Send message to Author

Short abstract:

This paper takes up Geoffrey Hinton's provocation that the future of large-scale AI is in "mortal" neuromorphic computers. By historicizing present fantasies of AGI with Carver Mead's 1980s research in neuromorphic computing, I offer a different history of the contemporary turn to brain-inspired AI.

Long abstract:

If the dream of posthumanism was immortality, the emerging computational paradigm is one of death. General purpose digital computers for large-scale AI are reaching their limit point in favor of new architectures like analog neuromorphic computation. Last year Geoffrey Hinton, one of leading figures of deep learning, argued that the future of AI was in its mortality. Hinton was referring to new research in neuromorphic computation in which hardware and software are one and the same. Data is stored as charges in memristors and floating gates rather than as code that can easily be transferred, such that when the computer “dies,” the data does too.

I take up Hinton’s provocation of neuromorphic “mortal computation” in which software and hardware are inseparable and designed together to chart a shift in the ontology of the chip centered on a rejection of the hylomorphic schema. I argue that this also constitutes a turn from software’s promise of immortal data, in the enduring ephemeral of the digital computer, to a mortal computation founded on temporary permanence. I do so by examining both Silicon Valley’s present fantasies of AGI and by turning to the late 1980s, when Carver Mead first began to develop neuromorphic computation at Caltech at the same time and in the same journals as deep learning and back propagation. By historicizing Hinton’s claims about the future of computing, I offer a different history of what I term AI’s mortal materialism and underscore neuromorphic computation’s beginnings in analog machine vision.