Most modern data collection uses embedding vectors. Those vectors are just a short list of numbers (for example, 128 decimal numbers).<p>However, <i>from</i> those numbers, one could make a pretty accurate guess about your sex life, political alignment, desire to buy a lawnmower, or how many rotten teeth you have.<p>In fact, even the developers of the system don't know what the embedding vectors will be useful for. They just know that if they take a bunch of data and create a user embedding vector, then that vector typically doesn't count as PII, yet can be used for almost all moneymaking schemes.