Thursday, September 22, 2005
Google's Big Problem, or Google vs the Social
NYT piece contrasting Google Answers with Wondir, etc.
What's Google's Achilles Heel? Well, there's a big glaring one at the moment. It's not (as every analyst seems to point out) that Google's a one trick (ie, contextual ads as far as the eye can see) pony - contextual ads across domains won't be commoditized for a very long time.
The problem is that in a world of media hypercompetition, algorithmic approaches aren't as efficient at large scale as social approaches. This is what I call connected consumption. Here's a nice article summarizing.
To make this intuitive, if it's not already, think about either finding new music via Last.fm, or relying on HSS. Which is more likely to be efficient? Pretty clearly, Last.fm.
But, as the article points out, even collaborative filters eventually break down in terms of novelty, and everyone gets hyperpolarized into their own microdifferentiated niches.
The interesting thing is that there seems to be a kind of exponential relationship between the algorithmicness of your approach and novelty - the more algorithmic it is, the less efficiently it can generate novelty. On the other hand, here's an example of a purely social approach generating some pretty cool novelty hyperefficiently (talk about digging through the crates!).
I mentioned this a few weeks back - if we're thinking in terms of the DJ metaphor, the best DJs also play tracks that infuriate and challenge people - not more of the same stuff over and over again. Think John Peel.
So what does all this have to do with Google? Simple - Google is heavily algorithmic. It isn't very good at social approaches (as you might expect from an algorithm-driven company). Most of it's efforts to build social competences ads have been soundly beaten and reduced to irrelevance by (simple) social approaches. It's a long list - Orkut, Google News, Froogle, Groups, etc. Google doesn't do communities (which most social approaches require) very well - it does the opposite; make hyperefficient markets.
Now, this is a big problem (at least, it is if you wanna take over the world). That's because as micromedia explodes, algorithmic approaches become less and less valuable. They'll certainly remain valuable in vertical and narrow domains (viz, SideStep), but the 100x gains will flow to a new kind of play - plays which, instead of going algorithmic, leverage peer production to engineer entirely new kinds of sociality. You know what I'm talking about - Skype, Flickr, Metafilter, Habbo, etc.
I haven't really named these social plays yet, but I think we should - any ideas?
I think Google knows the power of people, and will use a Last.fm approach to search. They recently introduced their "my personalized search" feature, and I think this will lead to something like the following:
1. using Gmail it can identify your friends, and can check their search histories and what they clicked on and recommend such sites to you when you do a search for a similar term.
2. it is only a matter of time before Google buys or makes an RSS reader. they will then know what blogs you subscribe to, can find other people who subscribe to the same blogs, and personalize the algorithm accordingly.
Very good post...not sure of what drives this statement:
>>That's because as micromedia explodes, algorithmic approaches become less and less valuable. <<
Isn't this definied by the nature of the algorithms rather than the type of algos used by Google up until this point?
i.e., couldn't one create algorithms that incorporate some of the social combinatorial characteristics that you're talkinga about and create a new type of algorithm?
Am I missing something? It seems like it would be very valuable for google to just add higher weight to the search results that are clicked on most often. This way they incorporate each searcher as a relevance analyst, increasing the value of their search.
Are they doing this already? It's hard to tell.
The problem with using what my friends are searching for, or what rss feeds I subscribe to, into the algorithm is that it's just not that big a dataset. I almost never search for things that are related to the rss feeds that I read. I search for information i am not getting from my feeds...
"It seems like it would be very valuable for google to just add higher weight to the search results that are clicked on most often."
their april 2005 patent filing claims that they are already doing this. a problem with relying on this too much is the "rich get richer" syndrome where sites that rank higher always rank higher because they are getting the most clicks since they are at the top.
identifying you based on your rss subscriptions and your friends does not necessarily limit the dataset, as google could make connections between you and your friends, your friends' friends', and so on. "small world" effects take over and the dataset quickly becomes large and, in my opinion, probably quite accurate. yahoo has already showed some savviness as they are developing search along the same lines with their "my web" search campaign.