Far more broadly, think of logic as just a low-dimensional approximation of something complex. Humans deal with logic and are taught “i before e except after c

Far more broadly, think of logic as just a low-dimensional approximation of something complex. Humans deal with logic and are taught “i before e except after c” etc. But even with describing human languages that may be inadequate. What AI does is it essentially makes a model by which you can search a latent space quickly — both to clasify input and to generate output. You can throw the recorded motions of the planets and stars at it and it might find physical laws that have 80 variables while humans want to deal only with simplified versions like Kepler’s laws of motion. In the vacuum of space those laws may be enough but when it comes to the complexity of chemistry, biology, genetics, politica, etc. the AI might have way better model that we can never understand. Like for dietary recommendations. Would people follow them? And what if they are wrong in some other ways? Like how humans beat AlphaGo through its blind spot or how you can fool face recognition by wearing a hoodie etc.