'The Economist' (p. 73; 7 Jan) reports that two US researchers, Thomas Griffiths of Brown University (RI) and Joshua Tenenbaum of MIT, have been studying people’s ability to predict and are suggesting that the evidence of their works indicates that the brain may act like a Bayesian probability ‘machine’.
Thomas Bayes was an English clergyman (1702-1761) who believed that the prediction of future events could be based on evidence of one or two prior events. His theory, though favoured by many, has generally been displaced by the “frequency” school of prediction, which relies upon sampling from a large population of data.
Bayesian theory is at the core of internet search engines and automated ‘wizards’. Some psychologists suggest that the human brain might be a Bayesian-reasoning machine because of the way that the mind perceives the world, makes plans, can master language, decides upon actions, and generally reason. This reseach study seems to support this view.
The key to Bayesian prediction is not an extensive, unbiased sample (the aim of the ‘frequency’ school of statisticians) but to have an appropriate assumption about the way the world works—in essence, a hypothesis about reality—that can be expressed as a mathematical probability distribution of how often events of a particular magnitude might happen.
The researchers gave individual pieces of information to each of the 350 participants in their study and asked them to draw a general conclusion. For example, given the amount of money that a film had earned since its release they were asked to estimate what its total earnings would be, even though they were not told how long it had been on release. They also asked about things as diverse as the time it takes to bake a cake (given how long it has already been in the oven), and the total length of the term to be served by an American congressman (given how long he has already been in the House of Representatives). All of these things have well-established probability distributions, and all of them were predicted accurately by the participants from the lone pieces of data provided.