Hacker News Reader

Comment by petilon | Hacker News Reader

At what point can we say AGI has been achieved? What is the test? AI is solving mathematical problems that humans have not been able to solve for decades. Is that not enough?

Sam Altman has said "If superintelligence can't discover novel physics, I don't think it's a superintelligence." Is that the test? How far away are we from AI discovering novel physics? It seems within reach.

View on HN

I think we previously assumed that AGI would need to come before superintelligence, but it kind of seems like that's wrong? Super intelligence in a narrow field has arguably already arrived (solving problems that were previously unsolved), but we are still pretty far behind human abilities in things like spatial reasoning or computer use.

It’s artificial, it’s general, and it’s intelligence. The people who believe “AGI” is an important and unattained goal need to start coining and defining their terms better.

A true AGI will continuously improve itself without periodic retraining from scratch. Just like humans.

That’s just an assertion. Why is that the “true” definition?

It depends on your definition. For me it would have to be able to do the stuff humans can do like make a cup of coffee (Wozniak test).

Just maths isn't really general enough for the G in AGI.

It can give you detailed instructions for making a coffee. Is that not enough? Actually making coffee requires more than intelligence, it requires eyes and limbs (i.e., robotics). Think about a human that is blind and does not have limbs. Does he not have natural general intelligence, even though he is not able to make a cup of coffee?

The Wozniak test has it with a robot body going into a house, finding a coffee maker and making a cup. I guess you can vary the rules as you like.

That's a good test for a robot, not for AGI. AGI should test only intelligence and should not require limbs.

Okay we can make a completely digital environment for it to make coffee in then. It would still fail unless you let it randomly try every combination potentially thousands of times until it stumbles upon the right path. It doesn't take intelligence to read off a recipe, it does take intelligence to read a recipe, understand it, adapt it to your specific tools and materials on hand which may differ from the recipe, and then actually still accomplish it in the first or maybe second try.

Why do you think it would fail? I think it’d be zero shot.

Reply on news.ycombinator.com