Networks

We have seen in the class that aggregating opinions and information from a large number of people is hard to do correctly.  One possible pitfall is the information cascade, where the consensus is seen as more accurate than any individual’s opinion, thus preventing new opinions from influencing the set consensus.  We have seen the problem with voting, how it is impossible to aggregate a group’s preferences in a way that satisfies all the conditions of Arrow’s Impossibility Theorem.

One interesting alternative to these viewpoints is Unanimous AI’s Swarm Intelligence project.  They run live experiments where a group of participants all share control of a pointer, which they individual try to move onto spaces representing the answer they believe is correct.  The questions that they ask these swarms range from policy questions (what should the minimum wage be?) to predictions (who will win the superbowl?) to questions of opinion (what candidate best represents your values?).  Experiments have shown that swarms often form better predictions than aggregating votes or asking experts.  The system allows there to be a real-time exchange of information among users.  Unlike traditional polling, individuals can learn what the current consensus is and factor that into their decision.  Unlike when information cascades occur, however, the individuals can express their own opinion, altering the consensus.  While more testing must be done to see if the technique is reliable and useful, it is certainly a promising idea that has a lot of potential.