Hacker News Reader

Comment by berkes | Hacker News Reader

> improve to where

I guess "improve on a vector we put it on".

E.g. the dystopian idea of "Give AI with autonomy some power and tell it to improve itself until it can solve climate change". A nice vector. Except one with a real danger of "solving climate change" by simply eliminating 99% of humans.

Or, more practically, a vector "become better than X¹ at doing Y so our company can take over the niche that does Y and make profit". Improve to write better code, improve to make popular music, improve to write more viral content and so on.

¹ X the variable, placeholder, not the sewer formally known as twitter.

View on HN

If the answer was to wipe out humans, then it would be a pretty dumb solution no?

It does seem like it would be a lot higher effort than others, but if the AI is easily able to engineer a plague maybe it's the easiest path

It's not solving "climate change" though is it?

Depends on the goal.

"Save the earth", then no; it's probably the only viable one even. But quite certainly easier than cutting carbon emissions in societies that don't want to change. "Save humans" then certainly dumb.

Both would be different vectors.

Assuming running billions of GPUs to run AI will actually be better off for the earth?

Yes. Entire IT industry is peanuts compared to what we do to produce food (meat), haul stuff around, make stuff, and haul humans around. It truly does not compare.

Witch does not mean I think we can ignore any win. But GPUs that run AI (or that secure cryptocurrencies for that matter) aren't the low-hanging fruit. Any win is important, so curbing that is important, IMHO. But, again, not the low hanging fruit. That really is flying/driving people around the world, shipping stuff, making stuff and eating (meat)¹.

¹ Emphasizing meat because when talking about low-hanging fruit, that's where the biggest win is. By far.

Well at some stage we have to admit that it all adds up. We have all the things you mentioned, now "AI" and while I agree some amazing breakthroughs have and still might come from "AI". It troubles me that everything is going "AI" and thinking about how much energy that will actually require.

I wonder about this. It never seems to conclude zebras are the problem. It's always humans.

Is eliminating humans the wrong solution, or just one that we don't like?

We see this with crime stats too (and as of late, voting results). Unfavorable outcomes surface, and interested groups immediately move to discredit it.

Because we can't fathom that the results might actually be accurate, nothing ever changes.

Hold on though, we're talking about "intelligence". It would be pretty fucking stupid if "Ultra smart AI" assumed the person who ordered "Please solve climate change so that I can live in a nice world" ended up with a solution which is more like, "I'll exterminate every last one of you...". Seems very frustratingly fucking dumb?

> Seems very frustratingly fucking dumb?

What have we done lately, on a global scale, that doesn't make us as a whole, as a species, not very frustratingly fucking dumb?

We are literally strip-mining and destroying the world that is sustaining us at a rate that means our (grand)children probably won't even be able to eat and live. What else but "very frustratingly F dumb" is that? And this is one example. Another one would be how we allow a few tech monopolies to monopolize our "attention" and all information and knowledge about us. We see it happening but do nothing, and always too late about it. We are unleashing AI systems that we know will disrupt lives, industries, maybe even societies, without anything in place to stop it if it does run out of hand.

We see ourselves as smart. And individually we probably are. But on a grand scale? F* dumb, if you ask me.

We do all of this, because we don't have the collective intelligence to overcome the problem. The point of developing Artificial super intelligence is that it can help us overcome those problems.

Reply on news.ycombinator.com