re: DeepMind's AlphaZero AI just taught itself chess in a few hours

quote:

---

Pretty cool/interesting/scary/wtf

---

It seems scary, but from my (admittedly limited) understanding of the state of AI, we are still a long way away from actual scary intelligence.

From the little bit I've looked into it, machine learning and neural networks seem to be just an interesting and effective way to solve problems. It's basically trial and error until a desired result is produced or goal achieved. AFAIK, nobody is close to producing an AI that can actually set its own goals.. we still have to set the parameters for "success" and the rules of the game.

I will start to get scared when someone sets a scary goal for an AI, gives it the tools (or means to acquire the tools) it needs to interact with the world outside its own system, and also a way to gather the inputs it needs to determine whether it is closer to reaching its goal. But if someone were willing and able to do all that, there are easier ways to achieve whatever it is they want to do.

I will only really get scared if/when an AI is ever able to actually set a goal for itself.