I am not talking about a thinking entity but a conscious one.
Internal mental states of humans are not important here. The task they perform is important.
I am talking about a system that is functionally identical to a brain, whether its constituents are humans, chips, or any other entities that work similarly to neurons. And I am not talking about thinking, but consciousness.
The mental states of humans affect how they perform their task. We are not automotons.
A system that is made up of a billion people cannot be functionally identical to a brain.
Our definitions of consciousness are different, which makes the conversation impossible.
How? All we are asking is that people do a simple task!
It is in the case of the China brain.
I already defined what I meant by consciousness.
Dawkins is correct.
In order to discount his hypothesis people need a theory of consciousness, and most people in this thread don’t have one.
You also need to understand how modern LLMs actually work, and again, most people here don’t.
You dismiss Dawkins’ hypothesis based on your interactions with ChatGPT, but he didn’t say anything about ChatGPT, he was talking about Claude. I’ve interacted with both, and Claude is very different.
There’s a reason why my interactions with Claude are more profitable than talking to the vast majority of humans.
Indeed, Claude is often a better interlocutor than the majority of respondents here.
I share your exasperation. It is often infuriating.
My money says it is not. But it will never be tested, so we’ll never know which of us is right.
As did I.
I don’t understand what you are talking about. I am not angry or irritated at all. All I am asking is whether the China brain experiences or not, given that the China brain resembles a neural network. If you’re not happy with people, replace them with chips.
It can be tested.
What is your definition?
How would that work?
Consciousness is fundamental. It is a property of matter, just as mass and charge are. But it is not a physical property like mass and charge are. It is subjective experience. Matter subjectively experiences because of it.
It’s not a mistake. It’s a contract between him and Anthropic to hype the hell out of the latest tulip mania. Funny how their logo is on all his social media releases and he never mentions ChatGPT.
Experience is a first-person phenomenon, so we cannot measure it. We just tell things are conscious or unconscious based on personal judgment. So we cannot test whether any entity is conscious or not if by testing you mean measuring the experience. We can, however, build a neural network, whether its nodes are humans, chips, or other entities that function like neurons. So, the question of whether a neural network experiences or not is valid regardless of what its nodes are. I would say that two different neural networks with different types of nodes experience the same thing if they are functionally identical, of course, if experience occurs when things are organized as a neural network. You agree that there are creatures in nature with a few hundred neurons that are conscious. This question is therefore valid: Does another neural network, which is functionally identical to the brain of these creatures, experience the same thing or not?
Are you talking about the property dualism?
I’m not sure how you can specifically justify stating that consciousness is a special property of the universe, since one can posit it to be an emergent quality arising from complex system rather than something baked into the universe at a fundamental level.
That said, to me the issue with ascribing consciousness to our present AI’s is that, aside from just what I mentioned about our present AI’s not having creativity, these AI’s do not have the embodied experience to understand the world on a first-person basis — in the end they are just extremely sophisticated language or image processors.
A good example of this is that we still have not made an AI capable of beating the ‘Coffee Test’ — i.e. an autonomous robot controlled by an AI shows up at any given random house which it has never been at before and is tasked with making a cup of coffee without error. In the process it will have to find where the coffee is stored, find the coffee machine, find where to get water, find a cup for coffee, figure out how to get the coffee into the coffee machine (including locating, operating, and replacing a coffee grinder if the coffee is not pre-ground) without error, figure out how to get water into the coffee machine without error, figure out how to operate the coffee machine without direction and without error, figure out how to return the unused coffee to whence it came without error, and figure out how to dispense coffee into the cup without error.
Until we have a robot that can beat the Coffee Test, IMO we still do not have proper embodied autonomous AI.
I would say we just guess things are conscious or unconscious based on personal judgment.
That’s not what I mean. What I mean is we cannot get one billion Chinese people, or one billion people at all, to take part in this experiment.
Is a succession of neurons in which ions pass through the membranes functionally identical to an electric current running through electronics? The material and form of energy isn’t important, as long as we shape things the same?
Well, sure. If it works the same way, then it works the same way.
However, a billion people doing whatever are not functionally identical to a neutral network.
Yes
I’m not positive, eh? It’s the guess that I think works best, but there’s no more evidence for it than any other guess.
But I do not think we can posit it to be an emergent quality arising from complex system. I mean, obviously, you can. But consciousness is not physical. At least not in any way inn aware of or anyone can point to. I don’t see making something non-physical out of physical components. That’s worse than making a house out of smoke, because at least a house and smoke are both physical.
I’ve never heard of there Coffee Test. That’s excellent!
Apologies. I forgot this.[quote=“MoK, post:93, topic:927”]
You agree that there are creatures in nature with a few hundred neurons that are conscious. This question is therefore valid: Does another neural network, which is functionally identical to the brain of these creatures, experience the same thing or not?
[/quote]
I agree that there are creatures in nature with a few hundred neurons that are conscious, because I think everything is conscious. Everything subjectively experiences it’s own existence. But proximity is not what gives conglomerates of matter a unity. Not as though two bits of sand touching are a teo-grain conscious unit, drop another grain on them and it’s a three-grain conscious unit, four, five, a thousand, a million, always a conscious unit just because they happen to be touching.
But sure, if a neural network is conscious as a unit, a functionally identical neural network also is. The question is, is a billion humans acting out the roles of the billion neurons in a billion-neuron neural network functionality identical to that neural network? And the answer is No.
But we can organize a few hundred individuals to perform tasks that are required to make them functionally identical to a small neural network. The question is whether this neural network experiences or not.
A neural network is a set of nodes, each of which receives electrical pulses from other nodes. Each node fires if it receives sufficient electrical pulses from other nodes; otherwise, it does not fire. A neural network can be made of people, neurons, chips, or any entity whose nodes work in the same manner as described earlier. We say that two neural networks are functionally identical if the way their nodes are connected is similar, and each of their nodes fires in the same manner.
What nodes are is not important unless you want to say that a neural network is more than a set of nodes.
I’m not aware of whatever you’re talking about, but if Dawkins is literally shilling for an AI that’s fucking hilarious
“The Claude Delusion” really cracks me up. There are layers to that joke.
It’s altogether unsurprising that some (many) people would think AI as a real person. People impute personality to machines that are far less bright than Claude. Your car, perhaps. 35 years ago I felt like my first computer (a Mac Plus) was sort of conscious. That was just an affectionate feeling towards a new gadget, nothing originating in the machine.
No doubt about it, ChatGPT, Claude, et al perform at an amazing level. Probably too amazing for our own good.
This is, I believe, a common mistake. You’re mistaking the criteria by which it was trained for the way it works. It was trained under circumstances where it was tasked to figure out the just likely next words given all the previous words. But that doesn’t mean that’s “how it works”. Any more than a wide receiver being trained to catch a football is how that wide receivers muscles or brain work. You wouldn’t say “A very simple way of understanding how Jerry Rice works is, being thrown a football, what’s the best way to catch it?” That’s what he was trained to do, not “how he works”.
I’m referring to the software; you’re referring to the hardware. LLMs are hardware-independent, except that it runs better on some chips.