This is a most excellent place for technology news and articles.
LLMs are not sentient, but this gimmick doesn't prove that.
There's nothing to indicate the human brain can't be implemented in a Turing machine too.
I don't think that is the case.
There's nothing that indicates the human brain has super-turing computation
Except that we've never done it, and we struggle to even define what consciousness is. But, other than that, sure.
There's fundamentally not much difference between our brain and a fly's, at the cellular level. We have fully simulated a fly's brain already. When given a virtual body, it promptly started acting like a fly.
I don't think LLMs are conscious or sentient. However, consciousness is likely just an internal illusion. There's no obvious reason we can't scale up from a fly to a human brain, other than difficulty. At that point you have a fully virtual brain that believes itself to be conscious, and can demonstrate sentience.
I'll believe it when it happens. There's a huge different between simulating 139,000 neurons and 50 billion neurons.
Besides, the authors looked at taste and touch sensory. That's not exactly predicting the entire behavior of a fly, and the authors (of which there were many), admitted they don't know if it can actually predict neural activity. Source
I'm not expecting anything any time soon either. Though I can see someone like musk pumping far too much money into it at some point.
My point was however that the difference is just one of scale. We don't need to predict the firings, just run it and compare it to nature. From what I've read, it behaves like a fly, including walking and grooming itself. This means there is no magic mystical difference between a real fly's brain and a virtualized one.
Projecting further, there is no difference, other than scale, between our brain and the fly. Implying there is nothing mystical about consciousness.
If a human brain can be conscious, then a virtualized human brain can be conscious. If a virtualized brain can be conscious, then so can the computer it runs on.
The question then becomes do we WANT consciousness in an AI, what would it look like, and how can we detect/measure it?
An illusion is a phenomenological experience, for which you must be conscious. Consciousness can not be an illusion. You must be conscious to experience an illusion at all.
It's a case of running out of terms. I was using it here as apparent, but not a "real" thing.
It's a bit like a lot of visual illusions, we can often all see them consistently, but they don't exist in the image itself.
In this case, consciousness is likely related to keeping our own mind functioning coherently. Providing a common virtual ground for the various parts of our brain to interact. There is no seat of consciousness, it's akin to the operating system on a computer. Not required, but makes a lot of tasks massively easier.
By the same logic, you could prove that we are not sentient.
Engineers need to stop trying to do the job of philosophers. LLMs are not sentient, but that tells us nothing about what gives rise to sentience or whether silicon is a substrate that could host a sentient consciousness.
I'm still not convinced most people I speak to at work are sentient.
They are probably sentient. Now sapience.. is another issue altogether.
I majored in Philosophy. Though I wouldn't want to be an AI ethics consultant for one of these big companies because I imagine that whatever I write down would have to preserve profits (first and foremost) and then gaslight everyone into thinking it's all perfectly ethical and okay and no one (including mother earth) is being harmed in the process.
You don't need an ethicist if you don't have ethics! /s
I just wish we could put this dumb sentient AI stuff to bed for now. Congrats, you guys built the most powerful (and wasteful) auto-complete. It's not sentient and your marketing doesn't have to be hyperbolic lies.
The researcher didn't make an LLM, he made a NAND gate, and in theory that could be expanded into an LLM
All computers are just a minimum of three layers of NAND gates
"Man built brain out of simple particles, proving brains are not sentient."
I'm still waiting for a Goatse powered LLM.
I think theres a big hole in that plan.
It would be quite a stretch.
Am I missing something, or is the point simply “I can build a computer program in a game that you may be surprised to learn is Turing complete, therefore ai is not sentient?”
I’m not here to argue that AI is sentient, but this is a really lousy argument.
The argument in the paper is a bit more nuanced - it's mostly arguing that we over-anthropomorphise AI, and this taints out results on AI sentience by pre-biasing towards some human like qualities already. Part of this comes from the AI attempting to emulated a human notion of identity.
Using a contrived goat powered AoE II AI would likely yield less assumed human-like quality bias. It would also be hilarious.
To be fair, humans anthropomorphising everything, of course they do that with software that can fucking talk to them.
It is incredible how we forgotten the whole point of the Turning test the moment it became relevant rather than some abstract though experiment.
People it this thread does not know what the Turing test actually tests. It is Ok, I guess the blame fails on pop culture.
The Turing test is not a test of sentience, it is a test if a machine can fool a human being into thinking it is sentient. And machines have passed this test for decades already. Who would have thought that our predisposition to anthropomorphize everything would make us so susceptible to thinking stuff is sentient 🤷
