But the challenge rate is not fixed. It increases at higher levels. So the lesson seems rather hollow: At some point, if you are successful at solving challenges, the rate at which new ones appear becomes too high for you.
...thus becoming useful object lessons to the rest of the species, and reducing our average susceptibility to reward systems with low variability. Not quite seeing the problem here.
But the challenge rate is not fixed. It increases at higher levels. So the lesson seems rather hollow: At some point, if you are successful at solving challenges, the rate at which new ones appear becomes too high for you.
Just like life. The reward for succeeding at a challenge is always a new, bigger challenge.
At which point you die, for lack of intelligence.
Actually a fairly good metaphor for x-risk, surprisingly.
Of course, it’s a lot easier to make a Tetris-optimizer than a Friendly AI...
I thought Tetris had been proven to always eventually produce an unclearable block sequence.
Only if there is a possibility of a sufficiently large run of S and Z pieces. In many implementations there is not.
It was either that or risk some people playing without stop until their bodies died in the real world.
...thus becoming useful object lessons to the rest of the species, and reducing our average susceptibility to reward systems with low variability. Not quite seeing the problem here.