PERSON

Hermann von Helmholtz

The nineteenth-century Prussian polymath whose doctrine that perception is unconscious inference became, a century after his death, the founding idea of the Helmholtz machine and the deep lineage of every modern generative AI system—a man whose two greatest achievements, the conservation of energy and the theory of perception-as-inference, turn out to have been one achievement all along.

Hermann von Helmholtz is the rare thinker whose relevance to artificial intelligence is not interpretive projection but a documented fact of the field's own naming. The most influential early architecture of unsupervised learning was called the Helmholtz machine by the people who built it—Peter Dayan, Geoffrey Hinton, Radford Neal, and Richard Zemel—because the thing they had built was, in its bones, Helmholtz's 1860s idea made mechanical: a system that perceives by guessing what unseen world would have produced the data it receives, and corrects the guess against the evidence. From that architecture runs a direct line to variational autoencoders, diffusion models, and the generative systems that now produce text, images, and code. Helmholtz also gave thermodynamics the free energy that Karl Friston later borrowed to name the principle by which brains—and Helmholtz machines—learn. He measured the speed of

Keep reading with YOU ON AI