Abstract

We have created powerful AI systems but we struggle to understand them. The problem is that AI sits at the intersection of computer science and psychology, yet cognitivism, the dominant approach in psychology, is flawed and misleads us about intelligence. This essay argues that radical behaviorism offers the proper lens. While behaviorism has been remembered mostly as a caricature since the cognitivist revolution, the success of deep learning vindicates its view: complex behaviors emerge from simple associative mechanisms at scale. Training is conditioning, gradient descent implements the law of effect, prompts work as discriminative stimuli, outputs correspond to overt behaviors, and hidden layers to covert behaviors. Novelty generation and emergent capabilities, often at the center of confusion and debate, follow naturally from this perspective. The framework also carries practical implications: contingency engineering, rather than representation engineering, is the right research direction, since edited internal configurations will always prove weaker than organically formed associations.