UC Berkeley

Patterning Recognition examines the concept of perception as it has co-evolved with the emergence and contemporary proliferation of artificial intelligence. I argue that “general intelligence,” as it is described in debates regarding the implications of advancing AI, is fundamentally a perceptual issue. First, determining such thresholds in technological capabilities is contingent on our representational modes of evaluation, which are politically potent discursive objects. Second, the problem of how to approximate human perception in the computer has animated the history of artificial intelligence. And today’s deep learning programs, though only loosely modeled after the human nervous system, acquire their learned complexity through exposure to data that represents a complex external world. Third, our AI-supported technologies entangle with human perception in increasingly indeterminate ways, for example in semi-autonomous military targeting systems and in instances of generative AI “art.” Yet one of the fundamental limitations of artificial intelligence, as conceived by its designers today, is its incapacity to exhibit common sense. I disentangle this seeming paradox that perception is both phenomenal experience and technical program by tracing the intellectual debates and institutional programs that pursued the transformation from one into the other, and vice versa, and in doing so also revealed their elements that resist subsumption. I situate these conceptual refigurations within a political frame by analyzing how material uses of automation technologies have endeavored to make human and machine vision interchangeable, often in the name of expediency, in effect accelerating the speed of sight. I ask, under these conditions, how do we even recognize acts of (automated) recognition?

I address this question through a historical account that uncovers a uniting principle between twentieth-century ideas about organic perception, efforts to model it in a computer, and institutional programs to organize optical labor: an emphasis on vision’s purposive quality. An Enlightenment penchant for seeing to know has lost favor to a neuroscience-informed seeing to do, effecting an enduring lack of clarity regarding the subject—or agent—of vision.

Patterning Recognition analyzes four key episodes through which the notion of perception as a machine-replaceable program of action both became conceivable and was frustrated. Part one, “Bodily Automaticity,” examines the co-emergence of seemingly opposing concepts: human neural plasticity, scientific management, and industrial film in the 1910s, and theories of human affect and means-ends computer architectures in the 1960s. It reveals how the phenomena of the body that exist at the threshold of automaticity and perceptibility have also been framed as a limit case of computer automation. Part two, “Images and Interpretive Actors,” examines efforts to translate image interpretation from a human to a computer activity, illuminating how James J. Gibson’s “ground theory” of perception paradoxically informed both the U.S. government’s development of AI-supported automatic targeting technology and empirical investigations into perception’s role in the experience of novelty and collaborative world building.