>>6767
>THOSE PICTURES ARE FAULTY THE ROBOT IS LOOKING AT FAULTY DATA
To me it looks like very small segments of some graph. We don't know what those are at all, aside from what it indicated by those repeated lines. I think this illustrates how AI learns, by adding an immense amount of data points. Compared to a human or animal child who will say some gibberish and instantly be corrected by his surrounding social interactions, the AI can't know what is right or wrong from the "social cues" very easily. There is "reinforcement learning" where the AI is given a "punishment" for bad learning, but then you'd have to tell it why it's bad, and in the end that's just an arbitrary filter. The best AI is learning blindly (like a virus) where the best DNA is the one that makes new connections to new data, and that is how it slowly maps everything. The only "punishment" is not finding more connections, but as the AI is being fed electricity it will keep running eternally and if it gets stuck there are ways to program it to get out of that route. (Early recommendations on youtube for example had this problem very clearly, and it's still an issue after that loli-alley was exposed; click once on a certain type of video and youtube would only recommend the same kind after that, but never search for it and no one finds them. It caused an outrage like 3 -4 years ago.)