (e.g., of a human) is all that matters strikes me as a pretty fringe view.
> Are you suggesting that Searle was saying that there was a difference between the rooms and that we just needed more advanced technology to see inside them
Perhaps you are trying to read too much into what the experiment itself is. I do not treat it as “Searle tried to tell us something this way”. If he wanted to say something more specific he probably had done it in relevant works. The thought experiment however is very clear and describable in a paragraph and is open to possible interpretations, which is what we are doing now. That is the beauty of thought experiments like this.
Second: the philosophically relevant point is that when you gloss over mental states and only point to certain functions (like producing text), you can't even really claim to have fully accounted for what the brain does in your AI. Even if the physical world the brain occupies is practically simulatable, passing a certain speech test in limited contexts doesn't really give you a strong claim to consciousness and understanding if you don't have further guarantees that you're simulating the right aspects of the brain properly. AI, as far as I can tell, doesn't TRY to account for mental states. That's partially why it will keep failing in some critical tasks (in addition to being massively inefficient relative to the brain).
> consciousness and understanding
After decades of this I’ve settled on the view that these words are near useless for anything specific, only vague pointers to rough concepts. I see zero value in nailing down the exact substrates understanding is possible on without a way of looking at two things and saying which one does and which one doesn’t understand. Searle to me is arguing that it is not possible at all to devise such a test and so his definition is useless.
Although for whatever it’s worth most modern AIs will tell you they don’t have genuine understanding (eg no sense of what pleasure is or feels like etc aside from human labeling).
The entire point of the thought experiment is that to outside observers it appears the same as if a fluent speaker is in the room. There aren’t questions you can ask to tell the difference.
This was why I have the tin of beans comparison.
The room has the property X if and only if there’s a tin of beans inside. You can’t in any way tell the difference between a room that has a tin of beans in and one that doesn’t without looking inside.
You might find that a property that has zero predictive power, makes (by definition) no difference to what either room can do, and has no use for any practical purposes (again by definition) is rather pointless. I would agree.
Searle has a definition of understanding that, to me, cannot be useful for any actual purpose. It is therefore irrelevant to me if any system has his special property just as my tin of beans property is useless.
> In reality the material difference between a computing machine and a brain is trivial
No it isn’t. You are making the strong statements about how the brain works that you argued against at the start.
> Among other practical differences such as guarantee of function over long term.
Once again ignoring the setup of the argument. The solution to the chinese room isn’t “the trick is to wait long enough”.
I don’t know why you want to argue about this given you so clearly reject the entire concept of the thought experiment.
I find the entire thing to be intellectual wankery. A very simple and ethical solution is that if two things appear conscious from the outside then just treat them both as such. Job done. I don’t need to find excuses like “ah but inside there’s a book!” Or “it’s manipulations are on the syntactic level if we just look inside” or “but it’s just valves!” I can simply not mistreat anything that appears conscious.
All of this feels like a scared response to the idea that maybe we’re not special.
The premise of the argument is that the Chinese Room passes the Turing Test for Chinese. There are two possibilities for how this happens: 1) the program emulates the brain and has the right relation to the external world more or less exactly, or 2) the program emulates the brain enough to pass the test in some context but fails to emulate the brain perfectly. We know that as it currently stands, we've "passed the Turing Test" but we do not go further and say that brains and AI perform "indistinguishably." Unless there are significant similarities to how brains work and how AIs work, on some fundamental level (case 1), even if they pass the Turing Test, it is possible that in some unanticipated scenario they will diverge significantly. Imagine a system that outputs digits of pi. You can wait until you see enough digits to be satisfied, but unless you know what's causing the output, you can never be sure that you're not witnessing the output of some rational approximation or some cached calculation that will eventually halt. What goes on inside matters a lot if you want a sense of certainty. This is simply a trivial logical point. Leaving that aside, assuming that you do have 1), which I believe we are still very far from, we're still left with the ethical consequences, which it seems you agree does hinge on whether the system is conscious.
You made a really strong claim, which is "I can simply not mistreat anything that appears conscious"--which is showing the difference in our intuitions. We are not beholden to the setup of the Chinese Room. The current scientific and rational viewpoint is at the very least that brains cause minds and they cause our mental world. I'm sure you agree with that. The very point we are disputing is that it doesn't follow that because what's going on on the outside is the same that what goes on on the inside doesn't matter. This is particularly true if we have clear evidence that the things causing the behavior are very different, that one is a physical system with biological causes and the other is a kind of simulation of the first. So when I say that a brain is trivially different from a calculating machine, what I mean is that the brain simply has different physical characteristics from a calculating machine. Maybe you disagree that those differences are relevant but they are, you will agree, obvious. The ontology of a computer program is that it is abstract and can be implemented in any substrate. What you are saying then, in principle, is that if I follow the steps of a program by tracking bits on a page that I'm marking manually, that somehow the right combination of bits (that decode to an insult) is just as morally bad as me saying those words to another human. I think many would find that implausible.
But there are some who hold this belief. Your position is called "ethical behaviorism," and there's a essay I argued against that articulated this viewpoint. You can read it if you want! https://blog.practicalethics.ox.ac.uk/2023/03/eth%C2%ADi%C2%...
> What goes on inside matters a lot if you want a sense of certainty. This is simply a trivial logical point
And yet entirely unrelated to this thought experiment. His point is not that the book isn't big enough, that the man inside the room will trip up at some point, or anything of the sort.
Now you might have a different argument about this all than Searle, and that's entirely fine. I'm saying that Searles definition of understanding is utterly pointless because he defines it as one that is not related to the measurable actions of a system but related to the way in which it works internally.
> The premise of the argument is that the Chinese Room passes the Turing Test for Chinese.
...
> enough to pass the test in some context but fails to emulate the brain perfectly
No. That is a far weaker argument than Searle makes. His argument is not that it'll be hard to tell, or convincing but you can tell the difference, or most people would be fooled.
From Searle, let's dig into this.
https://web.archive.org/web/20071210043312/http://members.ao...
> from tile point of view of somebody outside the room in which I am locked—my answers to the questions are absolutely indistinguishable from those of native Chinese speakers.
Already we get to the point of being indistinguishable.
> I have inputs and outputs that are indistinguishable from those of the native Chinese speaker,
Again indistinguishable.
And then he doubles down on this to the point of fully emulating the brain not being enough
> imagine that instead of a monolingual man in a room shuffling symbols we have the man operate an elaborate set of water pipes with valves connecting them. When the man receives the Chinese symbols, he looks up in the program, written in English, which valves he has to turn on and off. Each water connection corresponds to a synapse in the Chinese brain, and the whole system is rigged up so that after doing all the right firings, that is after turning on all the right faucets, the Chinese answers pop out at the output end of the series of pipes.
Searle has a problem - he looks at two different systems and says there is understanding in one and not in another. Then he ties himself in knots trying to distinguish between the two.
> The idea is that while a person doesn’t understand Chinese, somehow the conjunction of that person and bits of paper might understand Chinese. It is not easy for me to imagine how someone who was not in the grip of an ideology would find the idea at all plausible.
He cannot at all accept any sort of combination, he can't accept any concept of understanding being anything but binary. He cannot accept that it perhaps is not a useful term at all.
> in this paper I have tried to show that a system could have input and output capabilities that duplicated those of a native Chinese speaker and still not understand Chinese, regardless of how it was programmed
A programmed system *cannot* understand. It doesn't matter how it operates or how well, and again duplicating the capabilities of a real person.
As far as I can tell, since he leans heavily into the physical aspect, if we had two machines:
1. Inputs are received via whatever sensors, go through a physical set of components, and drive motors/actuators
2. Inputs are received via whatever sensors, go through a chip running an exact simulation of those same components, and drive motors/actuators
then machine 1 could understand but machine 2 could not because it has a program running rather than just being a physical thing.
Despite the fact that both simply follow the laws of physics, the very concept of a program is just how certain physical things are arranged.
To go back to my point because I'm rather frustrated yet again just pointing out what Searle explicitly says:
Searle defines understanding in a way that makes it, to me, entirely useless. It provides by definition no predictive power and can by definition not impact anything we want to do.
I am not arguing which of these things understands. I'm saying the term as a whole isn't very useful, and Searles definition has been pushed by him to a point of being entirely useless because he starts by insisting that certain things cannot understand.
So if you like, one is real and the other is fake. Or, one is physical and the other is symbolic or conventional. One actually had breakfast this morning and the other is lying about having breakfast to pass the Turing Test. One can feel pain, guilt, shame and the other one is just saying that it does because it’s running a program.
Searle says there is an empirical test for which domain a thinking object falls into (your machine 1 and machine 2)—to an outside observer, in the limit, there is no difference in behavior. They will do the same thing. For all that, if you have a metaphysical value for consciousness and “genuine” feeling, then you think the difference is important. If you don’t, you don’t.
FWIW—I think once AI has a full understanding of its ontology, even if it’s simulating a human brain perfectly, if it knows it’s a program it will probably explain to us why it is or is not necessarily conscious. Perhaps that will be more convincing for you.
If I understand your argument: if there's no empirical consequence, what's the point of the distinction, right?
Maybe he was cheating before or after, sure, but not during. No court would buy that.
...At least, that's how I interpret 'empirical consequence' - something observable or detectable, at very least in principle. Do you mean something different?
(Right this minute I'm coming from an empiricist framework where acts require consequences. If you're approaching this from a realist or rationalist view -which I suspect-, I'd be interested to hear it!)
btw--if you'd like to keep the conversation going, email is on my personal webpage in my bio.
I can imagine a lot of things, but the argument did not go this far, it left it as "obvious" well before this stage. Also, when I see trivial simulations of our biological machinery yielding results which are _very similar_, e.g. character or shape recognition, I am left wondering if the people talking about quantum wavefunctions are not the ones that are making extraordinary claims, which would require extraordinary evidence. I can certainly find it plausible that these _could_ be one particular way that we could be superior to the electronics / valves of the argument, but I'm not yet convinced it is a differentiator that actually exists.
There has to be a special motivation to instead cast understanding as “competent use of a given word or concept,” (judged by whom btw?). The practical upshot here is that without this grounding, we keep seeing AI, even advanced AI make trivial mistakes and requires the human to give an account of value (good/bad, pleasant/unpleasant) because these programs obviously don’t have conscious feelings of goodness and badness. Nobody had to teach me that delicious things include Oreos and not cardboard.
Well, no, that came from billions of years of pre-training that just got mostly hardcoded into us, due to survival / evolutionary pressure. If anything, the fact that AI is as far as it is, after less than 100 years of development, is shocking. I recall my uncle trounce our C64 in chess, and go on to explain how machines don't have intuition, and the search space explodes combinatorically, which is why they will never beat a competent human. This was ~10 years before Deep Blue. Oh, sure, that's just a party trick. 10 years ago, we didn't have GPT-style language understanding, or image generation (at least, not widely available nor of middling quality). I wonder what we will have in 10, 20, 100 years - whatever it is, I am fairly confident that architectural improvements will lead to large capability improvements eventually, and that current behavior and limitations are just that, current. So, the argument is that somehow, intuitively they can't ever be truly intelligent or conscious because it's somehow intuitively obvious? I disagree with this argument; I don't think we have any real, scientific idea of what consciousness really is, nor do we have any way to differentiate "real" from "fake".
On the other end of the spectrum, I have seen humans with dementia not able to make sense of the world any more. Are they conscious? What about a dog, rabbit, cricket, bacterium? I am pretty sure at their own level, they certainly feel like they are alive and conscious. I don't have any real answers, but it certainly seems to be a spectrum, and holding on to some magical or esoteric differentiator, like emotions or feelings, seems like wishful thinking to me.
I'm becoming less sure of this over time. As AI becomes more capable, it might start being more comparable to smaller mammals or birds, and then larger ones. It's not a boolean function, but rather a sliding scale.
Despite starting out from very skeptical roots, over time Ethology has found empirical evidence for some form of intelligence in more and more different species.
I do think that this should also inform our ethics somewhat.
On a side note: it's been a pleasure reading through the debates with you, and possibly we can continue over mail!
So what’s the physical cause for consciousness and understanding that is not computable? If for example you took the hypothesis that “consciousness is a sequence of microtubule-orchestrated collapses of the quantum wavefunction” [1], then you can see a series of physical requirements for consciousness and understanding that forces all conscious beings onto: 1) roughly the same clock (because consciousness shares a cause), and 2) the same reality (because consciousness causes wavefunction collapses). That’s something you could not do merely by simulating certain brain processes in a closed system.
1) Not saying this is correct, but it invites one to imagine that consciousness could have physical requirements that play in some of the oddities of the (shared) quantum world. https://x.com/StuartHameroff/status/1977419279801954744