What AI Knows About You

When you use an app powered by artificial intelligence, there is a model running behind the scenes that has been trained on behavioral data from millions of users. That model is constantly updating its picture of you: what you like, what you fear, what you want next, and how you are likely to respond to different stimuli. The picture it builds is called a user model, and it is often more detailed — and more accurate — than you might expect.

How AI Builds a User Model

A user model is a set of predictions and inferences about a specific person, built by a machine learning system from that person's behavioral data. It grows every time you interact with the service. Signals that feed a user model include: which content you click on and which you scroll past, how long you pause on an image, what you search for and what you choose from the results, what you buy and what you almost buy, what time of day you are active, what mood you seem to be in based on your activity patterns, and what life events might be happening based on shifts in your behavior. The model does not have a file labeled with your name that humans read. It exists as a set of learned weights and inferred parameters inside a machine learning algorithm — a mathematical representation of you that the algorithm uses to make predictions.

Inference: Knowing What You Never Said

The most significant capability of AI user models is inference — deriving facts about a person from indirect signals rather than from anything they directly stated. Inference allows AI systems to know things about you that you have never consciously shared. Researchers have demonstrated that AI models can infer political affiliation from music preferences, health conditions from purchasing patterns, sexual orientation from social media activity patterns, and emotional state from the timing and speed of typing. None of these inferences require access to private messages or confidential records. They emerge from behavioral patterns that seem innocuous individually but are highly predictive in combination. This means your user model may contain sensitive attributes — things you might consider deeply private — even if you have never disclosed them to the platform.

Recommendation Systems: The Most Personal AI

The AI system most people interact with most often is the recommendation system — the algorithm that decides what video plays next, what posts appear in your feed, what products show up first in a search, and what music starts playing when your playlist ends. Recommendation systems are explicitly designed to model your preferences and predict your desires. They are trained on what you have engaged with in the past to predict what you will engage with in the future. A recommendation system that has watched your behavior for months knows your content preferences with a precision that would be impossible for a human to match. But recommendation systems also have a structural bias: they are optimized for engagement, not for your wellbeing. Content that triggers strong emotion — excitement, fear, anger, curiosity — tends to generate more engagement than content that is calm or moderately interesting. A recommendation system optimizing for your clicks may systematically surface content that makes you feel worse in the long run.