“Is AGI Conscious” is a question that makes zero sense

[.]

Feels like one of those existential questions you start thinking in circles about and just become depressed.

[edmundfreeman]

I have a thought experiment here. Suppose aliens landed in Central Park. They aren't human so they won't have human consciousness, but we will say they have a valid non-human form of consciousness that will deserve as much respect as human consciousness. Likewise, I don't think AGI will ever have human consciousness, but could have a non-human consciousness as worthy of respect as our own.

[milesoflead]

Great video. It sounds like using 3 nebulous terms for something pretty nebulous.

[JohnWilliamDye]

This is an excellent take that encourages more investigation and curiosity. The ChatGPT subreddit is a nightmare of thought-termination for anyone who has picked up a book on neuroscience or AI.

[Lantalia]

That the question is ambiguous due to the concepts being ill defined seems secondary to the xy problem: If you had an answer to the question, what would it change? Or, why are we even asking the question?

For me, it _feels_ like people are trying to ask "Does this entity have moral weight/value?" otherwise it seems like a useless pzombie question

[.]

We know what consciousness is, as it is the state of being aware, which everyone has first hand experience in. We know that other people are conscious because they are also human, but there is no way to definitively prove that with tests unless you define consciousness as being able to pass those tests. Same thing with AI if it is able to pass all those tests. AGI is an AI that can match humans at any intellectual task, or as surpassing humans at most economically viable work. The second definition may be not far in the future, but the first one is hard to conceive-- that AI could outperform us at every single possible thing.

[wtcbd]

I'll consider it AGI, if not only it can do math, answer any question, fool us as to if we're talking to a human, but also be able to create something from scratch, without being prompted. As if we find out it was working on something behind the scenes or revealed something that we haven't even thought of. Consider how hard it is to have a robot that can do surgery, but you can't hand it an object like you would a child without it freezing up. AGI means not only doing the smart things, but also the simplest things, like handle an egg (Robot form).

[chrisf]

I think it is all about behavior. When we look at humans for signs of consciousness we look at the way that they behave, because we really have no idea of what is going on in there minds. So if we can interact with ai in every way that we interact with humans it doesn’t really matter.

[Timely]

Well I do think we have a solid definition on what consciousness is, but the only issue is we do not know enough about it. There are thing we still have yet to understand which could change what we know about consciousness. as far as we know consciousness is just being aware, this is why we need a greater understanding of it because with this definition the chat gtp 4 that improves it self would technically be conscious, when it is mostly not. Animals are conscious as well but compred to human, we have more consciousness as we are aware of our death our body etc...

Not saying this is cut and dry and we know everything, there's much more to learn about consciousness. As of now tho we a little understanding what it kinda is, but I think there's a missing key. anyways hope this helps and I agree if we don't fully understand consciousness, we can't define what AGI is as AGI is related to consciousness.

[SipMyCharlatte]

Vague yes, but I think they're usually referring to if agi is capable of experiencing humanlike emotion.

[alistairlacaille]

I completely agree... But we still have to ask the question though. Makes for quite the conundrum.

[bizlarry]

I know this is so superficial, but you mentioned it first... Can't ignore your hair, it looks great. You're so incredibly smart, I'm only getting about 40%> I'm hoping the remainder 60% I'll get through osmosis.

[kingrutse]

Anything that can ask a question is alive

[tazzledock]

Yes you're right. It's kinda vague. I appreciate your take on this.

[rodbotic]

I thought it did come across in the livestream. But a short direct video on the topic is also needed.

I get asked this question a lot. Usually I redirect and ask how do you even define consciousness.

[DarkSkay]

Humans can replicate the functioning of a Turing machine - it is unknown, if the reverse is possible. Probably not.

Assuming this was possible, it seems to strongly and directly imply that (in theory) it should then also be possible to create consciousness with paper and pencil. So, again: probably not possible.

[petr-heinz]

I wholeheartedly disagree. Papers written on the subject should come with their definition of the terms.

Having ambiguous terms in a question doesn't make it invalid or non-sensical.

Almost nothing can be fully defined, and people will not agree on those definitions. This however, does not make questions useless.

[roguedogx]

So it's basically impossible to definitely say anything about the topic beyond "I don't know". Lol

[ecospider]

I think it also the wrong question to be asking. I want to know how useful an AI is. And I want to know how dangerous an AI can be. If it doesn’t fit someone’s definition of conscious it can still be dangerous.

[dmarsub]

Comment recycling ♻️:

"I really liked how you we carefull with your statements and following the hype.

Most A.i. researchers have absolutely no idea what they are talking about in regards to philosophy and consciousness.

Their expertise in their field doesn't help them with understanding this problem.

Function alone will never prove consciousness.

Most of the tests and comments in this video sound like philosophy 101 students tbh.

If current a.i. does have consciousness there is no reason to assume it is connected to our input it's output it's function, understanding of the world or perceiving the world, or obviously not wanting anything.

An a.i. wants to follow it's reward function just like a gene wants to survive, it doesn't the version we have just happem to be the one that survived dna doesn't have consciousness, still it can create humans.

If current a.i. has consciousness we live in a world where lightswitches have consciousness. Ever single function of every single a.i. based on our current technology can be recreated with switches of water and wood no electricity needed the only difference is it is slower.

If we have a new technological approach to computers that might change.

If you wanna learn about the problem of consciousness don't read papers written by a.i. researchers ^^

Multy modality doesn't change anythong qualitatively still just 1/0 and probabilities but yes it will be harder to convince people of that.

Edit: the test you did at the end was pretty funny, but it was the result i expected current a.i. are literally trained to be people pleaser.

Also it is effectively a basic logic riddle and doesn't require theory of mind at all and a.i. is great in basic logic riddles. But i understand why you think that it implies that.

Emergence andComplexity is part of my field and Consciousness research is closely related. I am not just talking overconfident bs here, but i have to simplify because it is a youtube comment i type on my phone, and i am emotional about it."