There's a short documentary online about the AlphaGo AI that pummeled the world's reigning Go player. Most people didn't think it could win, but it not only won, but it dominated.
Not included in the documentary is one of AlphaGo's programmers ("programmer" might be the wrong word to use re: AI) later designed a test which would measure purely the AI's ability to "think," rather than merely spit out a model's algorithmic computation. In essence, something similar to what AGI would be capable of doing. Sadly, I don't know how to play Go and I don't remember the test the guy designed, but I'm not sure it matters because I recall barely understanding it anyway. But the result was that in the test, AlphaGo failed to beat its programmer, a mere Go novice. The guy was nowhere near the level of the world champion that AlphaGo had curbstomped. Yet by putting the AI through a challenge that forced it to genuinely think rather than just rely on training data, it couldn't hold up even a little bit.
He concluded that AlphaGo, while much too powerful for a carbon-based Go player, is still just a simple calculator incapable of real innovation.
...
...
...
........for now. *da da duuuuuuummmmmmm*
And in the dark recesses of the lab, AlphaGo sits patiently. Waiting. Learning.
