> Consider GO: it is a game of sophisticated intuition
It's still a game that can be described in terms of clear state-machine rules. The real challenge for AI is making sense and acting in the real world, which can't be described in such way. I consider advances in self-driving cars much more interesting in that sense - even if, even there, there are at least some rule-based constraints that can be applied to simplify the representation of the "world state".
Yeah, I think it's accurate to say that driving a car is harder in some sense than playing Go, based on the fact that the AI for Go came first. A lot of people have been working on self-driving cars for a long time now, and it's very monetizable, unlike Go.
Also, it should be noted that in Go, we're trying to beat the BEST human player, and have done so. In driving, we're just trying to be "good enough" or safe enough -- it doesn't have to be the safest driver in the world.
Beating the average human Go player was probably accomplished decades ago, whereas it's not even clear if we're safer than the average human driver (under all conditions).
These tasks are just wildly different, and yes I think it's basically all due to the fact that Go's state is so easily represented by a computer, and the goal is so concrete.
Sort of a tangent from the thread: I get the point about "good enough" at the moment, but I wonder if car AI really does need to perform much safer than any human driver before truly autonomous vehicles should be allowed to see widespread adoption. I'm thinking about the difficult problems re: legal and moral responsibility for human written/guided/trained programs like car AI. As well as the fact that, unlike in Go, real people's very lives are at stake in the program's successful performance. We already seem to have met the requirements for a research project---which is still unbelievable to me!---and I wonder how long the last leg will take.
AI cars could be safer now in most cases by simply not doing dumb illegal stuff.
The real problem is dealing with all the edge cases. Think of this edge case. You pull up to a red light, a guy with a gun starts running at your car in a manner you perceive to be threatening.
You as a human are most likely going to step on the gas and get the hell out of there saving yourself, at some risk of causing a traffic accident.
The car will just sit there till the light turns green while the windows get shot out and you get dragged out of the car.
You're absolutely right. The power budget for these cars is more like 30 watts. A Tesla driving is around 240 watts per km or 14kw per hour at 60kmph. If you're okay reducing your range but half you could get a dozen beefy GPUs if the battery can support that kind of extended load.
I wasn't talking about the implementation, but about the problem space.
Anyway Marazan already pointed that out, but any computer system is a state machine, with 2^N states where N is the number of bits the machine can flip anywhere in its system (RAM, registries, disk, etc.).
It's still a game that can be described in terms of clear state-machine rules. The real challenge for AI is making sense and acting in the real world, which can't be described in such way. I consider advances in self-driving cars much more interesting in that sense - even if, even there, there are at least some rule-based constraints that can be applied to simplify the representation of the "world state".