Hacker News Reader

We know what we want it to do, we want it to have some basic ability to think for itself. That is something we just simply can't do. Back propagation is far from a spanking for acting out of line. AI has no ability to understand it's acting like a fool or how to deal with uncertainty with emotions which cause us to act without regards to consequences and in many cases reality. It lacks understanding of what future consequences its trying to prevent such as our daily decisions to get up and go to work each morning. The AI has no understanding of it's future and the consequences of not going to work until it's fired 40,000 times for not showing up or it's children are taken from him/her/it.

I'm glad people are finally waking up to the fact that AI is not ML and AI is all hype at the moment. Google used algorithms quite effectively to adapt and learn, but they have no greater understanding of what we want, just what we and others have wanted in the past.

I agree that defining a goal along with metrics would be very helpful to make meaningful progress towards AGI. However, defining this test is extremely hard, to a point where I'm not sure if we could define a test like this. So far it seems, that just by defining a test we can come up with a narrow AI that optimizes for this test.

By this argument, humans also aren't general intelligences, because we haven't been able to write one yet either.

>We could say that an AI program should pass the Turing Test and many have written programs that more or less pass it. But so now, write a program that writes several different programs that all pass the Turing Test one better than the previous one.

That's essentially what machine learning is, though.