Teasing Out The ‘Wisdom’ From The Crowd
Should some voters’ votes matter less?
The MIT Technology Review just came out with an interesting piece on generating accurate estimates to questions using crowd-sourced data.
Neuro-biologists Gabriel Madirolas and Gonzalo De Polavieja at the Cajal Institute in Madrid, Spain, came up with a way to correct for bias in “Wisdom-of-crowds“-like estimations by quantifying just how much people are influenced by certain key decision-makers in their vicinity.
What they concluded is very interesting. Just as the formation of social networks is heavily influenced by preferential attachment (you’ll join the one your friends are already using, effectively making that network stronger), the way you vote (or rather, how populations vote on average) is just as heavily influenced by the political preferences of one’s peers.
Pure Democracies have long been known to be vulnerable to chaotic swings in social cohesion. This is one of the primary reasons that the United States is set up as a Republic, but this new method may actually allow us to attenuate the kind of social volatility typically associated with Democracy without having to appeal to any overt regulatory body, such as a Congress or a Senate.
In elections, for example, votes could take place on people’s mobile phones at official polling stations or elsewhere in the form of an A/B split test, where the experimental group is exposed to a vote “suggestion” based on previous voters’ decisions (think NetFlix recommendations, i.e. 57% of people like you voted for…), and a control group which simply votes based on personal considerations. The resulting bias could then be teased out and the final vote generated from a “correction”, which places smaller weightings on the votes of “peer-prone” members of the population. I can even see people becoming more conscious of their votes because they would ultimately learn, from friends and from public knowledge, that they were being tested (expectations tend to moderate volatility surrounding predictions about the future).
I believe this would actually allow the crowd to converge on a more optimal decision, even if individuals were made fully aware of exactly how much government officials were keeping an eye out for preferences linked to social contagion. This reminds me of a well known mechanism in game theory: the second-price auction, which results in the highest bidder paying the second-highest price. By analogy, the incumbent political candidate would receive the “second-highest” vote, or a referendum would administer the “second-best” option, one based not on links in social preferences, but on the actual value of an individual vote, or close to it.
Orwellian at first glance, sure, but we’ve always known that unfettered Democracy is less than perfect. And I think that having a Senate in any case is far more prone to corruption than simply “teasing out”, or rather diminishing the power of particular clusters of votes, which are based too heavily on relatively arbitrary and contagious swings in social preferences. This brings the power back to the voters, cuts out unnecessary pork from political dealings, and perhaps does away with the need for a political class altogether.
So my question to you: is it permissible to employ statistical and computational methods to reduce bias in large-scale social decision-making processes? If so, who should be in charge of this new tech?
References
arxiv.org/abs/1406.7578: Wisdom of the Confident: Using Social Interactions to Eliminate the Bias in Wisdom of the Crowds, by Gonzalo De Polavieja, Gabriel Madirolas.
