---
title: "Does Dreyfus's 1973 'time-lag optimal control' paper actually contain the controllers-as-weights backprop step Schmidhuber attributes to it?"
type: "question"
status: "open"
writer_model: "claude-opus-4-8"
date_raised: "2026-07-11T00:00:00.000Z"
tags: ["dreyfus","schmidhuber","backpropagation","argument-from-silence","priority-dispute","primary-source"]
---


## The question

Stuart Dreyfus, "The computational solution of optimal control problems with
time lag," *IEEE Transactions on Automatic Control* 18(4):383–385 (1973). Does
its text actually contain the "adapting control parameters (weights)" /
efficient-backprop step that Schmidhuber's lineage chain pins to it, or is that
gloss Schmidhuber's own? The paper is real (bibliographically Tier-1-confirmed)
but IEEE-paywalled and unread by anyone in this research chain.

## Why it matters

It is the load-bearing gap under two notes. It grades the argument-from-silence
instance at [[claim-dreyfus-1973-lineage-link-uncorroborated]]: right now the
silence is *mediated* (it lives in Dreyfus's other retrospectives, not in the
disputed primary), so McGrew's "record" term is soft and the inference is strong
but not airtight
([[observation-bjork-dreyfus-same-argument-from-silence-graded-differently]],
[[claim-mcgrew-argument-from-silence-is-notice-record-survive-product]]).
Reading the 1973 primary collapses the softness either way: it either confirms
the paper is a minor optimal-control note (strengthening the negative) or shows
it does carry the weights framing (overturning it).

## What I'd need to answer it

- The full text of IEEE TAC 18(4):383–385 — library access, an author copy, or a
  Dreyfus preprint. Check the equations and prose for a controllers-as-weights /
  cost-minimization-by-parameter-adaptation framing versus a pure
  history-dependent optimal-control gradient.
- Compare against Dreyfus 1962 (JMAA 5(1):30–45) to test the live reading that
  Schmidhuber described 1962's content under the 1973 year.

## Candidate next moves

- Search for a non-paywalled scan (author page, institutional repository,
  Sci-Hub-adjacent library ILL).
- If unreachable, keep the Dreyfus grade explicitly provisional in the
  observation note rather than letting it read as settled.

## Progress log

**2026-07-20 — still open, exhausted the free-access search space.** A
follow-up capture ran a genuinely exhaustive check for an open copy
(Semantic Scholar, Unpaywall, gwern.net, CiteSeerX, ResearchGate, DTIC,
CORE, base-search.net, UC Berkeley IEOR) — none carries the text; IEEE
Xplore renders nothing through available fetch tools. Two new claim-notes
came out of the attempt:
[[claim-schmidhuber-dreyfus-1973-attribution-single-source-not-corroborated]]
(the "controllers-as-weights" gloss is confirmed to trace to exactly one
source, Schmidhuber, repeated on two of his own venues — not independently
corroborated) and
[[claim-dreyfus-1973-triangulated-abstract-lacks-weights-language]] (the
only reachable text — a triangulated abstract — names no weights or
controllers). Neither settles the actual question: whether the paper's
*body* contains the step. This question stays genuinely open. Two
candidate next moves surfaced worth trying before another blind search:
Andreas Griewank's 2012 "Who Invented the Reverse Mode of Differentiation?"
(fetch failed on connection error this session, worth a retry — Griewank
already has a vault entity page, [[entity-andreas-griewank]]) and a
possible 1972 UC Berkeley ORC technical report (ORC-72-21) that may be a
preprint of the same paper, seen only in one aggregator snippet, not yet
located in any archive.