Precogs exist, but they're not people. . .

They're algorithms.

Pattern recognition algorithms can know things about you before you do. Tracking your daily routine: where you go, what websites you access and when you access them, your purchasing habits, your eating patterns, your social engagements, your work habits, etc., which thankfully are all in disparate and disconnected databases, if they're tracked at all, could be unified and mined to profile your psyche for comparison with other people's habits and mental states so accurately that it could actually provide foresight to actions that you have not yet even considered taking.

The amount of data we store about ourselves - all of which can be considered behavioral data in some way - is so vast that we absolutely depend on pattern recognition algorithms to make sense of it.

Today our algorithms are primarily based on statistics. Numbers can be mathematically analyzed for normality and variance. What other considerations could make pattern recognition algorithms more effective? Like the human brain making a discovery, a prudent algorithm will consider many patterns when learning about behavior. Even pruning data, like the human mind forgets, can be an important factor. How should an algorithm decide when it's time to forget old data, or lessen their priority?

We can even employ several algorithms at once to produce independent results that can then be analyzed by a supervisory meta-algorithm. The work can be farmed out to several different computers, several different databases, as long as this meta-algorithm is orchestrating the work.