-
Website
http://publishing2.com/ -
Original page
http://publishing2.com/2007/04/15/cumulative-advantage-explains-web-20-myspace-the-a-list-techcrunch-digg-and-so-much-more/ -
Subscribe
All Comments -
Community
-
-
Top Commenters
-
Popular Threads
I wrote about this with regard to Digg, but yes, it does broadly apply across all web media:
http://mashable.com/2007/04/15/is-digg-the-resu...
We used to call this "network effects", of course. Either way, it's pretty clear that cumulative advantage would be more noticeable on the web, where people are making decisions based largely on what others think. Top 10 lists, technorati top 100 etc all increase cumulative effects. You'd think we would have realized this by now and figured out that what people want is not popularity, but personalization.
No, I think people WANT to know what's popular.
We watch popular TV shows because it gives us something to talk about around the water cooler the next day. If we don't watch the show, we are out of the conversation.
We like to know what's popular on the web so we can talk about it minutes later over IM.
The problem with popularity vs. pesonalization is that, as Watts observes, we are intensely social creatures and thus personalization is actually a function of popularity -- our personal tastes are naturally driven by what other people think is popular.
BUT, that doesn't mean we don't have a sense of personal taste -- what happens is that our tastes are both driven by AND, in a sense, overrided by what's popular. We watch popular TV shows to be part of the conversation -- to Hashim's point -- BUT we do so even if we actually don't like those shows.
"the long-run success of a song depends so sensitively on the decisions of a few early-arriving individuals, whose choices are subsequently amplified and eventually locked in by the cumulative-advantage process"
This is more true or less true, depending on the type of network.
In networks that confer greater advantage to the most advantaged, this is more true. But in networks that limit the advantage of the most advantaged, this is less true.
By reducing the amplification any given individuals receive, we increase the probability that the message, whatever it is, will be filtered by people who are actually qualified to do so.
When the rich are allowed to get richer, through no merit of their own, but merely because they are rich, then they will spend money foolishly and wastefully. But when this same money is spread over a larger number of people, it will be spent more carefully and more productively.
This is often the case with anything that succeeds based on social approval/voting. It is human nature to want to "be part of something," and all too often I have observed people making foolish choices simply because they didn't want to be perceived as outside the norm.
That does sound good in theory. However, if for example, we reduce the amplification an individual receives simply because he may benefit from the outcome, we risk shooting ourselves in the 'collectivist foot' by losing the possible benefit of whatever he may be advocating. All because we rushed to suppress his influence on the group in order to prevent his gaining a financial benefit. The opportunity cost may be more than we think.
One might be tempted to buy into such a generalized indictment of the wealthy simply because it appeals to a misguided sense of social justice.
But the social reality tells a different tale. Typical high income/high net worth individuals share common traits like wise spending and investment habits.
It is typically the less-productive members of society who make foolish spending and investment decisions, as well as lifestyle choices.
So in the end...are some people more qualified to "vote" on the success of an idea than others? YES
However, when it comes to pop music/art/fashion and other less important ideas, it just doesn't matter.
Tune your radio to the local top 40 station. The fact that unqualified early adopters have an influence on the success of a song has been painfully obvious for decades.
Thankfully the market has a way of vetting things that truly matter.
While on one hand, it's nice to see something we know intuitively - that things already declared popular receive more attention than items without that distinction - on the other, the experiment as performed dramatically over-simplifies the situation for one very big reason: positive feedback loops.
The most successful emergent systems, those where valuable order arises by individuals acting autonomously, are successful through a balance of positive and negative feedback. Everything from thermostats to a human's sense of balance to ants' decisions on the best place to store their waste. Can you tell I just read Stevn Johnson's Emergence?
How dramatically would the results change if you simply gave users the ability to vote songs both up and down? Hard to say, but I would have to guess it would make a difference.
Creating a system that produces quality from simple rules is amazingly complex - netflix is offering a cool $1M for one that improves their movie recommendations. I wouldn't dismiss the crowd's wisdom from an experiment that doesn't allow the wisdom to emerge.
Your right - Google's algorythym is very different from the experiment, but not for the reason you state - that it doesn't show users the score and its seeds aren't random. And everytime someone searches, the score is right their on the page: results 1 - 10000000. The top search results get more exposure, more links, and longevity at the top - cummulative advantage in effect. And ask someone that has had a site running on the same domain since 96 if their google page-rank hasn't benefitted from their random early arrival to the page-ranking party.
Google is successful in many ways because it harnessed the positive feedback of links in, and balanced with all sorts of other network information feedback - a link from a site with a page rank of 1 is of less value than a 2, and links from virtual hosts on the same IP are downgraded to discourage gaming - the algorythym is their very complex secret sauce.
So my bottom line is that cummulative advantage exists - how else can you explain Sanjaya? - but the makers of effective software and systems know this and the look for rules that amplify the right signals and soften the right signals.
I've often criticized Digg for this reason - the site does everything it can to get you to vote for things that are already popular. You see the number of Diggs next to every article, and you can even sort by that value.
But other sites deal with it better. Google's magic is based on network effects where each inbound link is a vote, as every person who creates a link is doing so pretty much independently and unaware of what anyone else is linking to. Flickr's interestingness is opaque enough that it doesn't seem to be subjected to cumulative advantage. The blogosphere is a bit of a gray area where we can see both at work.
Cumulative advantage, incidentally, is also why publishing political polls ought to be illegal in a campaign season.
1. Build in a reputation system (like Slashdot) so that people with good taste have greater influence.
2. Make sure users have to sample a variety of songs.
3. Implement a tournament ranking system where users rate 5 random songs against each other at a time. The best 2 would go on to the second round.
These changes would dramatically increase the chances that the crowd recognizes both genuine hits and lesser-known gems.
Everything depends on the ratio of songs to users and users per group. Numbers matter because if there are too many songs per user, they will try only a small sample songs and so less likely find the objectively good songs.
Fortunately, the right architecture can keep the random rich from getting richer:
1. Build in a reputation system (like Slashdot) so that people with good taste have greater influence.
2. Make sure users must sample a variety of songs.
3. Implement a tournament ranking system where users rate 5 random songs against each other at a time. The best 2 would go on to the second round.
Such changes would dramatically increase the chances that the crowd recognizes both genuine hits and lesser-known gems.
His modeling of the less than dramatic impact of influentials on the spread of a contagion through a population was convincing--BUT--I don't think his model really applies all that well to the web...
Because in his model, *every* member of the herd had a *connection* to those around them through which they could infect(influentials simply had a higher probability to infect via these connections). But influentials on the web are different in that their "connections" are to DISCRETE populations--populations whose respective members don't necessarily have connections to the other population's members.
Watts's model describes the impact of promiscuous sheep on the spread of something within a specific herd. But THE IDEA VIRUS and THE TIPPING POINT describe the impact of a shepherd with access to multiple herds.