Machine Storytelling

Our power to tell stories about each other is one of the most important tools we have for understanding, connection, and joy. People tell stories to show who they are, to understand others, to establish relationships. Enjoying a good story is part of what makes us human.

When we first meet someone new we might ask a mutual acquaintance, “what’s her story”? It’s a question that can fit almost any situation. In a corporate meeting, we might mean, what team is she on? Or at a wedding, how does she know the couple? The answers to this question aren’t like other stories.  They matter in a special way, giving us a quick summary of who a person is, and of where they fit into the scheme of things. In fact, another way of asking basically the same question would be, “who is she?”

But it’s not only humans who can tell stories about who people are. Machines can do it too.  A machine that acts on information -- taking some in, processing it, and spitting out some other information — is called an algorithm, and many algorithms are made for telling stories about who people are.

Take the SAT, for instance. Over a long and controversial history, it has become a standardized way to turn multiple choice answers into a number that sums up a college applicant.

An SAT or IQ score might claim that the person it describes is gifted, or dull. A credit score might say that the bearer can be trusted to pay a loan back, giving them access to the compounding advantage of a favorable interest rate. On a dating app, a high number might mean that this guy is a good bet for the person who’s browsing the listing, a match worth grabbing a beer with. These stories, like the handmade human kind, are all open to interpretation by the listener.

It’s not just bureaucracies who love these mechanical stories: As individuals, we can’t seem to get enough of them either.

At the frivolous end of the spectrum, Cosmo and Buzzfeed have turned personality quizzes into an editorial franchise  (“What Harry Potter Character Are You?”). With higher stakes, corporate America loves the Myers-Briggs personality type indicator, which groups people into sixteen personality “types,” claiming to unlock deep insights into who we are that can help us navigate work, life and love.

And new machines for describing people have never been easier to build. That’s how I first came to think about this. I was working on civil rights advocacy in courtrooms, where “risk assessment” algorithms attempt to sort the truly dangerous people from the rest of us.

A mechanical story – like any story — encourage the audience to take a certain perspective. The courtroom algorithms, for instance, invite us to think that it is people who are innately dangerous, rather than danger being a product of circumstances. An IQ score invites us to see behavior as the expression of innate, durable intelligence.

The Myers-Briggs  — which if you’re reading this, you may well have taken yourself — implies a perspective in which our personalities are static. Merve Emre wrote a great  book on the test, The Personality Brokers. Emre argues that psychologists have found  a raft of serious methodological problems with the Myers-Briggs, including a lack of test-retest validity: A large fraction of test-takers get classified differently when they re-take the test. (The test maker now claims that 35% of all takers get a different classification when they re-take it.) At one point, Emre recounts a training session for people learning how to administer the test. But the trainer in Emre’s session, a woman named Patricia, argues that:

Personality is an innate characteristic, something fixed since birth and immutable, like eye color or right-handedness. "You have to buy into the idea that type never changes," [she] says, speaking slowly and emphasizing each word so that we may remember and repeat this mantra — "Type Never Changes" — to our future clients. "We will brand this into your brain," she vows. "The theory behind the instrument supports the fact that you are born with a four letter preference. If you hear someone say, 'My type changed,' they are not correct."

I can’t get that scene out of my mind. The people who sell personality tests have a natural reason to persuade us that personalities are testable, and stable. Much of the data-driven scoring that’s being introduced now somehow rides on, and amplifies, that same impulse to think we can get people’s number, that people even have a number, in whatever way we might want to rank them.

What is it about these mechanical stories that entices us so much? Can a story about who someone is be useful and good, without it being true? And what would happen to our stories, if we resisted this natural urge to pin one another down into stable, simplified identities?

More next week.