Easter 1929, New York. During the Fifth Avenue parade, a group of women light cigarettes in public. They are called "torches of freedom." Edward Bernays—the nephew of Freud and an adviser to the tobacco industry—stage-manages the event, linking a product to a claim for emancipation. The client sells cigarettes; the message sells freedom. The story.

Bernays tried to understand what people wanted, then find the message that could persuade them. A like seems much less important than a cigarette lit during a parade. But it leaves a clue: something appeals to us, draws us in, feels like us. Put enough clues together, and how much can you learn about someone?

In 2015, a study by researchers at Cambridge and Stanford tried to answer that. Using Facebook likes from 86,220 volunteers, researchers estimated five personality traits: openness, conscientiousness, extraversion, agreeableness and emotional stability. They compared those estimates with questionnaire results and the judgments of colleagues, friends and family. Could a computer describe us better than people who knew us?

THE COMPARISON · 2015 STUDY

10 likes · a colleague

70 likes · a friend or housemate

150 likes · a family member

300 likes · a spouse

The number of likes the model needed, on average, to outperform these people's judgments against questionnaire scores. Not a universal threshold for any one of us.

With around 300 likes, the model outperformed a spouse's judgment on average. It didn't know where you'd left your keys: it was better at estimating those traits than the test. At home, I'd still keep the consultation open when it comes to the keys.

A previous study, from 2013, found that one clue associated with higher intelligence scores was liking curly fries. Not because curly fries make you smarter: it was an association in the data, with no obvious connection. One like isn't enough to grade a person. And ordering a serving doesn't count as a master's degree.

A statistical clue, not brain food.

This is where what I call datacracy begins: the power of those who collect and interpret our traces. The word matters to me because it shifts attention from the data point itself to whoever can use it. I don't have to write, "this is my vulnerability." Sometimes what I let show is enough.

This is the part that worries me: trying to describe me is one thing; using that description to choose what to show me is another. AI doesn't make these predictions infallible. What concerns me is that we may treat them as truth, especially when it suits us. Minority Report is still a film; the question of who decides what we see is already ours.

When you like something, you're speaking.

THREE THINGS TO DO WITHOUT DISAPPEARING FROM THE INTERNET

1. Before you like something, ask whether you want to leave that particular trace.
2. When you see it, open "Why am I seeing this?" and check your ad preferences.
3. Read someone you disagree with, too. Not to throw the software off the scent, but to keep your world from shrinking.

I write about this in my book, Avviso ai naviganti. The title refers to bulletins for people at sea: they don't tell you to stay ashore, but to pay attention. There's a line I care about: "We went from fishermen to fish in a fish farm." I'd rather keep choosing where to swim.

NOTES FOR THE VOYAGE

For more: my article on datacracy (2019).

Studies: Cambridge and Stanford, 2015 · PNAS, 2013

Watch: The Truman Show and The Social Dilemma.

Read: Bernays, Propaganda; Zuboff, The Age of Surveillance Capitalism.

Have a good Sunday, from A margine: a small curiosity to carry with you.

Giovanni

A dedication to my dad, who turns 70. All he needed to know me was one glance. Happy birthday, Dad.

Three hundred likes

Ti è piaciuto?