Stubborn predictions

Bayesian theories of perception have traditionally cast the brain as an idealised scientist, refining predictions about the outside world based on evidence sampled by the senses. However, recent predictive coding models include predictions that are resistant to change, and these stubborn predictions can be usefully incorporated into cognitive models.

That is the abstract to a new article by Yon, de Lange, and Press, via Michelle Dawson.


Bayes meets Kuhn, curtesy of Michel Foucault, and against the philosophy of Henri Bergson.

Sometimes it’s obvious you don’t know what you are talking about. This is behav. Econ. stuff not Kuhn/Foucault.

Somebody has discovered heuristics without recognizing the concept? The wiki article has a pretty good summary of why some predictions are hard to change - or at least the way those predictions are made is hard to change.

'A heuristic technique (Ancient Greek: εὑρίσκω, "find" or "discover"), often called simply a heuristic, is any approach to problem solving, learning, or discovery that employs a practical method, not guaranteed to be optimal, perfect, logical, or rational, but instead sufficient for reaching an immediate goal.'

And here I thought economics was the only field that absurdly idealized human behavior for modeling efficiency.

"Common sense" is a heuristic that is applied to a problem based on an individual's observation of a situation. Or a computer's (i.e., AI's). Recent breakthroughs in AI will allow machines to recognize words and objects and reach a common sense decision. To illustrate, physicians often express the excuse that they don't rely solely on technology in reaching a diagnosis because technology lacks "common sense" (i.e., the physician's judgment based on her experience). Well, physicians soon won't have that excuse.

Free text of the article available here:

Dear Sir/Madam

I am MITIN NIKOLAY SERGEEVICH the Managing Director to ZAO REGIONNEFTETRANS OIL/GAS. We are a certified and authorized mandate that supply Oil and Gas products to buyers and consumers. We represent the major Oil Refineries to the Government of the Russia Federation to help buyers meet their various requirements and needs. We Can supply Aviation Kerosene,Jet fuel (JP 54-A1,5), Diesel (Gas Oil) and Fuel Oil Mazut (M100) REBCO, LNG/LPG,UREA ETC in CIF/FOB base email us ([email protected]) for more clarification and understanding as this is an arrangement and compromise with a take over bid. more of a government bond with incentives and relief. Kindly contact us with our information below for soft co-operate offer SCO.

Email: [email protected]
Tele: +7 (926) 801 5735 
Skype: mitinnikolay

I'd say the abstract is pretty misleading. They are proposing two separate systems, one for perception, and another for action. It is interesting recast goal-driven behavior as a weakened error feedback signal, but I don't find it compelling. Obviously there are lots of behavioral situations where a veridical model of the world is very important. Balance comes to mind.

If I had to place a bet, I think that this group will end up being "right" in the fact that in certain situations it can be beneficial to turn down the error signals in order to essentially accept risk in order to achieve some goal, but doing that as a steady state would never work. Ask Boston Dynamics.

Comments for this post are closed