Non-paywalled obit:
https://www.theguardian.com/world/2025/oct/05/john-searle-ob...
His most famous argument:
https://www.theguardian.com/world/2025/oct/05/john-searle-ob...
His most famous argument:
The human running around inside the room doing the translation work simply by looking up transformation rules in a huge rulebook may produce an accurate translation, but that human still doesn't know a lick of Chinese. Ergo (they claim) computers might simulate consciousness, but will never be conscious.
But is the Searle room, the human is the equivalent of, say, ATP in the human brain. ATP powers my brain while I'm speaking English, but ATP doesn't know how to speak English just like the human in the Searle room doesn't know how to speak Chinese.
Neither the man, nor the room "understand" Chinese. It is the same for the computer and its software. Jeffery Hinton has sad "but the system understands Chinese." I don't think that's a true statement, because at no point is the "system" dealing with semantic context of the input. It only operates algorithmically on the input, which is distinctly not what people do when they read something.
Language, when conveyed between conscious individuals creates a shared model of the world. This can lead to visualizations, associations, emotions, creation of new memories because the meaning is shared. This does not happen with mere syntactic manipulation. That was Searle's argument.
That's not at all clear!
> Language, when conveyed between conscious individuals creates a shared model of the world. This can lead to visualizations, associations, emotions, creation of new memories because the meaning is shared. This does not happen with mere syntactic manipulation. That was Searle's argument.
All of that is called into question with some LLM output. It's hard to understand how some of that could be produced without some emergency model of the world.
LLM output doesn't call that into question at all. Token production through distance function in high-dimensional vector representation space of language tokens gets you a long way. It doesn't get you understanding.
I'll take Penrose's notions that consciousness is not computation any day.
I know that it doesn't feel like I am doing anything particularly algorithmic when I communicate but I am not the hommunculus inside me shuffling papers around so how would I know?
Hopefully we have all experienced what genuine inspiration feels like, and we all know that experience. It sure as hell doesn't feel like a massively parallel search algorithm. If anything it probably feels like a bolt of lightning, out of the blue. But here's the thing. If the conscious loop inside your brain is something like the prefrontal cortex, which integrates and controls deeper processing systems outside of conscious reach, then that is exactly what we should expect a search algorithm to feel like. You -- that strange conscious loop I am talking to -- are doing the mapping (framing the problem) and the reducing (recognizing the solution), but not the actual function application and lower level analysis that generated candidate solutions. It feels like something out of the blue, hardly sought for, which fits all the search requirements. Genuine inspiration.
But that's just what it feels like from the inside, to be that recognizing agent that is merely responding to data being fed up to it from the mess of neural connections we call the brain.
You can take this insight a step further, and recognize that many of the things that seem intuitively "obvious" are actually artifacts of how our thinking brains are constructed. The Chinese room and the above comment about inspiration are only examples.
I cannot emphasize enough how much I dislike linking to LessWrong, and to Yudkowsky in particular, but I first picked up on this from an article there, and credit should be given where credit is due: https://www.lesswrong.com/posts/yA4gF5KrboK2m2Xu7/how-an-alg...
By the way, the far more impactful application of this principle is as a solution (imho) to the problem of free will.
Most people intuitively hold that free will is incompatible with determinism, because making a choice feels unconstrained. Taken in the extreme, this leads to Penrose and others looking for quantum randomness to save their models of the mind from the Newtonian clockwork universe.
But we should have some unease with this, because choices being a random roll of the dice doesn’t sit right either. When we make decisions, we do so for reasons. We justify the choices we make. This is because so-called “free will” is just what a deterministic decision making process feels like from the inside.
Philosophically this is called the “compatibilist” position, but I object to that term. It’s not that free will is merely compatible with determinism—it requires it! In a totally random universe you wouldn’t be able to experience the qualia of making a free choice.
To experience a “free choice” you need to be able to be presented with alternatives, weight the pro and con factors of each, and then make a decision based on that info. From the outside this is a fully deterministic process. From the inside though, some of the decision making criteria are outside of conscious review, so it doesn’t feel like a deterministic decision. Weighing all the options and then going with your gut in picking a winner feels like unconstrained choice. But why did your gut make you choose the way you did? Cause your “gut” here is an unconscious but nevertheless deterministic neural net evaluation of the options against your core principles and preferences.
“Free will” is just what a deterministic application of decision theory feels like from the inside.
Unless we suppose those books describe how to implement a memory of sorts, and how to reason, etc. But then how sure are we it’s not conscious?
I'm not even sure what you are asking for, tbh, so any answer is fine.
It's implied, since they enable someone who does not know Chinese to respond equally well to questions as someone with Chinese as a native language.
There are two possibilities here. Either the Chinese room can produce the exact same output as some Chinese speaker would given a certain input, or it can't. If it can't, the whole thing is uninteresting, it simply means that the rules in the room are not sufficient and so the conclusion is trivial.
However, if it can produce the exact same output as some Chinese speaker, then I don't see by what non-spiritualistic criteria anyone could argue that it is fundamentally different from a Chinese speaker.
Edit: note that here when I'm saying that the room can respond with the same output as a human Chinese speaker, that includes the ability for the room to refuse to answer a question, to berate the asker, to start musing about an old story or other non-sequiturs, to beg for more time with the asker, to start asking the akser for information, to gossip about previous askers, and so on. Basically the full range of language interactions, not just some LLM style limited conversation. The only limitations in its responses would be related to the things it can't physically do - it couldn't talk about what it actually sees or hears, because it doesn't have eyes, or ears, it couldn't truthfully say it's hungry, etc. It would be limited to the output of a blind, deaf, mute Chinese speaker confined to a room whose skin is numb and who is being fed intravenously, etc.
Indeed. The crux of the debate is:
a) how many input and response pairs are needed to agree that the rule-provider plus the Chinese room operation is fundamentally equal/different to a Chinese speakers
b) what topics can we agree to exclude so that if point a can be passed with the given set of topics we can agree that 'the rule-provider plus the Chinese room operation' is fundamentally equal/different to a Chinese speaker
Sounds like circular logic to me unless you make that assumption explicit
The success of LLMs imitating human speech patterns, often better than most people (ask an LLM to write a poem about some topic in a certain style and it will do better than 99% of people and do it faster than 100% of people) is pretty impressive. "But it is just a thought-free statistical model, unlike people". I agree it is a thought-free statistical model.
But most of the things we all say in conversation is of the same quality. 99% of the time in conversation words tumble out of my mouth and I learn what I said when I hear my words in the same moment by conversation partner does. How is that any different from how today's LLM models behave? Is such dialog any more thoughtful than what LMMs produce?
The problem with the people who buy Searle's argument is they don't really think through the magnitude of what would really be required to pull it off. It wouldn't just be a static book, or a wall full of encyclopedias. It would have to be a stateful system that modifies that state and deduces new rules that affect future transformations as flexibly as the human mind does. To me it is clear that such a system really does think in the same way that humans do, no dualism required.