>>12072894In Go, Chess, etc the physical nature of the board and pieces doesn't matter, only the abstract state that they represent.
This state can be given an explicit logical representation that the AI can work with.
Imagine instead, if a Go, Chess, etc AI needed to use cameras and robotic arms to read and manipulate actual physical pieces on a board, and use microphones to listen for when turns end, etc.
>>12075042>and humans are unpredictable drivers.It isn't even just humans.
Driven cars could be given some kind of system to communicate with driverless cars to make it much easier to handle.
The biggest problem is all the other shit going on.
What happens if a cat runs into the middle of the road, and the AI hasn't been trained on cats?
What happens if a black plastic bag is blown into the middle of the road, and the AI hasn't been trained on black plastic bags?
Do you run over the cat?
Do you stop for the bag, forcing everyone else to stop too?
How do you tell a cat and a bag apart?
What about every other possibility for random shit that could end up on the road?
There will always be something that the AI hasn't been trained on, and the behavior will basically always need to err on the side of safety, but if you need to shut down the entire road every time the AI doesn't know wtf it's looking at, it risks getting very inefficient.
