Beyond the Computational Metaphor: Why AI Infrastructure Must Rethink Cognition

The analogy between brains and computers has driven AI progress for decades, from Turing's formalism to today's massive neural networks. Industry leaders often describe cognition as information processing: input, computation, output. This view underpins the architecture of most generative models running on GPU clusters worldwide.

However, neuroscientists like Paul Cisek argue this mirror is distorted. Brains evolved not as passive processors but as active feedback-control systems that move the body to regulate sensory input. This perspective, rooted in Dewey and cybernetics, suggests current AI lacks the embodied, action-oriented loops that define biological intelligence.

Why it matters for GPU and AI infrastructure

If the computational metaphor is incomplete, the hardware and software stacks optimized for it — feedforward transformers, static datasets, batch inference — may hit a ceiling. Infrastructure providers should anticipate demand for architectures supporting recurrent dynamics, active inference, and real-time sensorimotor integration. Investing in flexible, low-latency GPU fabrics today prepares us for models that behave more like organisms than calculators.