Behaviorist intelligence and the scaling problem

摘要

This paper argues that the strict computational behaviorist position for the modeling of intelligence does not scale to human-like problems and performance. This is accomplished by showing that the task of visual search can be viewed within the behaviorist framework and that the ability to search images (or any other sensory field) of the world to find stimuli on which to act is a necessary component of any behaving, intelligent agent. If targets are not explicitly known and used to help optimize search, the search problem is NP-hard. Knowledge of the target is of course explicitly forbidden in the strict interpretation of the published behaviorist dogma. Also, the paper summarizes the existing neurobiological and behavioral realities as they pertain to behaviorist claims. The conclusion is that there is very little support from biology for strict behaviorism. Strict adherence to the philosophy of the behaviorists means that efforts to demonstrate that the paradigm scales to humansize problems are certain to fail, as are attempts to evaluate it as a model of human intelligence. The strict position thus cannot be what the behaviorists really mean. It would benefit the research community if they could elucidate their terms, and provide theoretical arguments that support claims of scalability.