I WENT LOOKING
It Knows What I'll Do. It Doesn't Know Why.
Spotify played the exact song I was about to go looking for.
Not a song like it. That song.
A few minutes later, TikTok served me a video I watched twice. Then Amazon suggested the thing I had almost bought on Tuesday.
And later that week, someone who has known me for years said, “I knew you were going to say that.”
She was right. I had been about to say that.
There’s a small jolt when something predicts you well. It feels a little like being known.
It also feels a little like being caught.
When does knowing our patterns become knowing us?
If something can predict what I’ll click, buy, watch, skip or want next, what exactly is missing when I say, “Yeah, but it doesn’t understand me”?
So I went looking.
It’s better at this than you’d think
In 2015, researchers compared personality judgments made by a computer with judgments made by the people in participants’ lives. The computer had one thing to work with: Facebook likes.
With about 10 likes, it judged personality better than a coworker. With 70, better than a friend or roommate. With 150, better than a family member. With 300, better than a spouse (Youyou et al., 2015).
So if the question is whether a machine can predict us, the answer is yes. Sometimes better than the people who love us.
But predicting and explaining are different jobs. Researchers who build predictive models draw a line between them: a model can forecast what happens next with impressive accuracy and still tell you very little about why (Yarkoni & Westfall, 2017).
Spotify doesn’t know why I needed that song.
It knows that people who played what I played tend to play it next.
Feeling understood is its own thing
This is where it gets strange.
You’d think feeling understood would depend on being understood correctly. Mostly, it doesn’t. Relationship research finds that how understood we feel is only modestly related to how accurately the other person understands us (Reis et al., 2017).
In one study of 199 newlywed couples, researchers measured two things: how accurately spouses knew each other, and how much they felt they understood each other. Only the feeling predicted how the marriages were doing. The accurate knowledge didn’t (Pollmann & Finkenauer, 2009).
Being known and feeling known aren’t the same.
And it’s the feeling we seem to live on.
We’re not great at this either
If understanding were only about accuracy, people would lose to the algorithm.
We’re not especially good at reading each other. Across 25 experiments, researchers asked people to predict what someone else was thinking, feeling or preferring. Some were told to imagine the world from the other person’s point of view. It didn’t help. If anything, it made them a little less accurate (Eyal et al., 2018).
What did help was simpler.
Asking.
When people could talk to the other person and hear it from them directly, their accuracy went up. The researchers called it getting perspective instead of taking it.
The most reliable way to understand someone isn’t to be clever about them. It’s to be curious with them.
So a friend can get our motives completely wrong, and we still leave feeling understood, because she asked, listened and cared about the answer. Meanwhile, an algorithm can predict our next move with unsettling precision and leave us feeling like a collection of data points.
Which raises a better question.
Are we judging understanding by what’s happening inside the other mind, or by how the response makes us feel?
Where it starts to feel creepy
There’s a point where personalization stops feeling helpful and starts feeling like someone read your diary.
Researchers found that point. When people realized an ad had been personalized using information collected without their knowledge, click-through rates dropped sharply. The ads were no less tailored. People just felt more vulnerable. When the same kind of information was collected openly, more personalization worked better (Aguirre et al., 2015).
So it isn’t only about how much something knows.
It’s whether we handed it over.
There’s a second line, too. In a series of studies on medical AI, people were more reluctant to accept care from an algorithm than from a human provider, even when the two were described as equally accurate. The concern driving it had a name: uniqueness neglect. People doubted a machine could account for what made their situation different. The more unique people saw themselves, the stronger the resistance (Longoni et al., 2019).
Which is most of us.
Known, but not reduced
We seem to want two things that pull against each other.
We want to be known.
We don’t want to be reduced to something predictable.
Prediction says, I know what you’re likely to do.
Understanding says something closer to, I know there may be reasons for what you do that I can’t see.
Prediction tries to close that gap. Understanding leaves room in it.
What we need from each other
Psychologists have spent decades studying what people need to feel satisfied in their life. One of the most researched answers comes from self-determination theory, which names three basic psychological needs: autonomy, competence and relatedness (Deci & Ryan, 2000).
Autonomy is the sense that our choices are our own. Competence is the sense that we’re capable. Relatedness is the sense that we’re connected to people who care about us.
Later research pointed to another source of well-being that stood on its own: beneficence, the sense that we’re giving something to others (Martela & Ryan, 2016).
Read that list next to the feed.
Autonomy gets harder to feel when something has already guessed my next choice. Relatedness needs someone on the other end who cares how I’m doing. And contributing requires someone who needs something from me.
Spotify has never needed me to show up.
People do.
That isn’t a small difference. A review of 148 studies found that people with stronger social relationships had about a 50% greater likelihood of survival than people with weaker ones (Holt-Lunstad et al., 2010).
Room to surprise
Maybe the difference isn’t how much something knows about us.
Maybe it’s whether we believe there’s still room for us to surprise it.
An algorithm is built to shrink that room. Every good prediction is one less surprise.
A person who loves us keeps the room open. They ask. They get it wrong. They ask again. And when we do surprise them, they’re glad.
The feed will be there tomorrow. It already knows what I’ll watch.
The people in my life don’t. Not completely. And they still want to find out.
That’s worth closing the app for.
Who in your life still gets to be surprised by you?
References
Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., & Wetzels, M. (2015). Unraveling the personalization paradox: The effect of information collection and trust-building strategies on online advertisement effectiveness. Journal of Retailing, 91(1), 34–49. https://doi.org/10.1016/j.jretai.2014.09.005
Deci, E. L., & Ryan, R. M. (2000). The “what” and “why” of goal pursuits: Human needs and the self-determination of behavior. Psychological Inquiry, 11(4), 227–268. https://doi.org/10.1207/S15327965PLI1104_01
Eyal, T., Steffel, M., & Epley, N. (2018). Perspective mistaking: Accurately understanding the mind of another requires getting perspective, not taking perspective. Journal of Personality and Social Psychology, 114(4), 547–571. https://doi.org/10.1037/pspa0000115
Holt-Lunstad, J., Smith, T. B., & Layton, J. B. (2010). Social relationships and mortality risk: A meta-analytic review. PLoS Medicine, 7(7), e1000316. https://doi.org/10.1371/journal.pmed.1000316
Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to medical artificial intelligence. Journal of Consumer Research, 46(4), 629–650. https://doi.org/10.1093/jcr/ucz013
Martela, F., & Ryan, R. M. (2016). The benefits of benevolence: Basic psychological needs, beneficence, and the enhancement of well-being. Journal of Personality, 84(6), 750–764. https://doi.org/10.1111/jopy.12215
Pollmann, M. M. H., & Finkenauer, C. (2009). Investigating the role of two types of understanding in relationship well-being: Understanding is more important than knowledge. Personality and Social Psychology Bulletin, 35(11), 1512–1527.
Reis, H. T., Lemay, E. P., & Finkenauer, C. (2017). Toward understanding understanding: The importance of feeling understood in relationships. Social and Personality Psychology Compass, 11(3), e12308. https://doi.org/10.1111/spc3.12308
Yarkoni, T., & Westfall, J. (2017). Choosing prediction over explanation in psychology: Lessons from machine learning. Perspectives on Psychological Science, 12(6), 1100–1122. https://doi.org/10.1177/1745691617693393
Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036–1040. https://doi.org/10.1073/pnas.1418680112