27 November 2009

Brembs (2006) Brains as output/input devices

I just finished reading an excellent blog post (paper?) by Björn Brembs entitled Brains as output/input devices. I admit that I too have tended to think of brains as stimulus-response machines, paying little attention to spontaneous behaviour and operant learning. Indeed, the old iPlant programming section on my website used to begin "Like a digital computer, the brain generates output from input". On the contrary, Brembs argues, the brain generates input from output. That is, the main function of brains is to control the environment they're in - and thus the sensory input and rewards/punishments they receive - by figuring out the right motor output for a given situation.

"input/output transformations may only account for a small fraction of what brains are doing. Maybe a much more significant portion of the brain is occupied with the ongoing modelling of the world and how it might react to our actions?"

Furthermore, Brembs argues, the variability we observe in spontaneous behaviour is a feature of operant learning: it is a way for the brain to find and develop patterns of behaviour that give it optimal control over its environment. "Faced with novel situations, humans and most animals spontaneously increase their behavioural variability", presumably in order to figure out how this particular environment responds to behaviour. It's the environment that responds, not the animal. Perhaps even the subtle variability we see in the invertebrate feeding system is an expression of the molluscan brain trying to figure out the best way to eat this particular sea-weed. If so, such variability should be selectively enhanced by reward learning. Is it? Other questions:

  • How is behavioural/neuronal variability generated in small and large brains?
  • What % of behavioural/neuronal variability is really subject to learning/operant control in different networks?
  • What features of neuronal activity are most likely to be subject to learning/operant control? In other words, where do we look? Spike rate of individual neurons? Network patterns? Duration of the different phases of motor programs?
  • How is reward conditioning/operant control of spontaneous variability instantiated in small and large brains?
  • How can we incorporate output-input functions in artificial neural networks and robotics? That is, what kind of tasks could such networks realistically perform today or in the future?
  • What is the cultural effect (within in the neuroscience community and generally) of treating brains as output/input devices rather than input/output devices?

P.S. It was particularly stupid of me to emphasise the input-output side of things on the iPlant website, as the whole point of conditional rewarding brain stimulation is to modify output-input learning by rewarding beneficial but endogenously under-rewarded behavioural variations (rigorous exercise in morbidly obese patients etc.) with electrical pulses to the reward system.
Post a Comment