On the computational utility of consciousness

We propose a computational framework for understanding and modeling human consciousness. This framework integrates many existing theoretical perspectives, yet is sufficiently concrete to allow simulation experiments. We do not attempt to explain qualia (subjective experience and feelings), but instead ask what differences exist within the cognitive information processing system when a person is conscious of some information versus when that information is unconscious. The central idea we explore is that the contents of consciousness correspond to temporally stable states in an interconnected network of specialized computational modules. Each module is an associative memory that operates in two stages: (1) a fast, essentially feedforward, input-output mapping that attempts to achieve an appropriate response to a given input, and (2) a slower relaxation search that is concerned with achieving semantically well-formed states. It is the stable attractors of the relaxation search that reach conscious awareness. To illustrate the operation of a module, we model performance on a simple arithmetic task and show that the sequence of stable states in our model corresponds roughly to the conscious mental states people experience when performing this task. What might be the computational utility of stable states within the cognitive architecture? Our simulations show that periodically settling to stable states improves performance by cleaning up inaccuracies and noise, forcing decisions, and helping to keep the system on track toward a solution.

Retrieve (gzipped postscript) (pdf)