Subnostr

Replying to

mark tyler

Depending on exactly what the definition of intelligence is, they are empirically intelligent.

And they also definitely can speak in a way that indicates the internalization of logic.

https://arxiv.org/abs/2304.03439

But maybe what I’m saying here does not conflict with what you’re saying?

Kc 2y ago

True. Also, the definition of intelligence changed a lot over time. There was a time when people said it needs intelligence to do arithmetic operations... Then the first mechanical calculators were invented.

Then they said you need intelligence to play chess... But someone later computer programs that played chess were built.

Some people say we can never build real artifical general intelligence because every time we think we did so, shortly after we just take this AI as another tool and move our perception of what AI must be able to do a bit higher. 😂

Reply to this note

Please Login to reply.

Discussion

mark tyler 2y ago

Haha yeah. Gotta separate the concepts of consciousness, which what some people mean by intelligence, and intelligence itself.

I doubt these things will be conscious anytime soon. But that doesn’t mean they can’t outcompete us for limited resources, in short, they don’t need to be conscious for evolution to preference them over us.

Kc 2y ago

Might be. Or maybe not.

In the end all current AI work with incentive or reward systems. And humans do this too. Evolution built us with a reward system that favors behavior that searches for energy sources and tries to replicate and tries to protect the replicated entities (or children).

But usually what we build as AI has a totaly different reward system. They are rewarded if they can distinct between cats and dogs. Or if they draw images that look realistic. Or if they write text that looks like other text from humans.

Why should we regard any such system as being conscious? It does what we want because it's programmed/rewarded for doing so. It won't sometime stop doing it and demand a salary or holidays.

mark tyler 2y ago

Yeah I don’t think they will be conscious soon if ever.

The way you write that though it sounds like you assume we can control the reward function ;)

Kc 2y ago

Oh, yes we can. Because that's how these systems learn. There's no learning without a defined reward function.

On the other hand, we could try to imitate the evolutions reward function. But as far as I know nobody really knows how to do that.