Verified Facts: Post 13 - The Machine That Learned Too Early (Perceptron)
Verified Facts: Post 13 - The Machine That Learned Too Early (Perceptron)
Date of fact-check: 2026-05-17
Post: /_posts/2026-04-09-The-machine-that-learned-too-early.md
VERIFIED CORRECT
Frank Rosenblatt biography
- Born July 11, 1928, New Rochelle, NY – confirmed Wikipedia + Cornell Faculty Memorial Statement (eCommons PDF, Stephen T. Emlen, Howard C. Howland, Richard D. O’Brien).
- Died July 11, 1971 (his 43rd birthday), drowning in Chesapeake Bay – Cornell Faculty Memorial confirms “Sunday afternoon, July 11, 1971, in a boating accident in Chesapeake Bay.” UMass page confirms “drowning on his 43rd birthday.”
- Graduated Bronx High School of Science, 1946 – confirmed Wikipedia.
- Cornell A.B. 1950, Ph.D. 1956 – confirmed Cornell Faculty Memorial.
- Cornell Aeronautical Laboratory in Buffalo, NY (research psychologist) – confirmed Cornell Faculty Memorial.
- Joined Cornell Ithaca campus 1959 as director of Cognitive Systems Research Program – Cornell Faculty Memorial.
- In 1971 was acting (co-)chair of Section of Neurobiology and Behavior – Cornell Faculty Memorial: “in 1971 he shared the acting chairmanship of the Section of Neurobiology and Behavior.”
- Active in McCarthy primary campaigns, Vietnam protests – Cornell Faculty Memorial: “He was active in the McCarthy primary campaigns in New Hampshire and California, in Democratic reform politics in New York State, and in a series of Vietnam protest activities in Washington.”
- Memory transfer through brain extracts research – confirmed Cornell Faculty Memorial: “transfer of learned behavior from trained to naive rats by the injection of brain extracts.”
- SETI / planet-detection research – Cornell Faculty Memorial confirms astronomical research. Nagy 2011 confirms NASA grant for Stellar Coherometer. Rosenblatt 1971 published “A two-color photometric method for detection of extra-solar planetary systems” in Icarus.
- Marvin Minsky one year ahead at Bronx Science – Cornell Chronicle confirms “same school as Marvin Minsky, one grade ahead.” Minsky born Aug 9 1927, Rosenblatt July 11 1928 – 13 months apart, supports one-year-ahead claim.
Project PARA / The Perceptron
- “Project PARA” = Perceiving and Recognizing Automaton, Cornell Aeronautical Lab, January 1957 – confirmed via Nagy 2011 slides (“The Perceptron, A Perceiving and Recognizing Automaton (Project Para), CAL Rept. No. 85-460-1, January 1957”); also Biblio.com listing of the actual report.
- Simulated on IBM 704 – confirmed Wikipedia + Cornell Chronicle.
- 1958 paper: “The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain,” Psychological Review 65(6) – confirmed via Nagy 2011.
Mark I Perceptron hardware
- 400 photocells in 20x20 grid (sensory/S-units) – confirmed Wikipedia Perceptron article + Nagy 2011 slides (“Mark I: 400 S-units”).
- 512 association units (A-units) – confirmed Wikipedia + Nagy 2011 (“512 A-units”).
- 8 response units (R-units) – confirmed Wikipedia + Nagy 2011 (“8 R-units”).
- Weights = potentiometers, learning via electric motors – confirmed Wikipedia Perceptron.
- Funded by Office of Naval Research – confirmed Wikipedia + Nagy 2011 (“Most of the ONR funding went to machine construction”).
- Now in Smithsonian National Museum of American History – confirmed Nagy 2011 + Cornell Chronicle.
- Publicly demonstrated June 23, 1960 – Wikipedia Perceptron. (Post says “constructed in 1960” – compatible.)
- 99.8% accuracy on shapes; 100% squares vs diamonds; 100% X vs E – Wikipedia Mark I Perceptron article confirms all three figures.
NYT 1958 press conference
- July 7, 1958: ONR press conference in Washington (UPI dispatch); story ran July 8 NYT page 25 – AITopics confirms dateline “Washington, July 7 (UPI)” + page 25 + July 8 publication.
- “Embryo of an electronic computer … able to walk, talk, see, write, reproduce itself and be conscious of its existence” – exact wording confirmed by AITopics archive.
- Rosenblatt’s quote: “the first machine which is capable of having an original idea” – confirmed Cornell Chronicle.
- Rosenblatt on assembly-line reproduction quote – attributed in NYT article (per AITopics summary).
Marvin Minsky / 1969 Perceptrons book
- Bronx Science class of 1945 (one year ahead of Rosenblatt) – consistent (Minsky born Aug 1927). Cornell Chronicle confirms “one grade ahead.” Wikipedia is silent on Minsky’s grad year but does not contradict.
- SNARC, 1951, Princeton, ~40 Hebb synapses, vacuum tubes, with Dean Edmonds – confirmed Wikipedia SNARC article. (Post says “40 neurons made from vacuum tubes” which is consistent though Wikipedia notes 300 vacuum tubes total, 6 per neuron, and the ~40 figure refers to Hebb synapses.)
- 1956 Dartmouth Conference – Minsky was one of four organizers – standard history; Wikipedia AI article. (Founding event of AI as a field.)
- 1969 Perceptrons: An Introduction to Computational Geometry, MIT Press – confirmed Wikipedia.
- XOR problem / linear separability arguments – confirmed Wikipedia Perceptrons (book) article.
- “Virtually nothing is known about the computational capabilities of this latter kind of machine. We believe that it can do little more than can a low order perceptron.” – quote IS authentic but Wikipedia attributes it to the 1971 Report of Project MAC by Minsky and Papert (paraphrased in the 1988 expanded edition of Perceptrons), NOT necessarily to p. 232 of the 1969 first edition. Footnote [^3] citing page 232 needs verification or rewording. See CORRECTIONS.
- Minsky’s “Necronomicon” comparison – confirmed Wikipedia Perceptrons (book) article.
McCulloch-Pitts 1943
- “A Logical Calculus of Ideas Immanent in Nervous Activity,” 1943 – confirmed Wikipedia Walter Pitts + Warren McCulloch articles. Published in Bulletin of Mathematical Biophysics.
- Warren McCulloch as neurophysiologist – confirmed Wikipedia: “an American neurophysiologist and cybernetician.”
- Walter Pitts as logician – confirmed Wikipedia: “an American logician.”
AI Winter / Lighthill / ALPAC / Mansfield
- ALPAC 1966 killed US machine translation funding – confirmed Wikipedia.
- Lighthill 1973 report – confirmed Wikipedia.
- Lighthill was Lucasian Professor of Mathematics, Cambridge (1969-1978/79), chair once held by Newton (1669-1702) – confirmed Wikipedia James Lighthill.
- “In no part of the field have the discoveries made so far produced the major impact that was then promised” – exact wording confirmed Wikipedia James Lighthill.
- British government cut AI funding “at all but two universities” – confirmed Wikipedia James Lighthill: “ended support for AI research in all but two universities.”
- Mansfield Amendment 1969 required DARPA to fund mission-oriented research – confirmed Wikipedia AI Winter.
- First AI winter ~1974-1980 – confirmed Wikipedia AI Winter.
Vindication
- John Hopfield 1982 paper, recurrent network model, physicist at Princeton – confirmed Wikipedia John Hopfield. Paper published April 1982 in PNAS 79(8): 2554-2558, “Neural networks and physical systems with emergent collective computational abilities.”
- Rumelhart, Hinton, Williams 1986 Nature paper “Learning representations by back-propagating errors” – confirmed Wikipedia Backpropagation.
- Paul Werbos 1974 (PhD thesis), Seppo Linnainmaa 1970 – confirmed Wikipedia Backpropagation.
- AlexNet 2012: Krizhevsky + Sutskever + Hinton – confirmed Wikipedia AlexNet.
- AlexNet 8 layers (5 conv + 3 fully connected) – confirmed Wikipedia AlexNet.
- Two Nvidia GPUs – confirmed Wikipedia AlexNet (“two Nvidia GTX 580 GPUs (3GB each)”).
- AlexNet top-5 error 15.3% – confirmed Wikipedia AlexNet.
- Submission September 30, 2012 – Wikipedia.
NEEDS QUALIFICATION OR MINOR CORRECTION
Telescope size
- Post: “12-inch Cassegrain telescope”
- UMass page: “Fecker 12-1/2” Cassegrain”
- Minor: should be 12.5 inch (or “12.5-inch”) not 12-inch.
AlexNet runner-up error rate
- Post: “26.2%”
- ImageNet official results: runner-up (ISI team) was 0.26172 = 26.172% which rounds to 26.2%. So the post is correct (rounded to one decimal).
- Wikipedia phrases the gap as “more than 10.8% above the runner-up” which is consistent with 26.1-26.2%. No correction needed.
“X and E with 100% accuracy after just 20 training examples”
- Post: “could classify the letters X and E with 100% accuracy after just 20 training examples”
- Wikipedia confirms 100% accuracy on X and E (with limited rotation) but does NOT mention “20 training examples” specifically. The figure may originate from a specific Mark I demo write-up but is not corroborated by the standard sources. Either provide a citation for the “20 examples” figure or remove it.
Mansfield Amendment year
- Post: “Mansfield Amendment of 1969” required DARPA-funded research to have military applications
- Wikipedia AI Winter says Mansfield Amendment shifted DARPA priorities to “mission-oriented direct research, rather than basic undirected research.” Whether the word “military” vs “mission-oriented” matches is a matter of nuance: the amendment did apply to defense department funding and required relevance to “military function or operation.” So “military applications” is fair shorthand but slightly tighter than the original wording.
Minsky-Papert hidden-layer quote attribution (footnote [^3])
- Post footnote [^3] cites “Minsky, M. & Papert, S. (1969). Perceptrons, MIT Press, p. 232.” for “Virtually nothing is known about the computational capabilities of this latter kind of machine.”
- Wikipedia Perceptrons (book) attributes this passage to the 1971 Report of Project MAC by Minsky and Papert (paraphrased in the 1988 expanded edition prologue).
- The quote may also appear in the 1969 book – it is widely cited from both – but page 232 of the 1969 first edition has not been verified by me. Action: either (a) verify the page number from a copy of the 1969 book, or (b) reword the citation to the 1988 expanded edition where Wikipedia confirms the quote appears.
PRIMARIES POSTED TO WAYBACK
Submitted via POST to https://web.archive.org/save/ on 2026-05-17 (all returned HTTP 200):
- https://en.wikipedia.org/wiki/Frank_Rosenblatt
- https://en.wikipedia.org/wiki/Perceptron
- https://news.cornell.edu/stories/2019/09/professors-perceptron-paved-way-ai-60-years-too-soon
- https://en.wikipedia.org/wiki/James_Lighthill
- https://en.wikipedia.org/wiki/AlexNet
SOURCES CONSULTED
- Wikipedia: Frank Rosenblatt, Perceptron, Mark I Perceptron, Perceptrons (book), Marvin Minsky, Stochastic Neural Analog Reinforcement Calculator, Walter Pitts, Warren Sturgis McCulloch, John Hopfield, Hopfield network, Backpropagation, AlexNet, James Lighthill, Lighthill Report, AI winter, ALPAC.
- Cornell Chronicle 2019: “Professor’s perceptron paved the way for AI – 60 years too soon.”
- Cornell eCommons Faculty Memorial Statement on Frank Rosenblatt (Emlen, Howland, O’Brien, ~1972).
- George Nagy 2011, “Frank Rosenblatt, my distinguished advisor,” Pace University talk slides (RPI archive).
- UMass Computational Phonology: “Remembering Frank Rosenblatt” (2021).
- AITopics archive: NYT July 8, 1958 article “New Navy Device Learns by Doing.”
- ImageNet ILSVRC 2012 official results table.