Late last year Netflix offered $1 million to anyone or any group that could improve their ‘matching algorithm‘. The Netflix ‘matching algorithm’ suggests films that you would like based on previous films that you have rated.
I looked into joining this competition for two reasons: 1) I've been
thinking about this subject for a long time and this was a good chance
to test my thinking; 2) this is a core question in marketing and human
networking. 'Matching' is a common subject friends talk to each other
about....finding movies in common, finding songs and food.
Please think about the issue for a moment. If a machine can ask you
a few questions and then know your tastes and interests to the point
where the machine is accurate 19 out of 20 times in the future,
wouldn't you be stunned? Wouldn't you be willing to let that machine
make nearly all minor decisions for you? Would it be fair to say that
that machine has a good understanding of how your mind works? My answer is 'Yes.'
Netflix pays $50,000 every quarter to participants in the contest based on the progress they have made.
I didn't join the contest after I looked at the size of the test
files that are used in building and testing the algorithms. The files
would barely fit on my computer and all the math programs I have would
find processing almost impossible.
The latest winner of the $50,000 progress award is a team at AT&T. Congratulations.
If you aren't a statistician you are unlikely to know that AT&T
Labs, formerly Bell Labs, was the home of most statistical inventions
in the past 50 years. I look forward to seeing the product of this
contest, it will be important for all of us.
This contest is also a lesson, a lesson that can not be repeated too
often...human beings will do more in a mildly competitive environment
than any other way. A mildly competitive environment is precisely what
commerce is all about. (Please don't write me about commerce
being like blood lust or warfare, it isn't. Commerce is sports-like
with striving as the mode of competition.)
Recent Comments