In 1962 David Hubel and Torsten Wiesel published “Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex” in The Journal of Physiology (volume 160, pages 106-154). The paper distilled years of experiments in which they recorded from single neurons in the primary visual cortex of cats while presenting carefully positioned spots and bars of light.
They described two kinds of cortical cells. “Simple” cells respond to a bright or dark edge at a specific orientation and position, with distinct excitatory and inhibitory regions in their receptive fields. “Complex” cells respond to an edge at a preferred orientation but tolerate movement of that edge across a range of positions - an early form of position invariance. The cortex, they showed, is laid out in an orderly functional architecture, with columns of cells sharing orientation preferences and others organized by which eye drives them.
The lasting significance for artificial intelligence is the picture this paints of vision as a hierarchy: light hits the retina, and progressively more abstract features - edges, then combinations of edges - are extracted at each stage. Kunihiko Fukushima’s neocognitron drew explicitly on Hubel and Wiesel’s simple and complex cells, and that lineage runs directly to the convolutional neural networks behind modern image recognition. The work earned Hubel and Wiesel a share of the 1981 Nobel Prize in Physiology or Medicine.