No accounting for…
If you could know for a certainty that you would like a movie before you actually saw it, would that be a good thing? That question is implied by an interesting article in The New York Times Magazine a couple of Sundays ago.
Titled “If You Liked This, You’re Sure to Love That,” the piece by Clive Thompson tells the story of how online DVD renter Netflix has tried to optimize its web site recommendation engine, called Cinematch. Stymied in its attempt to improve the success rate of its automated recommendations to customers beyond a small percentage, the company initiated a contest that offered a million-dollar prize to any programmer who could produce a 10-percent jump in successful film suggestions. The challenge has proved to be daunting, and the article provides compelling profiles of some of the competitors and their efforts.
Netflix’s motivation in this is clear. Its customers pay a monthly subscription fee to have DVDs sent to them. If the customers run out of movies that they are interested in seeing, they will in all likelihood cancel their subscription. So a company like Netflix has a major vested interest in helping people find movies they will like. Movie studios have a similar interest. If they are going to invest big bucks in the production of a movie, they want some assurance that large numbers of people will actually pay money to see it. But brand loyalty is less of an issue to the studios than it is to a subscription DVD rental service. Few, if any of us, queue up to see the latest release from say, Warner Bros, because, hey, I sure enjoyed the last four or five Warner Bros. movies I saw so they must know something about making good movies. Indeed, most people probably have no advance idea what studio produced the movie they are going to see on any given evening. If brand loyalty has been a factor, it more likely has to do with the stars of the movie or, perhaps, the director. Or perhaps the book or play or 1970s TV show it is based on. But most likely it has to do with the marketing that one has been exposed to and maybe the reviews one has read or, perhaps most importantly, what one has heard about the flick from friends.
So if studios can’t really build brand loyalty based on their own identity, they do what they can to put bums in seats. They look at what has drawn crowds before and come up with more of the same. If hordes flock to see Titanic, then they come up with more movies about sea disasters or disasters in general. If punters mob cinemas showing the Lord of the Rings movies, then they put out more swashbuckling fantasy flicks. In other words, they pick up the most obvious trappings or elements of the successful movies and try to come up with other movies that offer those same trappings or elements. (The owner of the rights of the successful movie has even more sure-fire options: making a sequel, a prequel and, eventually, a remake) And, broadly speaking, this is what recommendation software does, i.e. find existing movies that have similar elements or features to movies that the customer has indicated that he or she enjoyed.
Even non-programmers can appreciate intuitively how difficult such a program is to write. People’s tastes are obviously much more complicated than: if I like this cop movie, then I must like all cop movies. Any bit of software that was actually capable of accurately predicting one’s taste to the degree of consistently forecasting a human being’s reactions to movies would border not only on artificial intelligence but on the supernatural quality of an oracle. (The impenetrable nexus of this problem seems to be the movie Napoleon Dynamite. According to Thompson’s article, of all movies in Netflix’s database, this one is the most difficult to predict whether a given customer will love it or hate it.) After all, if you are like me, you yourself can’t even predict whether you will like a movie before you have seen it. You might have some good clues based on the subject matter and the creative talent involved, but how often have you gone into a cinema expecting to love a movie you had been eagerly looking forward to only to be disappointed? Or to have been dragged to a movie that you were sure you wouldn’t like and were pleasantly surprised?
But couldn’t the fact that we ourselves cannot predict what movies we will like not argue that the computer could actually do a better job? With the power of modern computing harnessed for the task, might not a clever program pick up on things that we ourselves are unaware of that provide clues to our taste? Possibly. But say for the sake of argument that a program could be written that reliably predicted what movies we would enjoy, would this be a good thing?
Of course it would, you are probably saying. After all, you are a busy consumer. You have only so many hours in your week, no, in your life, to watch movies. Why waste precious time watching ones you don’t like? How much more efficient to watch the ones you are sure to like. But look at it a different way. If a computer program could reliably predict the best person for you to marry, would you want it to arrange a marriage for you?
The reality is that being exposed to things that we think we won’t like, or that we do not like at first, changes us. Exposure to different people, to different ideas, to different art broadens us. Our tastes evolve, change, mature. And if we are exposed to too much of a thing we like, we sometimes get tired of it or outgrow it. And let’s not forget the thing that, maddeningly, studio executives always seem to overlook. Often it is the quality of a movie we like, not so much the plot or characters. If we liked the Lord of the Rings movies, it was because they were so good. We would gladly watch any other movie that good, regardless of whether it was a cowboy flick or a love story. But this bit of information is not particularly helpful to a computer programmer because, to a large extent, quality is a subjective judgment.
Years ago, I used to tell people that I would rather spend an evening watching a bad movie than watching a good television show. This was back when I was snobby about TV and before I found that there are actually some brilliant things made for that medium. But the point was that time spent watching a movie was never wasted even if I hated it. Because the act of watching a movie always had one very positive payoff. It always satisfied my curiosity about what the movie was like. Of course, this was back when I was curious about every movie I heard about. I am old enough now to realize that, while every movie potentially has a surprise in store for me, many of them don’t. And since I can’t possibly see most, let alone all, of the movies that get released, I have to pick and choose. And that forces me to, like a computer, make calculations about which one is likely to provide with the biggest payoff.
But I am happy to say that, still, some of the best films I see are the ones that I hadn’t planned to see. The ones that I wandered into by accident, usually at a film festival and maneuvered by scheduling complications. Perhaps to fill a gap in a time and place between two movies I thought I wanted to see. Or in place of a film that I couldn’t get into. I have never walked into a film screening completely by accident, but I wonder if that would not be a good way to discover something new and wonderful. Such an experience was actually the subject of a 1997 movie called Love and Death on Long Island, in which John Hurt wandered into the wrong auditorium in a multiplex, sat through a teen comedy that would never have entered his viewing list and wound up falling head over heels in love with teen idol Jason Priestly up on the screen.
Could a computer replicate the accidental viewing experience, by throwing a bit of randomness into its recommendation algorithms? (Not that all of us would particularly want to fall head over heels in love with Jason Priestly.) Possibly. But if a computer is really going to help broaden us in our tastes and enjoyment, it will probably do so because it can bring us into contact with so many different people.
-S.L., 4 December 2008
If you would like to respond to this commentary or to anything else on this web site, please send a message to feedback@scottsmovies.com. Messages sent to this address will be considered for publishing on the Feedback Page without attribution. (That means your name, email address or anything else that might identify you won’t be included.) Messages published will be at my discretion and subject to editing. But I promise not to leave something out just because it’s unflattering.
If you would like to send me a message but not have it considered for publishing, you can send it to scott@scottsmovies.com.