The artificial intelligence of film criticism
I’ve spent more than a few happy hours binge-rating entries on a website called MovieLens. It’s the public-facing aspect of a research project in “Big Data.” Each of us likely dreads that recurring issue of what to watch at night, which is about as first world a problem as you could imagine, suggesting we have no concerns about maintaining a roof over our heads and putting food on the table. Everyone trying to sell us products and services, including entertainment options, would like to predict what we will enjoy consuming. MovieLens is a scientific experiment, and, like the best of them, it resembles a game. It will bring out the geek inside any cinephile, if that identity isn’t already on display.
MovieLens has a catalog that exceeds what any individual, even the most devoted fan, could watch. There appear to be approximately 72,000 items within its regularly-updated scope. Thus, on any given page showing an array of miniature posters, there will be only a few choices that I can critique with any knowledge. That’s the prize: a picture I can click on to reveal the opportunity to assign stars and refresh my memory with the recollection of an enjoyable evening.
The number of these winners likely will decrease over time. I would have to see movies at a pace exceeding how rapidly I can rank them to maintain my odds. The graphic is integral to the appeal of the game. A list of text would be too plain.
The reason I (admittedly a bit OCD) would expend energy that could be spent on work is the same reason we enjoy checking email. “Variable interval reinforcement” is also why we gamble. Any of us can train ourselves or others, even with something so trivial as an audience member paying attention to a lecturer. The conditioning of a regular reward, however, is not as effective as the alternative of a random pay-off.
The late professor B.F. Skinner was the giant of this school of psychology, proposing society reinforce and inhibit behaviors through systematic stimulation. His theories, rejected for promoting authoritarianism, continue to influence the marketplace. MovieLens depends on incentives to engage. It is, beyond being a study of movies, a study of how to render surveys attractive.
That’s What I Like
My profile contains details I wouldn’t have anticipated. I’m more discerning than I would have guessed. My ratings range from 1.0 to 5.0, with only 52 at 1.0 or 1.5 and 208 at 5.0. The year of release for which I have rated the most films is 1995, when I was 27 years old. I rate documentaries the highest as a genre, with an average of 4.43, but this may be “preference falsification,” meaning our tendency to say what not quite what we believe but what it is beneficial to state we believe (for example, to impress others with how classy we are). I rate fantasy the lowest, at 3.48, but this also might be higher standards due to greater interest.
Some titles are “rare,” in the sense that few participants have given them grades. Thousand Pieces of Gold was not a huge hit. It is the Hollywood version of the true story of Polly Bemis, a Chinese woman sold by her family in the late nineteenth century, who becomes a pioneer in what would have been the territory of Montana prior to statehood. I saw it at the Opera Plaza cinema in San Francisco in 1991, on my own, and, besides me, there was only one couple at the matinee (Caucasian male, Asian female). Chris Cooper plays the “white savior” to Rosalind Chao’s heroine. I awarded 3.5 to this feature.
But millions watched the Six Million Dollar Man, the TV movie premiere of the 1970s television show about the “bionic man”. Lee Majors, Lindsay Wagner, and Farrah Fawcett were about as exciting as could be to a kid in the suburbs, staring at an old-school small screen powered by cathode ray tubes which, we were warned, were radioactive enough we had to sit a half dozen feet back, and, when burned out, could be replaced at the hardware store. I bestowed 4.0 on the principle that that score was appropriate for that point in my personal history.
You also can assess your taste against the general population. My unusual dislikes include Night at the Roxbury (average of 2.69; predicted for me, 2.47); Spaceballs (average of 3.42; predicted for me, 2.93); and Uncle Buck (average of 3.33; predicted for me, 3.18). I regard all of those as stupid, an insulting waste of a couple of hours of my life—the algorithm at least prophesied a relative aversion. My unusual likes include Dumb and Dumber (average of 2.92; predicted for me, 2.95); The Interview (average of 3.06; predicted for me, 3.52); and Hot Tub Time Machine (average of 3.1; predicted for me, 3.43). I treated these as idiotic from the get-go, and I was delighted to be proven right. Again, the formula forecast I would express more enthusiasm than typical.
It’s easy to make a mistake. Many titles are similar, in some instances deliberately confusing the potential viewer. In 1998, Pixar produced A Bug’s Life; Dreamworks, Antz. Accordingly, the “mock buster” direct-to-video Ant’s Life was an attempt to exploit the popularity of its animated superiors. I’m tempted to pan what I haven’t sat through. That would be wrong. I’m exposed to facts I wish I remained ignorant of: that some wonderful movies had terrible sequels. Scrolling systematically through lists of movie titles, it turns out, discloses secrets about the virtual universe.
Beneath the surface, MovieLens is serious scholarship. Since 1998, dozen of academic papers have analyzed its data . University of Minnesota professors developed the feedback loop to investigate phenomena such as pattern identification. They explain, “movies are a common interest, making algorithmic output easy to discuss.” The basic concept is if you loved ABC and hated XYZ, and I responded in the same manner, then if you also like new release D, I probably will, too. We just need to upload everything about ABC and XYZ.
The Algorithm Knows All
MovieLens is quantitative, not qualitative. It is a supplement, not a substitute, for a good essay about a good movie. To write about movies, I read about them. For most movies, I consult at least a review in advance, usually three; I do the same afterward, sometimes pulling up the same critique. There are critics whom I like, regardless of whether I agree with them. The late Pauline Kael, for example, is a worthwhile provocation, even if upon reflection her opinions seem to be poorly reasoned (or, as with Citizen Kane, based on false assertions). Others, I have concluded, possess an aesthetic judgment I do not wish to share; about the inoffensive We’re the Millers, the local newspaper critic wrote, “Comedies don’t need to be believable, but they do need to be logical,” which is the exact opposite of my impression of humor.
I looked at MovieLens. It was informative. Citizen Kane is 4.08 based on 22,419 ratings, predicted to be 4.91 for me, and given a 5 by me. We’re the Millers is 3.5 based on 4133 ratings, predicted to be 3.66 for me, and given a 4 by me. So consider this a meta-recommendation, a recommendation of a recommendation. Signing up for the MovieLens system is a must for anyone who cares about movies.