Egregoros

@jeremiah @Omega_Variant @cjd @nimrod @sun ok i didn't say AI was a brain, but brains *are* neural networks. it's more than semantics. also look at the definition of reason: "think, understand, and form judgements logically" the understand part is what's really in dispute. i would argue that understanding concepts is a requirement to engage in reasoning, and they demonstrate that ability.

forget about marketing. i'm looking at this purely from the perspective of what they can do. what useful work can they perform, based on my own experience.

@meowski @Omega_Variant @cjd @nimrod @sun brains are not neural networks in the same way an ANN is. This is another semantic hinging case where the terms "neural" and "neuron" were chosen, even though there's only a basic conceptual similarity between organic brain neural systems and software neural networks. A digital neuron != a brain neuron, and an ANN != the brain.

From a pragmatic perspective, sure, these tools can do jobs that previous software systems could not do. Somewhat. With a great deal of supervision, overhead, catastrophic and comedic misfires, and eventually deskilling people.

@jeremiah @Omega_Variant @cjd @nimrod @sun my dude, your brain IS a network of about 100 billion neurons with trillions of connections between them.

a digital neuron is modeled after a biological neuron. i would never say it's the same but the general idea is similar. some neural pathways are reinforced and others are discouraged as part of training a neural network. just like a child learns with positive and negative reinforcement. same basic idea.

people get hung up on what intelligence is- they think it means alive or having emotions or self awareness. we are going to have to relax this definition of intelligence because it's becoming undeniable

@jeremiah @Omega_Variant @cjd @meowski @nimrod I odn't think you have to to prove it learns. at amy rate II think our disagreement is I'm not bothered if it doesn't work exactly like a biological brain if you can prove it practices reinforcement training for example. the irony here is, this isn't how LLMs work at all, and if you look closer at "reasoning" LLMs, they do in fact do something that isn't really like human reasoning at all and arguably isn't any real kind of reasoning at all either.

@sun @Omega_Variant @cjd @meowski @nimrod i'm not sure it's even a disagreement; pragmatically, it doesn't matter if the fly and the simfly behave similarly in different environments as long as they learn a coherent strategy for existing in them, and if our theories about fly brains are accurate, the simflies and flies ought to behave very similarly. Good enough, move on.

I think if there's friction, it may be my autistic insistence on the importance of domains/categories bound to terms, and my observation that when terms hop between domains, they often carry inappropriate baggage that has knock-on effects (in culture, importantly.) Sort of like when someone says "your car" and is referring to your truck. Pragmatically, it's good enough, but it is not accurate.

@jeremiah @Omega_Variant @cjd @nimrod @sun machines are already physically stronger than humans and gorillas in so that's not even relevant. there's no question that a hydraulic press can exert 100 tons of pressure and no biological system can come close to that- but this doesn't threaten you psychologically. you just accept it as fact.

a $5 calculator can do calculations faster than a human too. i'm not saying that is intelligence. again it's the ability to learn. we're not even debating whether machines are smarter than humans or gorillas. we're simply debating whether they possess intelligence at all, to get to your original premise, and i say they do.