Actually AI as it currently exists as a technology is completely deterministic. Given the same model, the same training set, the same pseudo-random number generation seeds, and the same input data, you will get the same output... exactly, simply because every element in it, including the ‘random numbers’ that may appear in different parts of the algorithm are all deterministic functions. Only logic races on parallel hardware may violate that, but that’s considered ‘bad design’ practice and mitigated in modern frameworks such as Pytorch and TensorFlow.