AI is waaay more complex than that.

in fact, no one really understands how neural nets works when it does work. It just that it works

unlike programming which has codes and codes and codes of shit, and when the program works, the output works


for AI, the output works. the program? we dont really understand why it works, just that we know that, if we can feed the AI with insane amount of unique and good data, it will work. it just works.

people do understand like, approximately how it works.

but for example, when we created the neural net DeepMind AI which beat alpha go just by playing by itself, no programmer can understand what strategy deep mind employed nor "how" it thinks.

its just that deepmind AI is so good, it can beat the best go computer 100 to 0, and that the programmers just:

1) Thought it the rules of the game and win/lose condition
2) Let it play by itself X times.
3) Run above for 3 days.

and within 7 days, it beat the best computer alpha go, which beat the best player in the world.

the programmers or players wants to analyse deep mind, neural nets is such that, there is no way you can extract how the computer is thinking nor how it strategises. its just that, it wins games. it just wins.


Similarly, when we solve self driving, or when we use AI to allocate uber drivers jobs and delivery paterns, the neural net cannot explain why they work and why they are so efficient. it just works.

And when something goes wrong, you dont "go into the code and fix the code". you basically label them as wrong and pump more wrong data into the system, and it will "recallibrate its matrices" and eventually fix itself.

This post has been edited by Nachiino Etamay: Feb 19 2020, 12:17 PM