It's able to create knowledge itself

[Bob the Unbeliever]

Whoever calls GO a "very limited game" has never played GO.

I was thinking in comparison with reality, or chess. But I understand your point; yes, I have played GO, but it was decades ago...

 

[Nakosis]

I think the title is a wee bit of "click-bait" here.

I think the term "human knowledge" is a misnomer, in this context, as the entire context is limited to a game, which has a very short list of rules (which is in direct contrast to reality, which has a seemingly endless list). It's not surprising that a computer could derive all possible moves in such a limited scope-- a computer does not forget--ever-- once something is "learned", it has perfect recall. Thus it never repeats the same mistake, or even an experiment, a second time, it has no need to do so.

Neither does it need to sleep, to re-organize it's total experiences from short term memory to long-term, in fact, there is no real difference in all it's memory.

Finally, being electronic, it "thinks" at nearly the speed of light, again, in contrast to humans who only think at the "speed of chemistry".

I'm not remotely surprised at this, and I fully expect other games to be 'conquered' by computers as they grow more capable.

From the article...

“AI fails in tasks that are surprisingly easy for humans,” she said. “Just look at the performance of a humanoid robot in everyday tasks such as walking, running and kicking a ball.”

Yes, well humans rather clumsy attempt at trying to program computers to mimic human processes.
This is a self learning algorithm where one only needs to input the basic rules for it to surpass human knowledge in the field.

It wasn't taught how to play the game, it figured that out from the rules. Physics has rules. Lets program the rules/laws of physics and see what knowledge it can derive. We don't have to teach it physics, all we'd have to do is input the laws of physics. It could tell us what is true about physics well beyond what any human is capable of. Humans come up with theories and then have to test them. These theories could have all been tested, validated or disproven well before the time a human scientist had even finished developing their theory.

 

[Heyo]

I was thinking in comparison with reality, or chess. But I understand your point; yes, I have played GO, but it was decades ago...

The difference between chess and GO is that chess is complicated (many rules, different pieces) whereas GO is complex (high number of possibilities).
During a chess game you have to ponder on average 20 possible moves per move, with GO it's 180.
The number of possible chess games is about 10^120, with GO it's 10^170.
That's why chess computers could beat humans on a master level in the 1990s with algorithmic programming while AlphaGO only recently reached that strength with a neuronal network approach.