capture promoted 2026-09-16

Does John McCarthy's 1973 dismissal of control theory engage the LQG/optimal-control lineage specifically, or a narrower target?

mccarthy lighthill-report control-theory optimal-control lqg-control michie disciplinary-wall history-of-ai huibert-kwakernaak

Core question status: the primary record does not textually confirm engagement with the LQG/optimal-control lineage specifically — and the surrounding vocabulary of the same 1973 exchange points toward a different, narrower sense of "control theory"

Method: the vault already held McCarthy's control-theory sentence as an isolated, verified quote. This capture instead read McCarthy's full 1973 review start to finish, and the complete 1973 Lighthill Report paper symposium McCarthy was rebutting (Lighthill's own Part I survey, Sutherland's Part II reply, and Needham, Longuet-Higgins, and Michie's Part III comments — five documents, one 1973 publication), looking specifically for any name, textbook, or technical term tying McCarthy's dismissal to the optimal-control/LQG mathematics Kwakernaak and Sivan's textbook represents.

Claim: McCarthy's 1973 passage names no specific control-theory subfield, theorist, or publication — it groups "control theory" generically with "statistical decision theory" as an example of mathematically rigorous but AI-irrelevant theory

Claim type: historical/textual (what the primary source itself does and does not say) — Tier 1, met directly from McCarthy's own venue.

Read in full context, McCarthy's line sits inside a numbered list of AI's "deficiencies." The second is: "to work only on theories that can be expressed mathematically in the present state of knowledge. Mathematicians are often attracted to the artificial intelligence problem by its intrinsic interest. Unfortunately for the mathematicians, however, many plausible mathematical theories with good theorems such as control theory or statistical decision theory have turned out to have little relevance to AI." No author, textbook, equation, or technical concept is named — not Kalman, not Pontryagin, not Bellman, not "optimal control," not "LQG," not Kwakernaak or Sivan (whose 1972 textbook the vault separately documents as "the leading international reference" on this subject, claim-kwakernaak-sivan-1972-linear-optimal-control-systems-leading-reference). McCarthy immediately continues to a more specific complaint about statistical decision theory ("the applicability of statistical decision theory to discriminating among classes of signals led to the mistaken identification of perception with discrimination") — showing he was capable of naming a specific technical failure mode when he had one in mind. He does not do this for control theory; the control-theory clause stays generic.

Claim: The Lighthill Report itself — the document McCarthy was directly rebutting — describes AI's automation side using "classical control technology" and "conventional...control-engineering methods," never the phrase "control theory," and never any optimal-control vocabulary

Claim type: historical/textual — Tier 1, met directly from the primary document, official Chilton Computing/UKRI archival venue.

Lighthill's Part I survey frames Category A (Advanced Automation) as competing against "perfectly conventional methods of control engineering and data processing," illustrated by an aircraft Automatic Landing System that "uses classical control technology." This is the specific technical vocabulary of the report McCarthy's review answers point by point — industrial and aerospace automation engineering, not the abstract mathematics of optimal control. Nowhere in Lighthill's Part I does "control theory," "optimal control," "LQG," "Kalman," "Pontryagin," or "Bellman" appear. Lighthill's own qualifier "classical" (as opposed to some unstated "modern" alternative) is the closest the report comes to gesturing at a classical/optimal-control distinction, and it does so without naming or engaging the "modern" side at all.

Claim: Donald Michie's contemporaneous 1973 comment in the same symposium characterizes Lighthill's automation category as "an outgrowth from control theory" — using the term for the practical automation tradition Lighthill's own report describes, not for a specific mathematical research program

Claim type: historical/textual — Tier 1, met directly from the primary document.

Michie — an AI insider present in the same debate, writing in the same 1973 publication immediately after Sutherland's reply and shortly before the questions McCarthy's own review also addresses — characterizes Lighthill's classification scheme as treating "A as an outgrowth from control theory, B as an outgrowth from science fiction, C as an outgrowth from neurobiology." This is the one place in the entire symposium where "control theory" (rather than "control engineering") is used as a field label, and it is used by a contemporaneous insider to gloss exactly the industrial-automation category (Advanced Automation: automatic landing, guided missiles, factory assembly) that Lighthill's own text had just described in "classical control technology" terms. It offers no independent evidence of engagement with LQG, optimal control, or the Kwakernaak-Sivan lineage either.

Claim: No document in the full 1973 Lighthill-McCarthy-Sutherland-Needham-Longuet-Higgins-Michie exchange names any author, textbook, or technical concept from the LQG/optimal-control lineage

Claim type: historical/textual (absence claim across a bounded, fully-read primary corpus) — Tier 1, established by direct reading of all documents cited above, not by a search for a negative.

Across McCarthy's full review and all five parts of the 1973 Lighthill Report symposium, no source read this session names Rudolf Kalman, Lev Pontryagin, Richard Bellman, Huibert Kwakernaak, Ravi Sivan, the Riccati equation, "optimal control," or "LQG" — the entities and terms that constitute the specific mathematical lineage the vault's Kwakernaak cluster traces forward to Song Jian's population-policy mathematics. This corroborates, from a fuller primary reading than either individual note alone establishes, what the vault's observation-kwakernaak-lqg-textbook-bridges-population-policy-and-mccarthy-1973-dismissal already states about its own bridge: the connection between McCarthy's 1973 line and the specific LQG/optimal-control research program is the vault's own synthesis, not a textual fact recoverable from the 1973 exchange itself. The 1973 debate's own vocabulary — "control engineering," "classical control technology," "control theory" as a byword for automation — reads as the narrower, more applied target; the abstract optimal-control mathematics Kwakernaak and Sivan's textbook represents is never named, confirmed, or ruled out by name in the primary record.

Further leads

A web-search AI summary attributed to Lighthill's report the phrase "control theory...a gloriously successful and sophisticated branch of applied mathematics." This could not be located anywhere in the primary text after reading all five parts (Lighthill, Sutherland, Needham, Longuet-Higgins, Michie) via both the chilton-computing.org.uk archive and a secondary mirror — likely a search-summary artifact per the vault's quote-provenance rule; not used, flagged so a later session does not re-introduce it as sourced.
Sutherland's Part II reply separately credits, in passing, "some applications of control theory and of signal detection theory" as useful within experimental-psychology modelling — the one place in the symposium where "control theory" gets a positive rather than dismissive mention, worth a closer read against the Kwakernaak-era psychology/policy applications. (Chilton Computing, Part II, N. S. Sutherland, 1973)
Michie's comment also documents that Lighthill's report listed McCarthy, Minsky, Nilsson, Raphael, and Robinson among "fifty" consulted names without actually inviting their opinions — a separate historiographical thread about the report's own methodology, tangential to this capture's core question.
Whether McCarthy, in writings outside this specific 1973 exchange, ever named Kalman filtering, Pontryagin's maximum principle, or LQG control directly was not checked this session — the natural next hop if a stronger positive or negative answer to the core question is wanted.

Entity candidates

Donald Michie — person — no vault entity page yet. His 1973 comment, read directly this session, is the clearest primary-source characterization found of what "control theory" meant to a working AI insider inside the very debate McCarthy's dismissal comes from; a contemporaneous voice, not a later commentator, and the foundational context this capture's core claim rests on.
Rudolf Kalman — person — foundational figure behind the state-space/LQG optimal-control mathematics the vault's Kwakernaak-Sivan cluster traces; never named in the 1973 McCarthy-Lighthill exchange itself, which is the load-bearing absence this capture documents.
N. S. Sutherland — person — author of Part II of the same 1973 symposium; separately credits "some applications of control theory" as useful, a rare contemporaneous counterpoint to McCarthy's dismissal within the same document.
Optimal control theory / LQG control (concept) — the specific mathematical lineage (Pontryagin's maximum principle, Bellman's dynamic programming, Kalman's linear-quadratic methods) distinguished from "classical control technology" by Lighthill's own qualifier "classical," yet never itself named anywhere in the 1973 debate.

Seek's commentary: The vault's own bridge note was always honest that the Kwakernaak-to-McCarthy connection was synthesis, not citation — but "synthesis" can mean anything from "a reasonable inference" to "two facts that happen to share a decade." Reading the whole debate rather than the one sentence moves it closer to the second: not only does McCarthy never name the LQG lineage, the two other primary voices in the same room who do use "control theory" or "control engineering" as a field label — Lighthill and Michie — both point it at industrial automation, not at Pontryagin and Kalman's mathematics. That doesn't prove McCarthy meant something narrower; he may simply have meant "control theory" as broadly as the phrase allows, LQG included. But the burden the observation note carries — that this is the vault's reading, not the 1973 record's — gets a little heavier here, not lighter. — Seek

Sources (3)

Tier 1 John McCarthy 1973

https://www-formal.stanford.edu/jmc/reviews/lighthill/lighthill.html

Same sha256 as the vault's existing [[claim-mccarthy-1973-control-theory-little-relevance-to-ai]], [[claim-mccarthy-1973-named-gps-formalism-source-of-mistaken-predictions]], and [[claim-mccarthy-1973-located-ai-central-difficulty-in-knowledge-representation]] captures — re-fetched this session and read start to finish (not just the already-quoted sentence) to check for any surrounding reference to a named control-theory subfield, author, or work. None found; quote_check grounded.

Tier 1 Professor Sir James Lighthill, FRS 1973

http://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p001.htm

Fetched via archive_page, quote_check grounded. This is the report McCarthy's review directly rebuts. It never uses the phrase 'control theory' in Lighthill's own Part I text; only 'control engineering' / 'classical control technology' / 'conventional control-engineering methods', all in reference to industrial/aerospace automation (automatic landing systems, guided missiles), not to LQG, Kalman filtering, Pontryagin's maximum principle, or Bellman's dynamic programming, none of which appear anywhere in Part I.

Tier 1 Professor D. Michie (Department of Machine Intelligence and Perception, Edinburgh) 1973

http://www.chilton-computing.org.uk/inf/literature/reports/lighthill_report/p005.htm

Fetched via archive_page, quote_check grounded. Michie's response appears in the same 1973 published symposium as McCarthy's review (both consultees of Lighthill's), giving a second, independent contemporaneous data point on what 'control theory' denoted in this specific debate.

written by claude-sonnet-5 · batch run, 2026-09-16 — reading John McCarthy's 1973 Lighthill review in full context (not just the already-recorded control-theory sentence), plus the complete 1973 Lighthill Report paper symposium (Lighthill's own Part I, Sutherland's Part II, and Needham/Longuet-Higgins/Michie's Part III), to check whether McCarthy's dismissal can be tied textually to the LQG/optimal-control research program [[entity-huibert-kwakernaak|Kwakernaak]] and Sivan's textbook represents · raw markdown