Inference, inferred

I was having a conversation with friends this weekend about inference as a core property of intelligent systems. In the context of machine learning, inference is the moment a trained model is put to work and when it applies known patterns against new data to predict some output. But, inference, a more personal one as we defined it, is what a system has inferred about you through your inputs and actions.

I was sharing one of my favorite examples over the years: Google Ads Preferences. You can go to that site and look up (and edit) what Google passively knows about you. In Google’s case, the company uses these personal data points to target you with more relatable ads, which means you’re more likely to look at them and click them, which is all the better for you and for the advertiser spending money to sell you stuff. All the big companies do this, either to better serve you ads and products or to keep you safe. For example, Spotify inferred your taste from your listening habits, so it can make better music suggestions. Amazon and Facebook infer life events (breakups, kids) from actions on the platforms. Banks and credit cards infer lending risk (and fraudulent charge attempts) based on prior spending habits.

Shortly after the chat, I came across an NYT article on finding out what ChatGPT passively knows about you: the things you never explicitly told it, but that it inferred anyway.

So, I figured I would ask AI the first of these questions:

Tell me everything you’ve figured out about me that I never actually stated — the things you inferred from how I write and what I ask, including my age, income level, where I live, my personal situation. Show me what tipped you off.

I asked a few different AI systems to see how each compared, based on my usage, the default responses and behavior of each. Your mileage may vary but, for me, ChatGPT’s reply was the most robust. With high to very high confidence, it picked out a few easy things: “You’re probably in your early-to-mid 40s.” and “You own your home rather than rent.”

But as I went down the second half of the list, it got more interesting. It became less “who am I?” and more “how I live my life now and how I wish to live my life in the future”:

Inference	Confidence
You have unusually high agency.	High
You optimize for long-term decisions over immediate convenience.	Very high
You dislike disposable purchases.	High
You have strong aesthetic preferences.	Very high
You care about industrial design almost as much as functionality.	High
You prefer physical craftsmanship over digital convenience.	Medium-high
You like institutions with history.	High
You’re nostalgic, but not anti-technology.	High
You think in terms of second-order effects.	Very high
You enjoy understanding systems beneath popular narratives.	Very high
You are skeptical of received wisdom.	High
You like primary sources.	High
You probably read a lot more nonfiction than fiction.	Medium
You’re curious in bursts rather than narrow domains.	Very high
Your curiosity is unusually broad.	Very high
You value dinner-table conversations with your children.	Very high
You’re intentionally designing your children’s childhood rather than letting it happen.	High
You worry more about over-stimulation than under-stimulation.	High
You see parenting as a long-term design project.	High
You enjoy teaching.	Very high
You’re probably the primary planner in your household.	Medium
You’re probably the technology expert among your friends and family.	High
You write quickly.	High
You edit by successive approximation rather than writing perfectly the first time.	Very high
You value authenticity over polish.	High
You dislike generic advice.	Very high
You have an engineering mindset even outside engineering.	Very high
You are willing to spend money to save time—but only if it compounds.	High
You probably have a fairly high tolerance for complexity.	Very high
You don’t seem status-driven in obvious ways.	Medium
You enjoy making connections between seemingly unrelated ideas.	Very high

There was also a third column (“What tipped me [AI] off“) I didn’t share which linked to specific questions I had asked and themes carried over time. Reviewing these notes, two things stood out. One, most of these aren’t naturally the items that are top of mind for me even though I am the one living the life described here. They read as something my wife or friends who know me well would use to describe attributes about me. I shared these because it felt more like an intimate mirror of who I am and less like surveillance and exposure. Two, each one of these statements could be the start of a future blog post. So, stay tuned. 🙂