The biggest issue with AI is not that it is stupid but lack of definition for intelligence and hence lack of measure for it [1] [2].
Turing test is not a good measure because gorilla Koko wouldn't pass but she could solve more problems than lots of disabled human beings [3] [4].
It is quite possible that people in the future will wonder why so many people back in 2019 thought playing Go and other games in fixed simulated environments after long training had anything to do with intelligence and wasn't just a search/optimization problem like many others in computer science.
Intelligence is more about adapting/transfering old knowledge to new task (playing Quake Arena quite good without any training after mastering Doom) than it is about compressing experience into heuristics to predict outcome (playing Quake Arena quite good after million games after mastering Doom).
Human intelligence is about ability to adapt to physical and social world, and playing Go is a particular adaptation performed by human intelligence, and developing algorithm to learn to play Go is more performant adaptation, and developing mathematical theory to play Go might be even more performant.
It makes more sense to compare a human and AI not by effectiveness/efficiency of end product of adaptation (in games played between human and agent) but by effectiveness/efficiency of process of adaptation (in games played between human-coded agent and machine-learned agent after limited practice). Читать полностью »
