The old quarrel between rationalists and empiricists returns in AI as a design question: how much knowledge must be programmed into a machine by hand, and how much can it learn from data? Buckner (2024) proposes reading it not as a dichotomy but as a continuum: nearly every empiricist allowed a good deal of innate structure, and nearly every nativist a good deal of learning (2024, p. 12). What is at stake is the number and domain specificity of innate concepts and learning systems (2024, p. 18). At one end sits Fodor, for whom even concepts like CARBURETOR would be innate; at the other, radical behaviorism (2024, pp. 12, 14). AI has swung like a pendulum between these poles: from the rule-laden expert systems of GOFAI to deep learning, with minimal domain-specific programming (2024, p. 26), and on to the radical version of the bitter lesson (2024, p. 25).
Buckner's own position is a moderate empiricism. Locke and Hume never said the mind was a blank slate in its architecture, only in its content: they appealed to innate, general-purpose faculties such as memory, imagination, and reflection (2024, pp. 16–17). Buckner recasts Hume's Copy Principle as a Transformation Principle: simple concepts derive from systematic, domain-general transformations of sensory impressions (2024, p. 19). He names his program the "DoGMA" (Domain-General Modular Architecture): no innate ideas, but modules with general inductive biases that cooperate to extract abstract knowledge from experience; the name doubles as a reminder that this is a revisable empirical hypothesis (2024, pp. 30–32).
The concrete disputes turn on examples. Marcus read the translational invariance of convolution as innate knowledge; Buckner replies that it is a general bias, applicable to any data with the right mathematical structure (2024, pp. 20–21). He also charges some nativists with "anthropofabulation": comparing machines against an idealized picture of human cognition (2024, p. 29). The linguistic front is the poverty of the stimulus. What counts as "innate" in a system whose architecture an engineer chose?