[nnq]

...that's one of the reasons why I think 60% of the time we're applying machine learning "at the wrong end of the problem". We should:

(1) pick a set of goals we deem worthy

(2) use machine learning to find patterns of actions that push things towards our goals - that's why I think we should move as much as possible in reinforcement learning, and figure out ways to use RL for data-science too, both supervised learning and time-independent unsupervised learning are really dangerous and biasing tools, and I think they're sort of dead ends on the road to true-AGI

There's no such thing as "seeing the world for how it is". Looking for patterns in data using a tool/machine trained on data (that will be inherently biasedly selected) will just lead you to find the patterns that confirm your f up world view, out of the infinity of patterns that there always are in everything...

We need to figure out how to manipulate the world in order to shape it into the patterns that we'd enjoy more.

There's a sort of "Chinese perspective" in applied machine learning nowadays (which is probably not originating in China, but being associated with "ML empowered surveillance and social scoring" is easier labeled like this) that I think is very wrong and extremely dangerous... we should grow up out of it quickly, because later on properly purging it from our cultures will be a violent endeavor... Heck, I even hope we as species get a Mars base up as a backup if this conflict of world views goes hot because there's a high chance it will be very very hot, like in nuclear hot...