Netflix/Amazon/iTunes Suggests…

STEVE MURPHY: Websites have been trying to recommend things to me for years based on my prior activity, but they’ve never really been any good at it. has always been the worst offender, because it doesn’t seem to have any intelligence behind it at all.  I buy a camera… and it starts suggesting cameras to me.  Constantly.  You know what, Amazon?  I just bought a camera from you.  I do not yet need another camera.  Yes, it lets you rate what you’ve purchased and tell it when what you’re buying is a gift, but it never puts two and two together.  It never seems to suggest anything but what I’ve already been looking for, and often have already bought.

iTunes tried hard for a while, but couldn’t seem to get it right.  They’d suggest music based on your iTunes purchases, but couldn’t take into account any of the other music you own.  As my collection is about 10% iTunes purchases and 90% CDs I’ve purchased and music I’ve been (ahem) given, that recommendation engine never really connected for me.  But then, last year, came Genius.  Genius scans your entire library, all of it, and manages to put songs together into logical, clever little playlists that I typically enjoy very much.  They extended this into the iTunes Store, to suggest music I’ll like based on my entire collection.  I think it’s a great service, and while I don’t think I need help discovering new music, I’m glad it works as well as it does.

The only downside of Genius is that it’s a secret sauce: Apple won’t tell us how it works.  That’s very Apple of them, really, and not a surprise.  But it also means I have no way of improving my Genius results.  I can’t, say, find out what albums it’s not recognizing because I have the name typed wrong.  I can’t know how it’s relating songs together, or what effect a song’s “Rating” will have on my results.  I don’t like that.  I’m an efficiency guy.  I want to optimize, and Genius just makes that impossible.

Netflix, however, has gotten it right.  That’s because Netflix has a scarce product, and so they must rely on “the long tail” to for their business to do well.  Netflix has physical discs to handle, in limited numbers.  They have only a certain number of copies of the new, in-demand Harry Potter movie, so they want to make sure that those who can’t get the new releases are still presented with satisfying, not-new-but-still-great movies.  iTunes, on the other hand, doesn’t have that scarcity problem.  They don’t care if you buy a new song or an old song, because they don’t have to keep an inventory and they’re not running out of anything.

This aspect of Netflix’s business model demands big retheir recommendation engine has to be amazing, so Netflix spends tons of money on it.  They even famously issued the Netflix Challenge, which offered $1m to the first team of civilians to improve their recommendations by 10%.  This proved difficult, and the contest took more than a year before someone successfully completed the challenge.  That just shows how close to perfect the system already is.

When I log into Netflix, every movie has a rating in red, from one to five stars.  That’s how Netflix thinks I would rate that movie after watching it.  It even tells me which movies I rated that helped them choose this for me.  The more movies I rate, the more finely tuned my recommendations become.  These estimates are very often correct, and I am constantly impressed.  I remember once I saw a preview that had me sold, and even when Netflix told me I’d hate it, I rented it anyway.  Netflix was right.  It was terrible.  I stopped it in the middle.  I sent it back and rated the movie exactly as Netflix predicted I would.  I did that a few times before I learned to start trusting the machine.

So now I’ve stopped renting movies Netflix tells me I won’t like.  There’s no point in it.  I’ve surrendered.  Sometimes I rent a film Netflix highly recommends, even if I haven’t heard of it.  Sure, they occasionally guess wrong, but typically their recommendations are very much in sync with my tastes.  It’s the first recommendation engine I’ve truly loved, so simple yet so effective, like a personal movie concierge.  I can only hope that Amazon eventually catches up.

AKIE BERMISS: Read your science fiction, people.  If there’s nothing else we’ve learned from various sci-fi doomsday scenarios, we should certainly know that you should never trust the machines.  Machines are out to get you.  Maybe not today. Maybe not tomorrow.  But someday — those machines are going to take you down, man!  And so, while I love gadgets and devices and artificial intelligences, I have a healthy cynicism for computer-to-humanity relations.  And so I don’t trust any computer or program or algorithm to tell me what I like or may like or won’t like.  I have good friends that’ve known me for years and still couldn’t really guess what I like or don’t like.

Only I know that.

Why?  Because, on a pragmatic level that ignores genre, circumstance, and culture — everything is, in some way, likable (and obviously therefore also dislikable).  All sound, all silence, all light, and shadow… all these things are all that we can experience through the senses.  And where we fall in terms of liking things or not is dependent upon all the variables of our nurturing.  Where did you grow up?  What books/movies/music did your parents like?  What did your friends like? Where you saved from awkward adolescence by the drama club, or the math team, or the literary magazine in High School?  All these things will inform your opinion of things.  Whether you think Citizen Kane is still the greatest movie ever made or if that prize goes to The Usual Suspects.  Whether, when you think of high drama, you choose Austen or Byron or Hemingway.  And then, as you get older, and grow and accrue artifacts — a “sensibility” begins to distinguish itself.  Your record collection, your bookshelf, your jewelry collection, your tie collection… All of this, in an era now past, could have been read to tell us what kind of person you were.  And what you liked.  And what decisions you would make.

But these days everything is everything — or its getting closer and closer to something like that.  Now there is more to see than you record collection and private bookshelf.  What CDs do you keep in the car?  What do you have on your ipod?  What’s in your iTunes library? What’s on your back-up hard-drives? What shows do you follow on Hulu?  What your most searched item on Google?  Question upon question upon questions.  That is the only way to approach a holistic understanding of one’s sensibilities.  It can’t be answered with by using a five-star rubric to judge a couple dozen or a couple hundred movies? Just because I love David Lynch’s 1984 Dune doesn’t mean I give two bits about the rest of David Lynch’s movies (and I don’t.)  So, Netflix, why are you suggesting them to me?  Don’t you know that I love Dune because it was so wrong it was right?  That Patrick Stewart (whom I’ve loved not only in his more serious roles, but also as Captain Picard on Star Trek) plays Gurney Halleck and is bright shining star in the movie?  Also, Sting plays the sexy, evil counterpart to the film’s protagonist!  All these things are little reasons I love the film.  How could your algorithm pick up on that? It can’t — not without significant help from me.

I suppose I could go through and give all the rest of Lynch’s movies two stars and hope that guides the Netflix computer-mind to the decision that I’m not going for that stuff.  That Dune is a fluke.  Pure and simple.

Also, sensibilities change!  Have you ever run into a friend from childhood and tried to hangout and found that you guys no longer have anything in common?  That your friend still wants to eat skittles and watch cartoons all afternoon even though he’s a grown-ass man.  You like to drink scotch and watch pro-soccer.  Of course, should you decide to cave and have a little taste… maybe you get a sack of Skittles from Walmart and find the entire X-Men The Animated Series has been uploaded on Youtube… would you expect your new friends to understand?  Or be perplexed?  To feel confused and betrayed?  Would you want them to hold it against you?  And find that on your next birthday you got the complete Teenage Mutant Ninja Turtles from your wife?  I think not, my friend.  I think not.

And that’s the problem with people and computers.  People change.  Sometimes.  And sometimes only somewhat.  Somethings stay with us as a like or dislike all our lives?  Other things can change.  Somethings that we don’t like most of the time, we indulge in once in a while.  So you might fight my box set of Laura Ingalls Wilder books if you look closely at my bookshelf — but shouldn’t take that as an indication of what I’m usually reading.  Yeah, computers can’t be trusted.  And they can’t be trusted to willfully surprise you.  A book won’t watch a movie and decided: You may not like this, but you should see it.  A computer won’t force you to watch Donnie Darko because they think its great.  A computer won’t suggest that you try something new.  Computers are dependable.  And math is powerful.

But people are intelligent.  And with intelligence comes a bit of flaky unpredictability.  A potential for insanity.  For irrational thoughts and preferences.  That is the existence we exist in.  So computers can keep their suggestions… I search as I see fit.  I watch PBS and I listen to friends and I watch trailers.  And there’s a lot of chance involved in the things I do.  Certainly.

But I make up my own mind.

This entry was posted in Arts & Culture and tagged , , , , , , . Bookmark the permalink.