Thanks, in part, to Fortune Magazine. You see, Jeff O'Brien went and has beers and meat with some guys in Atlanta and now he's hot on recommendation engines. For those who might not know, recommendation engines are systems that watch what you do, what you buy, who you talk to, where you live, etc. to figure out, statistically, what you would like to buy.
Then it shoves opportunities to purchase them down your throat.
Thank you recommendation engine.
So, my wife loves little shiny bits on her clothing. "Tsim" stuff, in Cantonese. I am often subjected to her holding sweaters up with all sorts of glitter and blinding stuff and she says "Whaa, ho tsim! Do you like this?"
"That's pretty tsim," I reply.
Well, I am an attentive husband and, while I in no way share my wife's love of that clothing, I will buy it for her because I know what she likes.
Recommendation engines, which are watching me, know that I like to buy Ska CDs, books (of various titles), music software and major appliances. And ... shiny women's sweaters. It plugs in this information and comes out with recommendations. But wait! I didn't WANT a shiny sweater, I just bought one.
It is scenarios like this that make Mike at Techdirt and me a little woozy. He reminds us of the situation where everyone who watched certain shows were blessed "homosexual" by their Tivos a while back.
Recommendation engines will look at things I bought and swear I want something else - like perhaps more shiny sweaters or maybe a Hello Kitty Mitsubishi.
Recommendation engines are right up there with perpetual motion machines. The really perfect one is unobtainable. However, we can greatly reduce friction and probably build some fairly impressive attempts. But they aren't quite where Fortune might think.
Blogged using Windows Live Writer at the Marco Polo Gateway Hotel, Hong Kong.
Two thoughts: 1) I (and I think a large portion of the population) believe that my taste is fairly unique and original, and therefore the idea that a computer could figure out what I liked for me is a bit offensive and insulting (and I might just disagree with the computer just out of spite). 2) In order for an algorithm to really make good predictions, I'd have to tell it pretty much everything about me, which becomes a privacy issue. And I don't think it would be worth giving up my privacy just to get good movie recommendations, for example.
Posted by: Webomatica | 22 November 2006 at 00:44
Well, yes and no, statistically you aren't really that unique and recommendation engines don't have to be perfect, just pretty close to right. So if they hit your likes and dislikes accurately (say 65% of the time) that's still fairly useful.
What they are _not_ is accurate enough to predict everything you will like as the article suggested.
Even though others get annoyed at the misses by Amazon, I've enjoyed value from the Amazon recommendations which don't resuppose to know _you_ but say "Other people with your buying habits bought this."
Does that take the edge off it?
Posted by: Jim Benson | 22 November 2006 at 00:49