A History of AI Since the 1950s

INTS1301 Technology and Society: From Plato to NATO — Week 7

Brian Ballsun-Stanton

Macquarie University

2026-04-21

A conversation with a book

WHEN WAS THIS WRITTEN?

Simmons’ dream was that one could have “a conversation with a book;” the computer would read the book, and then the user could have a conversation with the computer, asking questions to be answered from the computer’s understanding of the book.

“The objective of this project is to develop a research methodology and a vehicle for the design and construction of a general purpose computerized system for synthesizing complex human cognitive functions. The initial vehicle, proto-synthex, will be an elementary language-processing device which reads simple printed material and answers simple questions phrased in elementary English.”

1961

Robert Simmons, SDC “Synthex” grant proposal

Accomplishing this “dream” would turn out to be as hard as AI itself.

— Nilsson (2010)

AI is a marketing term

WHEN IT WORKS, WE STOP CALLING IT AI

  • Technology you don’t notice because it’s working
    • The motion sensor in this room is three if-statements in a trench coat. That’s an expert system. Nobody calls it AI.
    • Your phone’s autocomplete, spam filter, face unlock: all AI from previous hype cycles. You don’t see any of it.
    • Google Maps routes you. Netflix picks what you watch. Siri sets your alarms. Background AI, every day.
  • Technology you notice, because it broke, because it’s new, because it’s threatening your job
    • ChatGPT (2022). First time most people talked to an AI. Still new. You still see it.
    • Clippy (1997). Mocked because of the googly eyes. The machine learning underneath runs your autocomplete now. You don’t see that part.
    • Self-driving cars. You still see them because they still go wrong.
  • AI is a synonym for technology: the stuff marketing wants you to notice

Your spam filter uses AI. ChatGPT uses AI. Why do we only call one of them AI?

Before AI was called AI

HOW A FIELD GETS ITS NAME

  • The military problem that became a field
    • WWII: how do you hit a fighter plane? Anti-aircraft predictors, gun directors, fire-control computers
    • Wiener’s Cybernetics (1948): one framework (feedback and control) that covers thermostats, gun directors, and the way your body regulates its own temperature
  • 1948 Hixon Symposium: two roads fork
    • Caltech, Cerebral Mechanisms in Behavior: cognitive science and AI’s common root
    • Simulate biology (neural nets) vs simulate behaviour (logic, symbols)
    • The logic-and-symbols side is later renamed GOFAI (good old-fashioned AI), Haugeland’s 1985 retronym, coined after the rival camp emerges.
  • 1956: the name “AI” exists because McCarthy writes the grant
    • Wiener is not invited to Dartmouth: the man, not just the framework
    • McCarthy’s own words: wouldn’t “accept Norbert Wiener as a guru or argue with him”
    • Simon and Newell preferred “complex information processing”; they weren’t writing the grant
    • Newell: “Like all names of scientific fields, it will grow to become exactly what its field comes to mean.” McCarthy held the pen

ELIZA, 1966

PEOPLE BELIEVED IT. HIS OWN SECRETARY BELIEVED IT.

  • ELIZA: symbolic AI’s first public landmark
    • Explicit rules for manipulating language: rules the programmer writes down
    • “I hate my mother” → “Why do you hate your mother?”
    • First chatbot most people ever saw. Symbolic AI reaching a mass audience
  • Weizenbaum’s secretary, who had watched him build it:
    • “She knew she was talking to a machine. Yet after a few sentences she turned to me and said ‘Would you mind leaving the room, please?’” — Weizenbaum (1967)
    • Parasocial relationships with computers are not new. 1966.
  • Weizenbaum warned the field from 1966. McCarthy dismissed him in print
    • Computer Power and Human Reason (1976): “there are certain tasks which computers ought not be made to do, independent of whether computers can be made to do them”
    • McCarthy’s review: “no argument is offered that might be answered.” Field-founder dismissing field-critic, first of a long pattern

Have you ever said “please” or “thank you” to an AI? Why?

Teaching computers by showing

THE OTHER CAMP: LEARNING FROM DATA, NOT RULES

  • Rosenblatt’s Perceptron (1958): the first neural net you could actually run
    • McCulloch & Pitts designed a neural net in 1943 as mathematical theory. Rosenblatt builds a working one 15 years later
    • Show thousands of labelled examples. The machine adjusts its own internal weights until it gets the answers right. The rules emerge from the data
    • This is where the word “training” comes from
    • Once trained, the weights freeze. The machine then applies what it learned, read-only. Training is not doing
    • Perceptron descendants run the high-voltage power grid today, catching spikes and transients before the lights flicker
  • Nilsson (2010, p. 77) names the split in retrospect
    • “Symbolic” AI (GOFAI): rules, logic, theorem-proving. ELIZA lives here
    • “Nonsymbolic” AI: neural nets, statistics, pattern recognition. Rosenblatt lives here
    • Both are called “AI” in the 1960s. Both will fight for 70 years. Neither wins permanently

ELIZA has rules a human wrote. The Perceptron has weights the machine found. If both count as AI, what is AI?

Where is the intelligence right now?

END OF THE 1960s. THREE THREADS.

  • Thread 1: notice / don’t-notice
    • Technology that works reliably fades into background infrastructure. You don’t notice your spam filter, you don’t think of the motion sensor as AI, and autocomplete is just how typing works now. What we still call AI is whatever we haven’t yet figured out
  • Thread 2: Amara’s Law
    • We overestimate what a new technology will do in the short run and underestimate what it does in the long run. Weizenbaum’s 1976 warning about parasocial attachment to machines arrived on time in 2022, forty-six years late
  • Thread 3: symbolic vs nonsymbolic
    • ELIZA followed rules a human wrote down explicitly. Rosenblatt’s Perceptron adjusted its own weights from labelled examples. Both were called AI in 1966; both camps are still with us; neither has ever permanently won the label

SNL “New Shimmer” (1976): floor wax AND a dessert topping

Where is the AI in your life right now? Which of these three threads is it on?

Reports that kill fields

SPECIFIC DOCUMENTS. SPECIFIC CONSEQUENCES.

“Work of excellence by talented young people was stigmatised as bad science … killed in mid-trajectory. … To speak plainly, it was an outrage.”

— Donald Michie, on Edinburgh AI after Lighthill (1972)

  • ALPAC, 1966: machine translation dies
    • A US National Research Council committee chaired by John Pierce reviewed Cold-War machine translation funding
    • The verdict: quality was poor, cost was high, and human translators outperformed the machines meant to replace them
    • Funding was cut across the US. Machine translation went quiet for a decade, 1967 to 1976
  • Lighthill, 1973: UK AI collapses
    • “In no part of the field have the discoveries made so far produced the major impact that was then promised” (Lighthill (1972), commissioned by the UK Science Research Council)
    • The critique: AI demos worked on toy problems but could not scale, because the combinatorial explosion punished every attempt to generalise

Who could write the 2026 Lighthill report? What field would it kill?

Expert systems rise

FEIGENBAUM’S ANSWER TO LIGHTHILL

“Our agents must be knowledge-rich, even if they are methods-poor.”

— Feigenbaum (1977)

  • DENDRAL and the knowledge principle
    • DENDRAL (Stanford, 1965 onwards) encoded a chemist’s judgement about molecular structure into explicit rules: the first expert system
    • Feigenbaum’s knowledge principle: intelligence comes from encoded domain expertise, not from general-purpose inference methods
  • MYCIN and XCON: the commercial arc
    • MYCIN (Stanford, 1970s) diagnosed bacterial infections better than junior doctors, but was never deployed: liability and trust, not capability, blocked it
    • XCON (Digital Equipment, 1980) configured VAX computer orders. By 1989 the system had 17,500 rules and returned over $40 million per year to DEC
    • Expert systems spread through the Fortune 500 between 1975 and 1985: boom time for narrow-scope AI

The autopilot in every modern aeroplane can land the plane without the pilot touching anything. Is that AI?

The Navy shelved it

DAYS ARE GOOD ENOUGH

“The Navy’s decision to mothball FRESH was because there was no compelling reason to keep it. It duplicated the expert judgement of Fleet planners, in a matter of hours rather than days. But the planners were not looking to retire, and in Naval warfare, days are good enough.”

— Walter Saunders, FCCBMP program manager (Nilsson, 2010, p. 291)

  • What FRESH did
    • FRESH was an expert system for Fleet battle management, delivered to CINCPACFLT in 1990
    • Built on Symbolics Lisp machines using commercial expert-system shells, squarely inside Feigenbaum’s knowledge-rich paradigm
  • Why the Navy shelved it
    • When DARPA funding ended, the Navy decided not to continue
    • Saunders, in hindsight, called the FCCBMP goals “an overreach for the state of the art in the 1980s”
    • Congress pressured DARPA to divert the funding to anti-submarine warfare. Geopolitics trumped technical merit

If your employer deployed an AI that did your job in hours instead of days, would you help train it?

Where is the intelligence right now?

END OF THE 1980s. SAME THREE THREADS.

  • Thread 1: notice / don’t-notice
    • What was new in 1969 is now background. Expert systems themselves are folding into plain software: XCON is DEC’s parts-ordering system; MYCIN-shaped tools sit inside hospital decision support; autopilot flies the plane. Nobody calls any of this AI anymore
  • Thread 2: Amara’s Law
    • The 1980s show the short-run overestimate: Feigenbaum’s “knowledge-rich” AI, Nilsson’s 1983 “predominant scientific endeavour”, XCON at $40M/yr, the Japanese Fifth Generation. Bust: 1987. The long-run half of Amara’s Law is invisible from inside the decade. Only from 2026 can we see expert-system descendants running tax software, airline booking, compliance engines. The tech dissolved into “software”. The name didn’t survive
  • Thread 3: symbolic vs nonsymbolic
    • Symbolic had the 80s: MYCIN, XCON, Japan Fifth Generation, Strategic Computing’s billions. Nonsymbolic stayed quiet at IBM Raleigh / Yorktown, at SRI, in pattern-recognition conferences. The pendulum hasn’t swung yet, but the weight has shifted

Jordan Harrod: “Does AI Shift Power?”

Where is the intelligence right now, at the end of 1989? Which thread carried the load?

The label goes radioactive

THE WORK CONTINUED, UNDER DIFFERENT NAMES.

“In spite of all the commercial hustle and bustle around AI these days, there’s a mood that I’m sure many of you are familiar with of deep unease among AI researchers who have been around more than the last four years or so.”

— Drew McDermott, AAAI “Dark Ages of AI” panel, 1984 (McDermott et al., 1985)

  • The winter lands (1984–1989)
    • McDermott’s 1984 panel is the warning; the bust follows. DARPA cuts funding. Lisp-machine companies collapse. Japan’s Fifth Generation fades. Grad students are told to strip “AI” from their CVs
    • AAAI membership peaks in 1985 and falls through the 1990s. The word “AI” becomes something you apologise for, not something you brand a product with
  • Architects’ names and venues: not “AI”
    • Judea Pearl publishes Probabilistic Reasoning in Intelligent Systems, a Bayesian-networks book in a CS department, not an AI department. Different venue, different vocabulary (Pearl, 1988)
    • At IBM Yorktown, Brown and Mercer translate languages using statistics. Breiman’s statisticians publish at NIPS and Machine Learning Journal. Working names: statistical NLP, corpus-based linguistics, machine learning

If a field’s name becomes a liability, what happens to the people doing the work?

Revival, not revolution

BOTH SIDES CLAIM THE TITLE OF AI.

  • The architects called it revival
    • “A resurgence of 1950s-style empirical and statistical methods.” Not a new paradigm, a revival of Shannon, Firth, Harris (Church & Mercer, 1993)
    • Algorithmic modelling had been practised “under other names for decades.” A paradigm that narrates itself as restoration doesn’t coin a revolution label (Breiman, 2001)
  • GOFAI: the retronym is the thesis in miniature
    • Haugeland (1985) coins Good Old-Fashioned AI, GOFAI. The symbolic camp needed a name only once a rival emerged
    • Retronyms appear only when a contrast forces them: acoustic guitar once electric exists; analog watch once digital exists; GOFAI once statistical methods claim AI

If neither side called itself “the winning side,” how do we know which one won?

Deep learning resurrects

THE NONSYMBOLIC CAMP NAMES ITSELF.

  • Hinton 2006: the camp names itself
    • Hinton’s Toronto group publishes “A fast learning algorithm for deep belief nets” (Hinton et al., 2006). The modern term “deep learning” lands for multi-layer neural nets
    • Church & Mercer called their shift a revival; Hinton’s camp chose a banner. Forty years of exile end with a word

Sutton’s warning: compute over algorithms. Forever true? Go look at Harness Engineering

ELIZA returns

1966 → 2022: 56 YEARS LATE.

  • The Transformer, and the five-year gap
    • “Attention Is All You Need” (Vaswani et al. (2017)). Zero uses of the word “intelligence” in the paper. The architecture that powers 2022 LLMs has no claim on “AI” at its birth
    • OpenAI execs call ChatGPT (November 2022) a “low-key research preview” (The Verge, 2023). 100M users in two months. Paper ≠ deployment
  • ELIZA at planetary scale
    • Weizenbaum 1966: the secretary knew DOCTOR was code. Still asked him to leave the room. Amara’s Law set the clock: 56 years from warning to planetary arrival
    • 2026: the chatbot sits in bedrooms, classrooms, boardrooms. Parasocial attachment at scales Weizenbaum could not have imagined

When does a warning actually arrive: when it’s given, or when we’re ready to hear it?

The bill arrives

THE WARNING LANDS.

  • The critique, and the field closing ranks
    • “On the Dangers of Stochastic Parrots” names the mechanism: statistical outputs, no understanding; environmental cost; bias baked in (Bender et al., 2021)
    • Google fires Gebru December 2020, Mitchell February 2021. Field-founder dismisses field-critic: McCarthy 1976 → Google 2020
  • Sutton’s closer
    • “General methods that leverage computation win, by a large margin” (Sutton (2019)). The decision becomes infrastructure politics: who owns the compute, who pays the externalities

Simmons, 1961: “a conversation with a book.” Today: yes, and the book talks back, at planetary scale.

Who owns the book?

Notes and References

This deck was built with the help of Claude Opus 4.7. Transcripts and tooling available on request. Rendered with Quarto. It is licensed with a Creative Commons 4.0 International – With Atribution License.

Amodei, D., & Hernandez, D. (2018). AI and compute. OpenAI blog post. https://openai.com/index/ai-and-compute/
BBC Travel. (2024). Air canada chatbot misinformation: What travellers should know. https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? 🦜. Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, FAccT ’21, 610–623. https://doi.org/10.1145/3442188.3445922
Breiman, L. (2001). Statistical modeling: The two cultures (with comments and a rejoinder by the author). Statistical Science, 16(3), 199–231. https://doi.org/10.1214/ss/1009213726
British Columbia Civil Resolution Tribunal. (2024). Moffatt v air canada.
Church, K. W., & Mercer, R. L. (1993). Introduction to the special issue on computational linguistics using large corpora. Computational Linguistics, 19(1), 1–24. https://aclanthology.org/J93-1001/
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., & Fei-Fei, L. (2009). ImageNet: A large-scale hierarchical image database. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 248–255. https://doi.org/10.1109/cvpr.2009.5206848
Dick, S. (2019). Artificial intelligence. Harvard Data Science Review, 1(1). https://doi.org/10.1162/99608f92.92fe150c
Dreyfus, H. L. (1972). What computers Can’t do: A critique of artificial reason. Harper & Row.
Dreyfus, H. L., & Dreyfus, S. E. (1986). Mind over machine: The power of human intuition and expertise in the era of the computer. Free Press.
Feigenbaum, E. A. (1977). The art of artificial intelligence: Themes and case studies of knowledge engineering. Proceedings of the 5th International Joint Conference on Artificial Intelligence (IJCAI-77), 1014–1029.
Grace, K. (2026). AI risk was not invented by AI CEOs to hype their companies. Substack (world spirit sock stack). https://worldspiritsockpuppet.substack.com/p/ai-risk-was-not-invented-by-ai-ceos
Halevy, A., Norvig, P., & Pereira, F. (2009). The unreasonable effectiveness of data. IEEE Intelligent Systems, 24(2), 8–12. https://doi.org/10.1109/MIS.2009.36
Haugeland, J. (1985). Artificial intelligence: The very idea. MIT Press.
Hinton, G. E., Osindero, S., & Teh, Y.-W. (2006). A fast learning algorithm for deep belief nets. Neural Computation, 18(7), 1527–1554. https://doi.org/10.1162/neco.2006.18.7.1527
Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science, 313(5786), 504–507. https://doi.org/10.1126/science.1127647
Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
Hutchins, W. J. (1995). Machine translation: A brief history. In E. F. K. Koerner & R. E. Asher (Eds.), Concise history of the language sciences: From the sumerians to the cognitivists (pp. 431–445). Pergamon Press. http://www.hutchinsweb.me.uk/ConcHistoryLangSci-1995.pdf
Kanal, L. N. (Ed.). (1968). Pattern recognition: Proceedings of the IEEE workshop on pattern recognition, held at dorado, puerto rico. Thompson Book Co.
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems 25, 1097–1105.
LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., & Jackel, L. D. (1989). Backpropagation applied to handwritten zip code recognition. Neural Computation, 1(4), 541–551. https://doi.org/10.1162/neco.1989.1.4.541
LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278–2324. https://doi.org/10.1109/5.726791
Lighthill, J. (1972). Artificial intelligence: A general survey. Science Research Council of Great Britain.
Martin, W. L. (1986). An assessment of artificial intelligence and expert systems technology for application to the management of cockpit systems (AAMRL-TR-86-040). Harry G. Armstrong Aerospace Medical Research Laboratory, Air Force Systems Command. https://apps.dtic.mil/sti/citations/ADA175456
McCarthy, J. (1976). An unreasonable book [review of weizenbaum, computer power and human reason]. ACM SIGART Bulletin, (58), 5–6. https://doi.org/10.1145/1045264.1045265
McCarthy, J. (2000). Review of “the question of artificial intelligence” edited by brian bloomfield. https://www-formal.stanford.edu/jmc/reviews/bloomfield/bloomfield.html
McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (1955). A proposal for the Dartmouth summer research project on artificial intelligence. Dartmouth College. http://jmc.stanford.edu/articles/dartmouth/dartmouth.pdf
McCorduck, P. (1979). Machines who think: A personal inquiry into the history and prospects of artificial intelligence. W. H. Freeman; Co.
McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The Bulletin of Mathematical Biophysics, 5(4), 115–133. https://doi.org/10.1007/BF02478259
McDermott, D., Waldrop, M. M., Chandrasekaran, B., McDermott, J., & Schank, R. (1985). The dark ages of AI: A panel discussion at AAAI-84. AI Magazine, 6(3), 122–134. https://ojs.aaai.org/aimagazine/index.php/aimagazine/article/view/494
Michie, D. (1982). Machine intelligence and related topics: An information scientist’s weekend book. Gordon; Breach Science Publishers.
Minsky, M., & Papert, S. (1969). Perceptrons: An introduction to computational geometry. MIT Press.
Newell, A. (1980). AAAI president’s message. AI Magazine, 1(1), 1–4. https://doi.org/10.1609/aimag.v1i1.84
Nilsson, N. J. (2010). The quest for artificial intelligence: A history of ideas and achievements. Cambridge University Press.
OpenAI. (2026). Enterprises power agentic workflows in Cloudflare agent cloud with OpenAI. https://openai.com/index/cloudflare-openai-agent-cloud/
Pearl, J. (1988). Probabilistic reasoning in intelligent systems: Networks of plausible inference. Morgan Kaufmann.
Pierce, J. R., & Automatic Language Processing Advisory Committee. (1966). Language and machines: Computers in translation and linguistics. National Academy of Sciences / National Research Council.
Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning representations by back-propagating errors. Nature, 323(6088), 533–536. https://doi.org/10.1038/323533a0
Russell, S. J., & Norvig, P. (2021). Artificial intelligence: A modern approach (4th ed.). Pearson.
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., Driessche, G. van den, Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., Dieleman, S., Grewe, D., Nham, J., Kalchbrenner, N., Sutskever, I., Lillicrap, T., Leach, M., Kavukcuoglu, K., Graepel, T., & Hassabis, D. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484–489. https://doi.org/10.1038/nature16961
Sutton, R. (2019). The bitter lesson. http://www.incompleteideas.net/IncIdeas/BitterLesson.html
The Verge. (2023). OpenAI execs dubbed ChatGPT a low-key research preview. https://www.theverge.com/2023/12/5/23989871/openai-execs-dubbed-chatgpt-a-low-key-research-preview
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems 30, 5998–6008. https://arxiv.org/abs/1706.03762
Weizenbaum, J. (1966). ELIZA — a computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36–45. https://doi.org/10.1145/365153.365168
Weizenbaum, J. (1967). Contextual understanding by computers. Communications of the ACM, 10(8), 474–480. https://doi.org/10.1145/363534.363545
Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. W. H. Freeman.
Wiener, N. (1948). Cybernetics: Or control and communication in the animal and the machine. MIT Press.

Coda: Carnival barkers

EXPERT SYSTEMS THEN, AGENTS NOW

“We’ve built a better brain … Expert systems reduce waiting time, staffing requirements and bottlenecks caused by the limited availability of experts. Also, expert systems don’t get sick, resign, or take early retirement.”

TIMM brochure, AAAI trade show, 1984 (via George Johnson, (Nilsson, 2010, p. 272))

“Enterprises can now deploy agents … to perform real work. For example, companies can use it with OpenAI to deploy agents that automatically handle tasks like responding to customers, updating systems, and generating reports, all within a secure, production-ready environment.”

Cloudflare / OpenAI press release 13 April 2026

One is from 1984. One is from 2026. Which is which, and how can you tell?