Virtually everything about you is encoded in virtually everything you do. With enough intelligence and enough data, it's probably possible to determine with a high degree of accuracy what kind of mint you prefer (spearmint, peppermint, that other one) based on hearing you talk for 30s.

In a world with AI systems that can be trained in an unsupervised way against basically all the data we can collect, the amount of information that can be accurately predicted about you is probably proportional to the number of bits of observational data about you available to to the AI, and I would suspect there is a roughly logistic relationship between bits available and the % of information it can reliably guess.

I'm pretty sure you're right. There are a vast number of correlations waiting to be discovered. Were just waiting on sufficient data.