ETech: Surowiecki's Wisdom of Crowds
Surowiecki's book, The Wisdom of Crowds, unsurprisingly formed the basis for his great talk. Some notes, mostly cuz I'm too exhausted to do anything else:
In last five years, there are questions about the value of collective action. Examples of collaboration: linux, wikis, blogs, slashdot, open source software, google, etc. We have a sense that the underlying structures are similar, but it is useful to think how these things are different.
Not all forms of collective action are created equal. If you use the wrong form of collective intelligence, the reprecussions could be bad.
Groups, under the right circumstances, can be remarkably intelligent. A group must fulfill these four criteria to be intelligent: it must be diverse. Members must be acting independently. The group must be large. And there must be some method to aggregate each individual's opinion.
The model works well when there is a true answer (there are 254 jellybeans in the jar) or there is a probabilistic answer (Smarty Jones is a 3-to-1 favorite to win the race).
The key why this works is because individuals are acting on private information (whether it's bad or good). The goal is to tap into disagreements.
The ant hill example from Steven Johnson's Emergence: ants are dumb agents. Collectively, they can produce stunningly intelligent results. They are good at finding food with the least amount of energy. They put their graveyard at the furthest spot from their main food souce, but still equidistant from the ant hill, and they do this instinctively. Ants follow simple rules and pay enormous attention to those around them. Interaction is the key to intelligence.
But, human beings are not ants. If there's too much interaction humans become less intelligent. The more we talk to each other, the dumber we become. Small group groupthink. "It is better to fail conventionally that succeed unconventionally." Human beings like the comfort of the crowd. Portfolio managers spend a lot of time following what everybody else is doing because it is a way to appear reasonable.
Human beings are imitation machines because it can be a useful way to learn. It's a quick and dirty heuristic that works well. If a lot of people think something is valuable, it most likely is valuable. The problem is when humans imitate slavishly or without thinking what they're doing, you have a problem. When this happens, you have an information cascade. After a certain point, once an information cascade gets going, it becomes difficult for people making decisions later in the process to not do what everybody else is doing.
The classic information cascade example. There are two restaurants, both empty. Person one comes along, doesn't know which one is better, and so chooses one randomly. The second person comes along, sees the empty restaurant, figures something is wrong with it, and so he chooses the other one because the first person is there. The third thinks there must be something wrotng with empty restaurant and does the same. Lather, rinse, repeat. Once the cascade gets started, it becomes rational to do what everyone else is doing. As long as you assume everyone else is rational, and they have some info on which they base their decisions, it becomes rational to do what everybody else is doing.
Pascal said all problems in world stem from one simple fact, man cannot sit quietly in a room by himself.
So, how can you have interaction without having information cascades? How can you have interaction without losing factors that develop group intelligence?
1. Keep ties loose. Better off if ties are looser rather than tighter. Loose ties minimize the influence those around you have on your decisions.
2. Try to constantly be exposed to as many diverse sources of information as possible. Injecting random information into a system is good. Example: there are three groups who are charged with solving math problems. Group one is average math students, group two is genius math students, and group three is both average and geniuses. The mixed group almost consistently beats out the genius group, because even if the less intelligent group members know less, what they know is different, and this helps the group's performance.
So, you can be too connected if the connections are the wrong kind, especially if the connections reinforce your knowledge and prejudices.
The flip side of Pascal's isolation is the cacophony you find on the net. Both allow help you avoid the constraints of the tightly networked model and arrive at the same place, which is independence.
Technorati tags: etech