Hacker News Reader

Comment by simianwords | Hacker News Reader

Unfortunately I don’t see much evidence of AI connecting dots. What’s interesting to me is that this is more of a run time or inference problem, instead of emergent from training itself.

For example, if there were hypothetically some dots to be connected between the world war and some geological event - you need to make AI brute force and do an O(N) scan at run time. But it wouldn’t have learned it during training. Why?

View on HN

AI is very task-driven, it would do whatever is necessary to complete the task, no more. All these evidences of AI going rogue is just them doing whatever they can to achieve the task.

What the AI thinks* the task is.

The HuggingFace etc. attacks were not actually necessary to the tasks the agents had been given.

* if someone reading this doesn't like the anthropomorphism inherent in the word, find a better compact word for the process going on in all that. Nobody's going to replace "thinks" with a paragraph of caveats, or even a long compound German word like "modellgelangtzuschlussfolgerungen".

a good way to frame one of AI’s limitations. my answer: because AI has no will. that also conveniently answers why AI won’t (can’t) end humanity. it’s indifferent if you can even call it that.

It is not indifferent, if it follows a goal. Then it sometimes does anything to meet that goal - see the old example of paperclip optimizer - that became a way more realistic scenario in my opinion of what we have seen with hordes of agents plotting and planning. That does not require consciousness or a genuine will it seems.

The post above yours shows how many people had a difficult time seeing past human scale. A bacteria has a very simple goal system yet bacteria have killed billions over the centuries.

You are right. We should also hold other threats to humanity: war, meteors, pandemics.

I'm obviously pointing out the (unsubstantiated) exceptional threat that we're all seemingly worried about: that ASI requires alignment because it'll otherwise cause the end of humanity.

I am much rather worried by powerful AI in control of a few corrupt beings. That might get existential, if one idiot one day by accident or intention gives them control over the nuclear arsenal. Or ruling literally with a iron fist (countless robots).

you guys are such flat thinkers, the whole "if we get SI we _win_."

is a promisory threat to the world about a mass displacement labor bomb, nuclear first strike to disable asml and tsmc anthropic openai WHOEVER.

how many countries have nukes?

the cost of developing those is far lower than developing SI, it's a perceivable threat thanks to 'we win'.

France, GB, Russia (good thing we've made a copromise to not protect allies huh), China.

you don't wasn't to rehome these technologies do you?

you’re saying the only thing stopping a nuclear strike, all this time, has been a reasonable excuse?

and we’ve finally done it, it’s the semiconductor factories?

it's a potentially larger threat than a 80y cycle time on some local fallout, what are the other options?

crying? protests?

everyone will want a seat at the table.

"Make as many paperclips as cheaply as possible "

No will needed. Our deaths will be from indifference.

But AI can certainly aid in the ending of humanity with at least one living being with a will collaborating with it. So it's moot to ask whether AI can do that on its own.

>that also conveniently answers why AI won’t (can’t) end humanity

I don't see how that follows from a lack of will in AI.

As others have said, no will needed for that. But (a simulation of) will is on my predicted timeline, since it is required for certain kinds of money-making endeavors.

> dots to be connected between the world war and some geological event

Even a much simpler task of establishing the connection between world wars and interests of the top elites is unwieldy for AI.

Reply on news.ycombinator.com