Broadly speaking, the hygiene hypothesis is the current leading explanation. To make an analogy, in AI/ML terms, it amounts to a training set problem: you're n

Broadly speaking, the hygiene hypothesis is the current leading explanation. To make an analogy, in AI/ML terms, it amounts to a training set problem: you're not exposed to a diverse enough set of microbes and pathogens, or you're exposed to a biased sample, or you're not exposed to enough microbes. Your immune system's role is to act as a classifier to discriminate between good and bad, self and non-self. When you have a bad training set, you increase the odds that good things will be identified as bad, or that self antigen will be recognized as foreign.