> Point being — can ML techniques be used to reverse hashes or find collissions? Immovable object vs irresistible force? That's actually a good philosophical o

> Point being — can ML techniques be used to reverse hashes or find collissions? Immovable object vs irresistible force? That's actually a good philosophical observation. And there is an answer for that. If you try to use ML on an encrypted data set then it will blow up and won't converge. The encrypted data contain a lot of randomness while the ML would assume the presence of a not insignificant amount of continuity, linearity and correlation. Trying to train a network on encrypted data would blow it up as it would not be able to find a convergence point in any observable time. The effort would be equivalent to trying to bruteforce the encryption. But if you have a few million years, a few PBs of storage and some extravagant network training algorithm based on mutations, who knows.