They wield power not available to individual humans. They achieve things individual humans can’t. They get people to sacrifice themselves (up to a literal karoshi) for company’s interests.

Are you sure you don’t do things that appear stupid to your cat?

>They function much more like a drunk, demented DNN optimizing for some unknown set of variables than like a human using its intelligence to solve problems.

That’s a weird sentence. It’s written as if there was some fundamental difference between the two.

> They wield power not available to individual humans. They achieve things individual humans can’t. They get people to sacrifice themselves (up to a literal karoshi) for company’s interests.

Do these have something to do with intelligence? I see power and intelligence as largely orthogonal. Ditto control and intelligence.

See Donald Trump during his presidency, for instance. Mountains of power, control of the nuclear arsenal, massive private wealth. People threw their careers away to tell lies for him.

Still an idiot.

> Are you sure you don’t do things that appear stupid to your cat?

Not at all. I see your point here and it's a valid one (though I still can't do anything but reason from the reference frame I've got).

Though I'm not sure my cats even have a category for "stupid".

> > They function much more like a drunk, demented DNN optimizing for some unknown set of variables than like a human using its intelligence to solve problems.

> That’s a weird sentence. It’s written as if there was some fundamental difference between the two.

I'm not sure there's a fundamental difference, but I'm also not convinced there isn't.

Certainly the output I've seen from image recognizers, GPT-N, image generators and similar have seemed recognizably different to me from skilled human makers.

They're occasionally quite good at aping a particular style or topic, but I don't remember seeing anything so far that reminds me of the flashes of inspiration, insight, and understanding I'm used to seeing from other humans while they're problem-solving.

I've also seen people much more informed than I make cogent-seeming arguments that DNNs are necessary but not sufficient for something like human intelligence, and that we'll need other paradigms to combine with them before we really see progress.

I've seen the opposite claimed too, in a reasonable and coherent way - that DNNs are all we need and they just need more scale than we've been able to throw at them so far.

I'm not sure which stance I think is right (if either - not too sure human intelligence is actually a tractable problem, given that we don't even have a coherent definition of it, nor a clear idea of whether the not-understood nature of consciousness may have anything to do with it).