Can the whole brain be simpler than its "parts"?
share
This is the first in a series of connected papers discussing the problem of a dynamically reconfigurable universal learning neurocomputer that could serve as a computational model for the whole human brain. The whole series is entitled "The Brain Zero Project. My Brain as a Dynamically Reconfigurable Universal Learning Neurocomputer." (For more information visit the website www.brain0.com.) This introductory paper is concerned with general methodology. Its main goal is to explain why it is critically important for both neural modeling and cognitive modeling to pay much attention to the basic requirements of the whole brain as a complex computing system. The author argues that it can be easier to develop an adequate computational model for the whole "unprogrammed" (untrained) human brain than to find adequate formal representations of some nontrivial parts of brain's performance. (In the same way as, for example, it is easier to describe the behavior of a complex analytical function than the behavior of its real and/or imaginary part.) The "curse of dimensionality" that plagues purely phenomenological ("brainless") cognitive theories is a natural penalty for an attempt to represent insufficiently large parts of brain's performance in a state space of insufficiently high dimensionality. A "partial" modeler encounters "Catch 22." An attempt to simplify a cognitive problem by artificially reducing its dimensionality makes the problem more difficult.
READ FULL TEXT