Each dot links to its primary source. The works below are the histories and essays on which the framing rests: a common origin in cybernetics, followed by alternating periods of convergence and divergence. The six-phase periodization used here is a synthesis; no single source states it in this form, and the final section lists those who would dispute it. The selection leans toward vision, the author’s own field; the same story could be told through memory, navigation, motor control, audition or language.
The shared origin
Papert, S. (1988). One AI or many? Dædalus 117(1):1–14. JSTOR · PDF Opens with “the standard version” of the history, told as a fairy tale — “two daughter sciences were born to the new science of cybernetics” — and then disputes it. The origin of the image drawn here, from a protagonist, with his own caveats.
Dreyfus, H. L. & Dreyfus, S. E. (1988). Making a mind versus modeling the brain: Artificial intelligence back at a branchpoint. Dædalus 117(1):15–43. JSTOR Two research programs founded together in the early 1950s; one starved of funding by 1970; the winners’ account is a “retrospective illusion”; 1988 reopens the fork.
Heims, S. J. (1991).The Cybernetics Group. MIT Press. doi The standard history of the Macy conferences: the “one community” phase was institutional, not incidental.
Abraham, T. H. (2016).Rebel Genius: Warren S. McCulloch’s Transdisciplinary Life in Science. MIT Press. doi The McCulloch biography; documents that the 1943 paper was largely ignored by neurophysiologists and taken up instead by von Neumann and automata theory.
Dupuy, J.-P. (2000).The Mechanization of the Mind: On the Origins of Cognitive Science. Princeton University Press (reissued MIT Press, 2009). Cognitive science as the partly disowned heir of cybernetics.
The split: Dartmouth, Perceptrons, and the winters
Kline, R. R. (2011). Cybernetics, automata studies, and the Dartmouth Conference on Artificial Intelligence. IEEE Annals of the History of Computing 33(4):5–16. doi Archival evidence that McCarthy chose the name “artificial intelligence” to distance the field from cybernetics and from Wiener. Book-length treatment: Kline, The Cybernetics Moment (Johns Hopkins University Press, 2015).
Olazaran, M. (1996). A sociological study of the official history of the perceptrons controversy. Social Studies of Science 26(3):611–659. doi The account in which Perceptrons ended neural-network research was written by the winners and rewritten when the networks returned. Essential context for 1969.
Haigh, T. (2023–2025). Historical Reflections, Communications of the ACM: Conjoined twins: AI and the invention of computer science, 66(6):33–37 doi · There was no ‘first AI winter’, 66(12):35–39 doi · How the AI boom went bust, 67(2):22–26 doi · Between the booms: AI in winter, 67(11):18–23 doi · Artificial intelligence then and now, 68(2):24–29 doi A historian’s periodization: “AI” as a label founded partly in reaction to cybernetic claims, with neural networks outside it until the 2010s.
Cardon, D., Cointet, J.-P. & Mazières, A. (2018). Neurons spike back: The invention of inductive machines and the artificial intelligence controversy. Réseaux 211:173–220. doi Co-citation analysis, 1943–2018, of the alternation between symbolic and connectionist paradigms, with deep learning “reviving the spirit” of cybernetics. The closest scholarly treatment of the cycle, although its axis is symbolic versus connectionist within AI, and it attributes the 2010s revival to data and computation rather than to biology.
Reconvergence and the present divergence
Sejnowski, T. J. (2018).The Deep Learning Revolution. MIT Press. doi The source of the phrase “deep learning revolution” for 2012, and a participant’s account of the 1980s.
Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks 61:85–117. doi; and (2022) Annotated history of modern AI and deep learning, arXiv:2212.11279 Argues that neural networks and deep learning “are conceptually closer to the old field of cybernetics than what was traditionally called AI.” Idiosyncratic on questions of priority.
Hassabis, D., Kumaran, D., Summerfield, C. & Botvinick, M. (2017). Neuroscience-inspired artificial intelligence. Neuron 95(2):245–258. doi “The fields of neuroscience and artificial intelligence (AI) have a long and intertwined history. In more recent times, however, communication and collaboration between the two fields has become less commonplace.”
Zador, A. et al. (2023). Catalyzing next-generation artificial intelligence through NeuroAI. Nature Communications 14:1597. doi · Doerig, A. et al. (2023). The neuroconnectionist research programme. Nature Reviews Neuroscience 24:431–450. doi The present divergence stated programmatically, and the case for renewed exchange.
Linsley, D., Feng, P. & Serre, T. (2026). Better artificial intelligence does not mean better models of biology. Trends in Cognitive Sciences 30:599–610. doi The case that the present period is a divergence: as vision models improve on benchmarks, their alignment with primate neural and behavioral data has stopped improving or declined.
Arbib, M. A. (2025). Artificial intelligence meets brain theory (again). Biological Cybernetics 119:16. doi A founder-generation figure places the 2024 NIH BRAIN NeuroAI workshop in a series of such meetings that began with cybernetics.
Objections to this framing
Neural-network research did not stop in the 1970s (Grossberg, Kohonen, Amari, Fukushima, Widrow); the “split” is in part the winners’ account (Olazaran 1996; Papert 1988). Haigh (2023) argues that there was no first AI winter. The two fields may never have been one: McCulloch and Pitts were ignored by neurophysiologists (Abraham 2016). Neuroscience’s contribution to modern AI has been described as thin and decades old (Ullman, Science 363:692–693, 2019; Zador et al. 2023), on which view 2012–2014 was neuroscience adopting AI’s tools rather than a reconvergence of ideas. The present divergence is a claim about vision models and neural rather than behavioral fit; the alignment of language models with brain recordings still improves with scale (Antonello, Vaidya & Huth, NeurIPS 2023).
←→ step · 0 all years · drag to pan · scroll to zoom