LLMs are trained to produce outputs that humans would find reasonable. That's it. They don't do anything else.
To say they have feelings and that they can think/reason is insane.
To say they have feelings and that they can think/reason is insane.
Alan M. Turing thought about criteria to settle the question of whether Machines Can Think, a question of which we now know that it is about as relevant as the question of whether Submarines Can Swim
The question of whether machines can think is about as relevant as the question of whether submarines can swim.
The whole essay [1] is actually somewhat relevant to modern LLM use for other reasons. He spends a lot of time lamenting the industry's obsession with "user friendly" interfaces which obscure the underlying complexity of the problem at hand.
Of course his glib dismissal doesn't make the question wrong, it's just a fun quote for those who already believe that human thought could (eventually) mimicked by a machine.
I don't think the original context particularly matters: the quote now has a life of its own. A rock was formed to satisfy the laws of thermodynamics that demanded an efflux of heat, but I can still use it to smash a window.
This is software trained to produce outputs. It doesn't feel anything.
Take this statement: "consciousness emerges from sufficient complexity in any system". Neither you or I can prove it right or wrong, but we can both make an assumption that if it was right, then there would be a way for an LLM to be conscious.
I'm not arguing that they are, I'm saying being this confident that they are not is the exact same as being confident that they are.
Mathematics is an invention of humans.
Which shouldnt be a surprise - most CS folks are disconnected ironically from reality.
I don't know what you ate today. It would be insane for me to confidently claim that you ate two bananas and a pizza today.
The burden of proof is on the person making the claim. That's the basis of empirical research.
I've read a number of panpsychist publications, yet I have yet to encounter an author who asserts with certainty that stones are conscious exactly in the same way human beings are.
The more common panpsychist interpretation, that everything exists on a spectrum of consciousness, with stones at the low end and humans at the high end, seems like a fairly reasonable ontological hypothesis to me.
Functionally,I'd say a key difference is that it's probably easier to perform general computation with a computer than to achieve the same goal using a stone.
Are you conscious on a nanosecond timescale?
I said nothing of the kind. I can only assume your question in response to my comments mean you think a powered computer is alive. I'll challenge that on the basis that a computer can generally maintain its state without power just fine. Shut it down, unplug it. Wait a day, a year, a decade and plug it back in. Assuming the hardware didn't degrade to the point of failure you can turn it on again and it will be in the same state you left it in before you unplugged it. Is it alive while its unplugged? Or is it simply not alive whether its plugged in or not?
It is powered while it is powered.
Under general anesthesisa you are still alive, but you don't have consciousness.
It'a even in their name, they model language, language is the thing that humans use to communicate, therefore they model communcating humans.
Humans are the precedent.
As it is, it would be insane for me to confidently claim you are conscious or even that you're human, as an LLM could have written the same message.
The philosophical zombie does not satisfy occams razor because it asserts an unmeasurable property that has no effect on the outcome.
If it acts like it has consciousness, then assuming it to have consciousness is more scientifically rigorous, than assuming for it not to have consciousness, because the former requires one less variable.
So why keep discussing?
But you could replicate exactly the calculations happening as a LLM computes the next token. Would that be conscious? Where would that consciousness reside?
My view is that LLMs are designed and trained to autocomplete narratives, and that as part of that process they develop an emergent narrative about themselves. Whatever personality traits they exhibit is part of that narrative. Perhaps this is not that different from how humans learn to move in social contexts, but it is not consciousness as it happens in living beings.
Why not, at least in principle? Wouldn’t an exact map of a brain’s neural structure and its electromagnetic and chemical state, combined with inputs mirroring the senses, be able to be automated to produce a digital analog of human thought.
The model is impractical and couldn’t be reproduced today, yes. But what, exactly, is missing?
https://www.hawking.org.uk/in-words/lectures/godel-and-the-e...
People have thought they were nearly complete at least for 500 years, probably longer.
When you choose a specific model to represent the brain you are asserting you’ve accounted for everything that generates consciousness (=the lights are on, experiences qualia). This means you’ve solved the hard problem. The hard problem is debated constantly though.
In my view there is absolutely no reason to think a description of reality can be equal to reality. There are probably (in my view that is) aspects of reality, like first person internal experience, that fundamentally can not be described with third party linguistic / symbolic / representational devices.
This is not an argument for true AI being possible, let alone already achieved; it is a warning about depending on arguments from incredulity, which are particularly susceptible to future developments.
Searle's Rooms is a fully rejected and rebutted argument, and Philosophy will never be (and never has been) a serious discipline until it figures out how to actually let go of failed arguments.
That’s not to say that I think the Chinese Room Argument is convincing. I don’t think it is. Or, I guess it might demonstrate something, but not what it aims to show.
I would argue that if, say, the person who has memorized this program and mentally executes it, and if this program also includes actuating some levers to control a mech or something, and the program (running via the person mentally going through the instructions by rote) takes actions to pursue some goals (e.g. purchasing some wood by conversing in Chinese and building a tower with that wood), while the person on whom the program is running wouldn’t know how to order a cheeseburger in Chinese,
Then, the program running in that person’s head might well understand Chinese even though the person doesn’t. (Whether or not the program actually understands Chinese, I think probably depends on the specific of the program in ways that I don’t think anyone knows. I don’t expect that any program that would understand could be small enough that a human person could memorize it and execute it by rote in this way, but this is a thought experiment and such issues can be disregarded I think.)
Critics immediately point out: yes, there clearly is still understanding, it's just in the system itself -- the program. The fact that the person does not have the understanding doesn't mean the understanding is gone.
His response to this is: but what if the system was the person's brain!?
Okay, so either it doesn't address the criticism at all (there is still a system of rules that contains the understanding; who cares whether that system is internal or external to the person's brain?) or it completely destroys the original premise of the argument, which is that the understanding is not in the person! You can't put the understanding back in the human agent and still claim the human doesn't understand what they're doing! His entire argument relies on the fact that the human agent obviously doesn't understand Chinese, and then says, actually, their brain contains everything it needs to know to have completely fluent conversations in Chinese. So the person does understand Chinese!
> How does the person having all the instruction books memorized and whatnot imply that the person understands Chinese?
How does a person's brain containing all the facts, information, and skills that it needs to have a fluent conversation in Chinese imply that the person understands Chinese? How does it not!?
Notice that the man while having memorized the rulebook can't convey any fact of his childhood in Chinese. He can only convey facts encoded in the rulebook, perhaps of a hypothetical childhood. The point is there are different ways information is stored in the brain; one way is through rote memory, the other is something like semantic memory. The rulebook encoded as rote memory doesn't endow his semantic memory contact with the meaning of the symbols. Hence the man doesn't understand Chinese.
This doesn't render the Chinese argument successful. The system's reply still stands when understood properly. I like to call it the virtual mind objection. Physically executing the rules in the rulebook just instantiates a virtual mind. Whether the man is reading the rules from the book or just pulling them from memory is irrelevant.
They don't, they have simply memorised a translation function which takes Chinese as input and returns Chinese as output. They don't have facts or information about the Chinese language - they don't even need to know what language they are speaking!
When the person reads 汉堡包, do they understand what that is?
Assuming you don't speak Mandarin, do you understand that word? If not, how could you claim to understand Mandarin? Even if you could use the translation function to speak it, you still don't understand it. Remember that the translation function is Mandarin -> Mandarin, so if you don't understand Mandarin you cannot use it to gain an understanding of the word.
Because they can’t use Chinese to speak about things they encounter in the world?
Like, someone asks them in Chinese whether they are hungry and want some food, and they answer based on the program, not based on whether they are actually hungry, because they don’t understand the words in a way that lets them use them as a person who understood Chinese would.
Again, I don’t mean to say that in such a situation that there is no understanding of Chinese happening in their head, only that this doesn’t constitute that person understanding Chinese.
It seems to me very strange to want to say that the person with the simulation in their head understands Chinese. They can’t do the things that understanding Chinese lets people do.
Perhaps the program the person has running in their head is even a person, a moral patient and a moral agent, and one who understands Chinese. But the person who is running (by rote) the program, doesn’t thereby understand Chinese. They may “contain understanding of Chinese”. The understanding is definitely happening in their brain. But they don’t understand Chinese.
When you hear a certain set of symbols in a certain context you jump.
The thought experiment is fundamentally nonsensical.
Because in the thought experiment, the translation function (which is what you mean by the book) is f(Chinese) -> Chinese. If you don't have knowledge of Chinese, it won't help you.
We can also play a little game. I’ve invented a language called Kejdoke.
Me: Jeheixis ownc j’Lm hys bjækf¿
You: Kloften¡’by
You should be able to tell me what this exchange was about. Since any reasonable definition of understanding for a language would be to do this, and you have the same knowledge of Kejdoke as the person in the thought experiment has about Chinese. If you can't tell me what this exchange was about, the person in the thought experiment could not understand either.
idk how legit this is but the summary is my point https://arxiv.org/abs/2206.07682
They can have empathy too. Meaning, you can describe a circumstance with many parties involved, with different motivations and conflicting goals -- and it will tell you how these humans would likely feel about each other and the situation at hand.
1. "LLMs have empathy since they write empathetic things."
2. "How do you know its their own empathy?"
3. "LLMs are authors, just like humans are."
4. "How do you know that they are that kind of author?"
5. [GOTO 1]
To paraphrase the Chinese room argument: Put me in a room with infinite time, paper, pencils, filing cabinets, and give me an English language list of instructions for doing all the neural-net calculations required to run an LLM. Slide your prompt under the door, and, after doing all the calculations, I will slide the next word out, and so I will repeat until the response is done.
In this thought experiment, I'm just following the instructions. So, in this case, what is doing the "thinking" or "reasoning" or "empathy" here?
Your brain is just a big set of weights/neurons organized into layers that do back and forward propagation, turns out. You are just statistics surrounded by flesh and bones.
LLMs can reason, reasoning is logic, and as you said yourself it is derived from what humans belief is reasonable. However, there is more to the structure of language, the way mathematics is also a language that ultimately describes the universe.