After #IndyRef 4: Credulous Bayesians and the importance of maintaining diversity of thought

Screen Shot 2014-09-29 at 16.43.54Earlier today I was asked to speak on the business of shared space and the political effects of social media in Northern Ireland at an event in Liverpool in the new year. Not long afterwards I came across this paper which was written in 2007 by Edward Glaeser and Cass Sunstein.

In the wake of the largest and in many ways most successful social media campaign we’ve ever seen in Scotland I think it explains some of the unmet expectations involved:

The information of the crowd provides new data, which should lead people to be more confident and more extreme in their views. Because group members are listening to one another, it is no puzzle that their post-deliberation opinions are more extreme than their pre-deliberation opinions.

The phenomenon of group polarization, on its own, does not imply that crowds are anything but wise; if individual deliberators tend to believe that the earth is round rather than flat, nothing is amiss if deliberation leads them to be firmer and more confident in that belief.

Curating cognitive diversity within groups may provide a useful counterbalance. Dr Scott E Page of the University of Michigan rather sparely puts it ‘Crowd error’ = ‘Average error’ – ‘Diversity’. As Nick Cohen recently argued in the Observer, the serious lack of diversity within modern (often, but by no means exclusively governmental) institutions has driven out difference and/or dissent creating what Chris Dillow calls “Bubblethink“.

Glaeser and Sunstein again…

… we suggest that social learning is often best characterized by what we call Credulous Bayesianism. Unlike perfect Bayesians, Credulous Bayesians treat offered opinions as unbiased and independent and fail to adjust for the information sources and incentives of the opinions that they hear. There are four problems here.

First, Credulous Bayesians will not adequately correct for the common sources of their neighbors’ opinions, even though common sources ensure that those opinions add little new information.

Second, Credulous Bayesians will not adequately correct for the fact that their correspondents may not be a random sample of the population as a whole, even though a non-random sample may have significant biases.

Third, Credulous Bayesians will not adequately correct for any tendency that individuals might have to skew their statements towards an expected social norm, even though peer pressure might be affecting public statements of view.

Fourth, Credulous Bayesians will not fully compensate for the incentives that will cause some speakers to mislead, even though some speakers will offer biased statements in order to persuade people to engage in action that promotes the speakers’ interests.

Our chief goal in Sections V-VIII is to show the nature and effects of these mistakes, which can make groups error-prone and anything but wise, especially if they lack sufficient diversity.

Or as Carol Craig noted during #IndyRef…

…there are times when it makes sense to be optimistic (or use optimism building techniques if you are prone to pessimism) and times when it is better to be pessimistic. He writes: ‘The fundamental guideline for not deploying optimism is to ask what the cost of failure is in the particular situation.

If the cost of failure is high, optimism is the wrong strategy’.

This may be one reason why ‘smart’ hierarchies still retain a capacity to beat the excessively optimistic and risk taking crowd.

, , ,

  • terence patrick hewett

    Makes important points for any insurgents.

  • mickfealty

    Or indeed, established institutions. this is part of the poison within the public appointments system in the south, and to a certain extent in NI.

    Rampant insiderism!

  • Kevin Breslin

    The scientific paper was on social networks, but the Craig comment was on individual decision making. Your comment on risk-put analysis is also confined to individual decision making. However since lone individual risk-put behaviour is pretty much a survival instinct in all living creatures, it is a very bad example of diversity among uniform groups. Diversity in this case comes from “people with the strategic vision of a lemming” … not the real lemmings but the memetic myth of course.

    However the big flaw in this argument is that “thought diversity” has limited impact on society, only an individual’s impact on society or not.

    A person could have diverse thoughts and limited communication networks.

    A credulous Bayesian could for example accept a diversity of facts, but because he is credulous he has an indeterminate bias.

    An incredulous Bayesian could reject a diversity of facts, but also has an indeterminate bias over what he does accept.

    The “perfect Bayesian” accepts of all facts he senses and reaches some random equilibrium based on some perfect reject/accept formula… some sort of esoteric perfect bias.

    However this particular self-optimised “gnosis” or even “agnosis” will still present as an imperfect bias to a social relative or norm in a social Bayesian hierarchy.

    Such is the case with the memetic formulae …

    a) If the cost of failure is high, optimism is the wrong strategy

    b) Smart Hierarchies retain a capacity to beat optimistic and risk taking crowds.

    The big problem with these is a network is not a person nor even a social interpretation of a person, but the collective. The first is quite clearly a case of correlation means causation, when it not necessarily so, the second is true even under Cromwell’s law, as it does not highlight any mechanism at work, just probabilistic possibility that could happen randomly anyway.

    A social network relies on communication, and the Bayesian network is the simplest ad-hoc mathematical model for grouping up individuals. So if we sum individual “formulas” for deciding the behaviour of the group, a new social “functional” is created that determines the crowd behaviour.

    In this functional there are several “failures”, “costs”, “strategies”, “hierarchies”, “optimisms”, “risks” competing against one another. In this functional there will also be “successes”, “benefits”, “experiences”, “pessimisms” and “puts” as well. And all these individuals have their own different cost formulas. This creates a very complicated persuasion matrix/tensor of commonality of both individual counterbalances and social counterbalances.

    But even in a perfect cohesive group that believes failure costs are high, there is no “gnostic” truth to believing that optimism is either a right or a wrong strategy for that group. Take a group of soldiers in a war zone, they could dessert, mutiny, fight, kill themselves, surrender but doing so in an optimistic or a pessimistic manner does not determine if what the person or group has done is right according to their conscience or reason, nor does it have control over the choices of other agents or the environment around them. That may be down to analyser bias.

    The other thing with the premise, that diversity in society is a sort of absolute importance is that it becomes credulous and naively so, as diverse thought is no panacea to removing group uniform errors, and because diversity can also remove uniform accuracies and orthodoxies as well.

    For example, Why would a doctor need to know a diverse set of options on how to treat a patient, when he’d want to have a tried and trusted method over a diverse rang of infinite monkey quackery, failed methods, and unknown risk?

    This leads to the creation of the “know-it-all” … the know it all does not know everything, but knows enough of society’s diverse range of opinions to only accept, even with limited skepticism, the things that are specific to his life and experience.

    We are all Know-it-Alls … therefore the myth of a group think that does not allow fluctuation due to diverse opinions is created in the social conscience. However like gasses, magnetic fields and radiation levels … just because you can’t see it, doesn’t mean it’s not there.