Stuff Smart People Like
Amazon.com never did reply to my letter protesting the change in their recommendation process (other than the instaformlett I got within minutes, claiming it “understand[s] my concern”), thanks to which you’re no longer able to just view an item’s page, click “I own it” if you bought it elsewhere, and rate it by selecting 1-5 stars to improve your recommendations. Now, you have to “share your thoughts with other customers,” which requires creating a review. You can continue to rate Amazon purchases through “Improve my Recommendations,” but you’re SOL on adding non-Amazon purchases to the recommendation engine’s database unless you create a full review.
I noted the tone-deafness of recommendation engines in that post, a thought somewhat but rather sloppily seconded by Walter Kirn in the Times magazine this morning, though “ur doin it wrong” springs to mind when I read how he’s using the system(s). He admits he’s “polluting” his results by using his girlfriend’s account, which doesn’t account for the irking fact that browsing for “True Grit” (the novel) and Mark Twain’s autobiography leads Amazon to bomb him with Civil War books, which he finds offensively classes him “as some sort of drawling son of Dixie who re-enacts the battle of Shiloh on weekends.” Netflix in turn insulted him with Woody Allen’s “Radio Days” on his list, which he hated. (Walter, you have to click “one star,” “not interested,” or “do not use for recommendations” on things like that to get them off your list. In Amazon’s recommendation system, you can still click “not interested” with zeal on the innumerable bad choices it throws at you, until at last you go back to your original purchase and click “Do Not Use For Recommendations” because loving the last Star Trek movie absolutely does not mean you want to see Aliens Vs. Predators XLVIII.)
Maybe Amazon has hand-corrected the Matrix since Kirn’s review (available online no later than yesterday), because the page for Portis’ novel now recommends more Portis as well as other “outlier” novels in the same vein like “Little Big Man,” “Treasure of the Sierra Madre,” “The Ox-Bow Incident,” “The Big Sky, and “Cool Hand Luke” – all also good novels turned into great films. No Civil War history in sight, on either the fine “Battle Cry of Freedom” end or the Southern-revisionist Shelby Foote axis.
Why is software incapable of figuring out that Kirn loves Flannery O’Connor? Why can’t it give him her books if it can give him Barry Hannah, based on Charles Portis? To a large extent, you get what you put into a recommendation engine, but there’s no doubting that software by itself is still terrible at assisting good taste, but there’s a reason for that.
The design of recommendation engines is based on one primary object: driving sales. Not to assist those of us who treasure outliers and novelty and variety and obscurity. Most people who read Lee Child do not see that he is head and shoulders above the rest of the thriller-writing crowd, an exception to be made on a reading list like mine that may not otherwise include anything from that genre. Most people are as content with the next brick from the James Patterson writing factory as they would be with the next Reacher book. So a system that recommends Patterson from a Reacher page or vice versa is exactly what those consumers want and exactly what will sell more books.
Recommendation engines are engineered to provide you with what most people want: more of the same thing. Romance novels and Saw movies and hapless-dad shitcoms are all the same, and yet that’s what most people want – the same thing over and over. In my dream world, and presumably in Kirn’s, you tell a system that you love “The Wire” and instead of recommending “It’s Always Sunny in Philadelphia,” it recommends, oh, I don’t know, let’s say Trollope’s “Palliser” novels, because it’s another huge sprawling story of a city, covering large amounts of time and space, encompassing sex and class and crime and money and politics, high and low. In fact, I think there’s good money to be had in a website – let’s call it Fleemob.com – for people who want that diversity of taste in recommendations. The way that Mint.com hoovers up all your online financial data, this site could hoover up your Netflix and Amazon tastes (and who knows, favorite Vimeo and YouTube videos and blog writers and postings for that matter) and let you “curate” a collection which could then be pattern-matched with that of other members, and you could say, show me what someone like me wants to read or watch next, based on all these tastes and not just a desire for more of one of the same of one of them. Reddit is always replete with people saying “Hey I just watched the Wire what do I do next?” The market is not and will never be as vast as that for those wanting more Saw movies, but there is indeed a market.
When you “tune in” to Google or Amazon, it’s like turning on the radio in the days of “Austerity Britain” – you can listen to BBC. Or BBC. (You can buy a car in any color as long as it’s black.) “The algorithm” is THE algorithm, designed for one and all. Imagine if you could tune into Google or Amazon and select a “station” – imagine if you could “tune” out the tastes of people who don’t like “Black Swan” even though they thought it had “suburb acting” (sic) as a Fandango reviewer posted recently. Can this really be so difficult?