Elephantstrunk

Reading here about flexible perceptions of reality illustrated a shortcoming of Bayesian updating as a model of learning.

In the Bayesian model learners have prior expectations about the outcome of an event, the event occurs and the probability of the outcome given expectations is used to update the expectations. In this model learners obey a mathematical theorem and their beliefs converge. So far so sensible but the problem is that particular outcomes are rarely well defined. More plausibly learners select an explanation of an event and only then adjust their expectations and they will choose the explanation that requires the least adjustment of expectations.

It is possible that there is a less obvious Bayesian process going on but mostly there is a refusal to learn implemented by deferring the realisation of events. So, if I believe that I am the greatest genius since Gauss and I do badly in a maths exam I can hypothesise that the exam was fair and accept that I am not the greatest genius since Gauss or I can decide that the exam was unfair because I am the greatest genius since Gauss and therefore a bad result means that exam was unfair.

Politicians in their inability to admit wrongness are particularly prone to this, at least in their public rhetoric but talk about the UN is most revealing about this. The UN has no significant power of its own and yet it is blamed for almost everything. It is also why some apparently trivial issues can assume an apparently exaggerated importance. Sometimes the obviousness of the wrongness of something is more important than the magnitude. For example, the Valerie Plame affair is a less serious issue than launching a war on Iraq but much harder to provide alternative explanations for.

This makes the current practice of going to war without really saying why horribly damaging.