Nina and I were in the Rock ‘n’ Roll section of a used record store.

(Yes, such places still exist.)

“Papa,” she said, “What about the Doors?”

“Hm,” I said. “Heavy emphasis on poetic lyrics. Surrealism. But with a spooky psychedelic late 60s sound.”

“Oh,” she said. “It’s just, someone whose opinion I trust — a lot — said I should check them out.”

I picked out a Doors album for her; she took it to the old turntable in the corner of the store and gave a listen.

As it turns out, the Doors weren’t her cup of tea. That day, she bought albums by Billie Holiday, the Supremes, the Police, and Prince, but left Jim Morrison to languish in a fog of patchouli on the shelf.

My daughter’s trusted informant — whose name I didn’t dare ask; teenagers hate that kind of question — had missed the target by recommending the Doors. But the recommendation interested me.

Nina has grown up in a house of music. Every night when we sit down to dinner, there’s something on the stereo. Blues. Jazz. Malinese singing. American Folkways. Gilbert and Sullivan. The Arcade Fire. Whaling shanties. Yo-Yo Ma. Alison Krauss.

This list may sound somewhat random, but there are commonalities. We tend to like virtuoso musicianship; music with deep and authentic roots; and complex rhythms and melodies.

The music industry has a word for our kind of taste: “eclectic,” by which they mean “defying categorization.”

This makes us something of a marketing nightmare.

Say you’re a music company, and you want to sell us a record. We’ve bought from you before, which gives you an important piece of information about what we like. Last time, it was Duke Ellington, so that means we like big band jazz, right? So how about this newly remastered Cab Calloway?

Yeah…we’re just not crazy about Cab Calloway. The Duke was a brilliant composer. Cab was more about the show than the music.

Okay. Let’s try something else. You like whaling shanties, right? Work songs with a nautical flavor. Let’s see… Here’s something. How about songs of the Russian navy?

Actually, we’re not really into military music.

Things were simpler twenty years ago, before the Internet had devoured the music business. If we wanted to hear something new, we could go to an independent store staffed by knowledgeable people whose taste — and whose estimation of our taste — might be a guide to undiscovered music.

Some of those brave and wonderful stores still exist, but they’re increasingly rare, and most are hanging on by their fingernails.

More and more, the recommendations we get for music, books, and movies, are generated by computers.

Netflix was a pioneer in this field. Early on, they realized a crucial fact about their business model, which was based on monthly subscription: if people watched a lot of movies, they stayed with Netflix. If, on the other hand, they were overwhelmed by the tens of thousands of movie titles to choose from, and therefore didn’t watch many movies, there was a very good chance they’d cancel their Netflix account.

The trick was to suggest movies a customer was sure to like.

The executives at Netflix realized they were sitting on a treasure trove of data. They already knew a lot about their customers. Each movie rental was a valuable data point — a dot that, taken with all the other dots, painted a detailed portrait of a customer’s taste. And by asking people to rate movies they didn’t like as well as movies they liked, the profile got richer and deeper.

Netflix threw a lot of money and talent at the problem, and “Cinematch” was born, a computer program that could extrapolate from a customer’s past rentals and generate suggestions from Netflix’s vast video library. Cinematch worked by comparing customers with each other using a process called “collaborative filtering.” If you gave Star Wars five stars, the highest rating, the computer looked up other customers who’d given Star Wars five stars, and recommended their top picks to you. The odds were good that if you loved one movie in common, you’d love others in common, too.

Cinematch’s recommendations were statistically solid, but not perfect. After all, people can like Star Wars for very different reasons. And there were some movies, like Napoleon Dynamite, for instance, that people either loved or hated in ways that were impossible to predict. So in 2008, Netflix offered a million dollar prize to anyone who could improve Cinematch’s performance by ten percent.

A million dollars might sound like a lot to pay for a ten percent improvement, but Netflix stood to gain enormously if they could figure out how to lose fewer customers.

The outsourcing strategy worked. Teams around the globe competed feverishly for the prize; one team found a solution within a year.

But anyone who has used Netflix knows that the new and improved recommendations still leave something to be desired. The question of why we like what we like is infinitely complex. The fact that I love the Seven Samurai doesn’t mean I have a thing for samurai movies. It means I have a thing for Kurosawa movies.

On the other hand, human beings aren’t perfect recommenders, either. Witness Nina’s friend and the Doors. But I think a sensitive, intelligent person will always have an edge on a machine. Algorithms — even massively complex and sophisticated ones like Cinematch — rely exclusively on a person’s past choices, which aren’t necessarily predictive of future ones. And collaborative filtering, which is being used more and more frequently by companies like Amazon and Google, has limits, too. Call me delusional, but I’d like to think there’s something original about me, something that makes my taste my own, and not reproducible in a computer lab.

Isn’t it possible that, despite a lifetime of choices, we don’t yet know the full outline of what we like to listen to, read, or watch? People change. They move in new directions. That’s bad news for companies that would have us grow more and more alike, the easier to make a sale.

Me, I vote for eclecticism. Inconvenient for the folks in marketing, perhaps, but it’s hard to imagine anything more human than that.

This column was published in the Perry Co Times on 05 April 2012

For more information, please contact Mr. Olshan at writing@matthewolshan.com