What a Curly Fries Like Could Give Away

Why would a fries page help predict a test score?

In a 2013 Facebook study, a Like for Curly Fries was linked to higher scores on a test of reasoning.

That is a strange job for a food page. A Like tells you which page someone clicked. Why should it hint at how well they solve a puzzle?

The study found no clear reason for that link. The result was not evidence that eating fries makes anyone smarter.

But it revealed a curious gap. A model could find a useful clue where a person saw no obvious meaning. It did not need to explain the link to make use of it.

How a click became a clue

Michal Kosinski and his team had two kinds of records. One showed people's Facebook Likes. The other held facts and test scores from those people.

The reasoning part came from a small group of 1,350 volunteers. They took a test that asked them to spot patterns in shapes. It was a measure of a specific skill, not a full account of what someone could do.

The team used some people's records to build each model. Then it asked the model to predict scores for people left out of that step. The real scores gave the team a way to check its guesses.

This mattered. A model that only knew the records it had seen would not tell them much. The test asked whether a pattern could help with people it had not been fitted to.

A list of Likes could compete with a friend

In a later study, Wu Youyou and her team put Likes up against friends' views of a person.

People rated their own traits, such as how open they were to new things. The team compared those ratings with a model's guesses and a friend's answers. Around 100 Likes were enough for the model to beat an average friend in that sample.

Here, "beat" meant closer agreement with scores for five broad traits. Friends answered ten items; people rated themselves on 100. It was a narrow test, not a contest over who knew someone's life best.

The page you chose is only part of the story

Both studies used old Facebook records from people who chose to join a research app. They cannot tell you what your own Likes reveal, or what Instagram infers today.

The useful lesson is about the gap between an action and the clues it can leave. You may know exactly why you clicked a page. A model may group that click with others to answer a question you never had in mind.

You chose a page about fries. You did not choose what else that choice might help someone guess.