Sunstein started by laying out the results of a 'Colorado experiment'. Here's how it went. Boulder is a left-leaning community in Colorado while Colorado Springs is right-leaning. Groups were pulled together from these two separate communities to talk about climate change, affirmative action and same sex marriage. Their views on these topics were measured before the discussion. In private, many people tended to be uncertain and tentative about their opinions. Then Boulder people had a short discussion about this topics with other Boulder people while the Colorado Springs people talked to others from Colorado Springs. After these discussions with like-minded people, they become more confident, more unified and more extreme. As Sunstein said, this was not all that surprising, but it was useful to verify the process of people being influenced by others in a group experimentally rather than relying on intuition.
Sunstein went on to describe the effects of mixed and homogeneous composition of judicial panels. A laborious analysis of the decisions of three-person judiciary panels, comparing homogeneous (as to who was the President when they were appointed) and mixed (people appointed by different presidents of different parties) panels showed a rather surprising result. The best predictor of their decisions was not their own political leaning but the leaning of the others on the panel.
He gave many examples of people's resistance to new data, when they have firm opinions already. Add to that the fact that most Americans get their news from Facebook, which filters news to suit their tastes - this generates more clicks and more opportunity to sell advertising. He showed us a quote from an earlier FB statement of how their NewsFeed works:
Our success is getting people the stories that matter to them most. If you look through thousands of stories every day and choose the 100 that were most important to you, which would they be? The answer would be your News Feed. It is subjective, personal and unique - and defines the spirit we hope to achieve.
It's easy to see how this approach would solidify extreme positions and lead to polarization. Facebook is now making moves to modify how News Feed works, under considerable public pressure. Sunstein made a strong point that we shouldn't be algorithmed into being extreme, which the original Facebook algorithm certainly did. He got a good laugh when he said we needed to made algorithm a verb, but we needed a shorter word. Unfortunately algored wouldn't really work.
Sunstein was discussing topics from his latest book #Republic: Divided Democracy in the Age of Social Media. I'm looking forward to reading the book and hearing more about his research, because I'm sure I haven't done justice to what he talked about.
After the talk I made a suggestion to Sunstein. The New York Times often runs a section entitled What the Conservative Media that it might be interesting to see the relative click rate on The New York Times' section called What the Conservative Media are Saying (or some such title). Compared to the laborious research on the judiciary panels, this would provide simple quantitative results. I know what I'd be betting. But Sunstein was scrupulous in not claiming positions for which he didn't yet have experimental verification. His book should be a good read.