Cass Sunstein (2006), Infotopia: How Many Minds Produce Knowledge, Oxford: Oxford University. Elaboration on Sunstein’s more academic (2005) “Group Judgments: Statistical Means, Deliberation, and Information Markets,” New York University Law Review (80), 962–1049.

Sunstein critiques (but does not condemn or abandon!) deliberative decision making. In deliberative groups, informational influences and social pressures prevent useful info from coming forward (they can also prevent bad info from cluttering up the show). Sunstein cites various research that finds statistical groups (no deliberation, just ask separate individuals and average their responses) can perform just as well as and often better than deliberative groups. He notes that prediction markets overcome some of the problems that arise in deliberative settings. In prediction markets, folks with good info have a strong motivation to participate: they can bet their money and win! Likewise, folks with faulty or no info have a motivation not to participate: they’ll lose their money!

Some key theory:

  • Condorcet’s Jury Theorem:

    • Assume the following:
      • A group of people is trying to decide a question by majority vote.
      • There is a way to evaluate whether that vote is correct or incorrect.
      • Each voter’s probability of being right is independent of other voters’ probabilities.
    • Condorcet says:
      • If each voter is more likely to be right than wrong (“p > 0.5”), then the more the merrier (including more voters increases the chances of obtaining a correct majority vote).
      • If each voter is more likely to be wrong than right (“p < 0.5”), then the more the murkier (including more voters increases the chances of obtaining an incorrect majority vote).
  • Price System Theory:
    • Friedrich Hayek (1945), “The Use of Knowledge in Society“: short paper, big idea
    • open market price system aggregates dispersed, unshared information to produce a pretty accurate assessment of the value of goods and services
    • central planning could never do as a good a job (said Hayek, fighting the Cold War)
    • Hayek didn’t discuss deliberation, but Sunstein says Hayek applies!
    • Hayek certainly provides basis for prediction markets, where participants put a value on statements about what may happen

Some key passages:

  • [p. 67] “People are much more willing to say what they know if other dissenters are present and if a principle of equality is widely accepted within the group.” Suggests to me that being contrary can serve the public good!
  • [68] groups with cohesive ties exert more social pressure, cause more self-censoring. Investment clubs with close social ties lose lots of money!
  • [85] “cognitively central” vs. “cognitively peripheral”: If you are cognitively central to a group, you have ideas that are more familiar to more people in your group (team, board, neighborhood, community, etc.). If you are cognitively peripheral, you have ideas that aren’t shared by many others. You know things that perhaps no one else does.
    • “Well-functioning groups need to take advantage of cognitively peripheral people. But it turns out that cognitively central people usually have a disproportinoate influence in discussion, and they also show higher levels of participation in group deliberation. By contrast, cognitively peripheral people end up having little influence and do not much participate.”
    • My observation: this finding suggests the key to winning a debate or public vote may be less providing new persuasive information and more framing one’s positions in terms of popularly held knowledge. This is not how things should be; the greatest leaders find a way to tell us what we don’t want to hear but not losing our support. Still, the practical politician can’t ignore the influence of cognitively central people and our natural desire to hear our shared knowledge reinforced.
  • [201] “Consider here a fundamental redefinition of what it means to be a team player. Frequently, a team player is thought to be someone who does not upset the group’s consensus. But it would be possible, and a lot better, to understand team players as those who increase the likelihood that the team will be right—if necessary, by disrupting the conventional wisdom.”
  • [202–203] See Luther Gulick (1948), Administrative reflections on World War II: The U.S. outperformed the Axis powers in WWII because dissent and criticism within democracies produces better decisions, even in wartime! Totalitarian leaders live in information cocoons.
  • [211] Designating a devil’s advocate in your group can be useful, but authentic dissent is much better for group performance.


Sunstein isn’t sanguine about blogs. He sees lots of polarization rather than good collaborative deliberation and knowledge-building. But might that merely be a feature of political blogs? Blogs for scrapbookers and cancer survivors don’t polarize, do they? Maybe there’s something in the rhetorical nature of political blogs that differentiates them. Political blogs are persuasion devices, not decision support systems or community-building exercises. Now that’s using political in the smaller definition of the word, not the grand Aristotelian meaning.

See also Sunstein’s list of prediction markets.

Hey Baby!” — article on IBM researcher Mark Wattenberg’s NameVoyager app, a cute little data analysis tool that allows quick and easy comparisons of the popularity of baby names. The app got 500,000 hits in its first two weeks, with no ad campaign or other press.

Key features that Wattenberg thinks make this interface popular:

  1. Easy interaction: not complicated! Users just type in the first letters of names they are interested in and move the mouse over the graph for details.
  2. Social data analysis: people interact, converse about what they find, encourage other people to try experiments. They learn from each other and stimulate each other’s curiosity.
  3. Immediate information: you type in a few letters, and pow! you get a refreshed graph.

So consider these factors with the HULK. Want to get people interested? Make those stats available immediately. Make it as easy as possible to get information on who’s voting how (and how they’ve voted historically). Make it easy to run numbers and then to tell others how to run those same numbers, to replicate any particular interesting bit of data mining and experimenting.