A thread:

When neuroscientists “investigate perception”, they require some sort of behavioral measure to distinguish between different “perceptual outcomes” (e.g. appetitive vs aversive stimuli). In humans, this can often be just a verbal report, but in nonhuman animals it 1/n

usually requires training the animal on some task. Neuroscientists will take the results of these experiments and attempt to generalize them beyond the context of the task they trained the animals on. But why do they believe they can do this? Why do they think that the 2/n

phenomena they observe have something to do with a generic “perceptual processing” and not the task specific dynamics of the organism-environment system in the context of a learned behavior? Training the animal on the task being used for the behavioral measure of perception 3/n

means that the organism is learning to assemble the resources available to it into a task specific dynamical system that solves the task at hand (e.g. mice licking in a particular direction in response to whisker stimulation). It is not clear (and probably dead wrong) that 4/n

one can interpret these experimental results as a window into the neural basis of generic perceptual processes. And that’s because there’s no such thing as generic perceptual processes or computations. In particular, we cannot take the behavioral measure being used as a raw 5/n

“report of internal perceptions and computational processes”; it is not possible to divorce the report from the process of perceiving, since in the experimental set-up, performing the task which serves as the report is only possible via ongoing perceptual coupling to the 6/n

environment. The animal does not perceive some stimulus, do some computations, and then act in a separate linear fashion. Rather, the process of perceiving the environment selects the action that will be used as a report because of the animals learning history (being trained 7/n

on the task) and the fact of the environmental context (the experimental set-up). In other words, what a neuroscientist would call “perceiving the stimulus” is not some generic process that is the same across all tasks (e.g. seeing a blue light and then pressing a button vs. 8/n

seeing a blue light and then giving a verbal report) but is in fact dependent on the action that is used as a behavioral report due to the subject’s learning history and context. One must then ask, “Are neuroscientists really measuring general perceptual processes?” The 9/n

answer depends on the theory of perception that one invokes. If a neuroscientist hooks an EEG up to a human subject, shows them some visual illusions, and asks for some form of behavioral report from them, they are not measuring generic neural dynamics associated with the 10/n

“processing of visual illusions”. They are measuring the specific neural dynamics associated with the coupling of the visual information in the task environment to the behavioral report they asked for. This coupling has been established ontogenetically during the subject’s 11/n

learning history. This is why naturalistic tasks are so important. If you want to know about real behavior in real environments, you need naturalistic tasks in naturalistic settings, not artificial tasks in artificial environments. Ulric Neisser put it best in his book 12/n

"Cognition and Reality" (p. 6): "As the concept of information processing developed, the attempt to trace the flow of information through the 'system' (i.e. the mind) became a paramount goal of the new field... The proliferation of these ingenious and scientifically 13/n

respectable methods seemed at first - and still seems to many - a sign that the new cognitive psychology would succeed in avoiding whatever pitfall had claimed the old. Such optimism may have been premature. The study of information processing has momentum and prestige, but 14/n

it has not yet committed itself to any conception of human nature that could apply beyond the confines of the laboratory. And within the laboratory, its basic assumptions go little further than the computer model to which it owes its existence." 15/15