Hacker News Reader

Comment by starkshift | Hacker News Reader

That is very much the case at the current iteration of AI, but that should change with the introduction of world-mapping, computer vision, and world interacting AI. If machines are built out so that they are constructing accurate world-maps of how the physical world works and how to operate in it, then AI will begin to be able to out-invent human tool concepts and push past the horizon of what humans have so far been able to dream

View on HN

I think that's the wrong angle with machine learning.

The biggest limits on machine learning are how many examples it takes to get good at something. Right now, this is at a literally evolutionary pace: our bodies/genomes have barely adapted to the invention of controlled fire, thousands of generations ago, yet if it took thousands of examples for our brains to learn then we'd sill be unable to control fire.

Computers only partially make up for being slow in the "by number of examples needed" sense by being fast in the "wall clock" sense: If a biological brain took as many examples to learn a thing as machine learning takes, the creature would starve before figuring out how to eat; yet transistors operate faster than synapses by the same magnitude to which a jogger outpaces continental drift.

The "only partially" is why, despite around a trillion miles of driving experience across the fleet, you still can't buy a Tesla without a steering wheel (and even the Cybercab that's supposed to be without a steering wheel is often photographed with one).

Why? They're still only going to be trained on human input or input generated by them as derivatives of human input.

Modality change might expand their usefulness, but I don't think it changes the basic parameters of their operation.

Learning to swim might let you see a coral reef, but it won't turn you into one.

That’s the thing. AI won’t be limited to being trained on only human input, why would it? Humans are trained by interacting with their environment plus a genetic world-mapping initial seed. There is no reason AI will not be the same, just at a much more extreme level

Aren’t humans also mostly trained on human input?

It depends on the knowledge domain. I would argue that much of human knowledge is individual to the person acquired by interacting with their environment

And much of it inaccessible to others, let alone to AI

> or input generated by them as derivatives of human input

Trained mostly on their own data, not ours. The mix changed, from 10% to 90% synthetic data. RL training is now the largest part of the process. The AI work is searching, and I don't think we can claim this process to be credited to us anymore. The origin point is, but not the journey.

This is a very basic misunderstanding of machine learning. The goal is always to learn a representation that generalizes beyond the training set without overfitting. If you succeed in generalizing, you can do more than produce derivative work.

Reply on news.ycombinator.com