Inference: How AI Knows Your Secrets Without You Telling Them

You've never told Facebook you have diabetes. But an AI model can probably infer it from your online behavior. You've never listed your political beliefs on LinkedIn. But AI can estimate them from pages you've liked and groups you've joined.

This happens through a process called inference—when AI makes educated guesses about information you didn't directly share, based on patterns in data you did share.

How Inference Actually Works

Inference relies on statistical correlations. If AI notices that people who search for diabetes symptoms, join diabetes support groups, and purchase blood glucose monitors all have similar browsing and shopping patterns, it learns that connection. Then when a new user matches some of those patterns, AI can infer they likely have diabetes—without that person ever confirming it.

This is powerful because it reveals secrets you thought you kept secret. You might openly discuss your love of hiking while keeping your anxiety condition private. But if AI notices that people with anxiety also tend to watch certain types of content, buy certain products, and visit certain websites, it can infer your anxiety from your hiking interests and content consumption.

The scariest part: you can't control inference by controlling what you explicitly share. You can't prevent AI from connecting dots you didn't intend to connect.

Real Examples of Inference

Target famously used purchase inference to identify pregnant customers before they'd told anyone. When women bought certain products in certain combinations—unscented lotion, vitamin supplements, certain books—Target's AI inferred pregnancy. They received targeted ads and coupons for baby products, sometimes revealing pregnancies people hadn't yet announced.

Similarly, employers use inference to estimate job hunting behavior, health conditions, financial stress, and even likelihood of quitting. Your search history and app usage patterns can reveal these things without you ever explicitly stating them.

Insurance companies use inference to estimate health risk. Your zip code, online searches, app downloads, and shopping habits all contribute to risk calculations you're unaware of.

Why This Matters for Privacy

Inference creates privacy risks you can't directly control. You could have tight privacy settings, share minimal information, and still have intimate secrets inferred from innocent data points. The problem is that you don't know what combinations of data reveal what secrets to an AI system.

Additionally, inferred information is often wrong. AI might infer you're pregnant, sick, or job hunting when you're not. But companies act on these inferences anyway—changing what ads you see, adjusting insurance quotes, or altering job recommendations. You're impacted by false assumptions about yourself.

Minimal Protection

True protection against inference is nearly impossible. But you can reduce the attack surface:

• Vary your online behavior—don't let patterns form
• Use different email addresses for different activities to prevent data linkage
• Avoid creating public profiles that aggregate your interests
• Be suspicious of "free" services that have no obvious business model—the model is inferring secrets from you and selling them

Try this: Think of one secret you keep private. Now list innocent-seeming data points about yourself that AI might correlate with that secret—searches you've done, products you've bought, content you've consumed, communities you've joined. You'll realize how much an inference algorithm could piece together without you telling it anything directly.