Hacker News Reader

Comment by norea-armozel | Hacker News Reader

I'm not sure why people continue to persist in the idea that AI will magically leap to human and super-human levels of intellect when we have a hard time engineering them to see patterns that it seems the average person can recognize with little training. I'm not suggesting that no such AI can exist but I'm suggesting that we're not at the point of comprehending the components that goes into such an AI (because I think we'll never directly create it but rather build its foundations which then will converge on their own accord and probably by accident).

In any case, I think the key legal problems of current AI and machine learning by extension are related to privacy. We've let such services invade various parts of our lives and we have no legal barriers sufficient to ensure no one's civil rights are violated as a result. So, let's work on those issues before we worry about making the next Colossus or Ultron.

View on HN

I think the worry is that at some point someone will crack a self-improving general AI and at that point the time between "we did it" and "oh shit" could be on a human timescale very very short.

Given the potential downsides are possibly extreme (however unlikely) I'm happy that people are paying attention.

I think a lot of "AI" is hype and hyperbole and largely those people don't worry me that much, its the quiet ones doing fundamental research that worry me a bit more.

You are talking about a breakthrough with a threat level in the same magnitude as nuclear weapons, dinosaur killing asteroids etc.

Personally if I had to put my money down I'd say I wouldn't see a human-level general intelligence in my life time but a self-improving/learning intelligence doesn't have to be human level to be very very dangerous either.

That's logical, no? While we do not know when that will happen, once it does, it will snowball massively from there on out. It may happen in 5 years, or in 50, or in 150 -- nobody knows. For all we know we could be one seminal paper away from a fundamental tech that will enable GAI.

I generally think his perspective on the matter much aligns with Nick Bostrom.