Herbert Simon named and dismissed perceptron research, citing Rosenblatt 1958, as a negative case in his 1983 chapter "Why Should Machines Learn?"

In a dedicated subsection ("2.4.3 Perceptrons") of his chapter "Why Should Machines Learn?", Herbert Simon addressed Rosenblatt's perceptron program directly, described its mechanism in his own terms — a system that computes features of a stimulus, discriminates among classes by linear additive functions of those features, and reweights on correct or incorrect output — and delivered a verdict:

"I have to conclude (and here I don't think I am in the minority) that this line of research didn't get anywhere. The discovery task was just so horrendous for those systems that they never learned anything that people didn't already know."

This is Simon's own direct, named judgment on perceptron-style learning as a research program, offered as one of several "classical" (≥20-year-old, by his 1983 clock) negative examples of AI learning research. His stated reason is narrow and specific: not that the mechanism was wrong, but that it failed at discovery — it never surfaced anything its designers didn't already know. This is a distinct claim from, and does not engage, Minsky and Papert's formal multilayer-sterility conjecture (claim-perceptrons-multilayer-sterile-was-conjecture); Simon's verdict rests on decades of subsequent research output, not on a structural proof.

A secondary source (Ben Recht's "arg min" blog, Tier 2) attributes this text to a plenary talk at a 1980 Carnegie Mellon workshop retrospectively treated as ICML's genesis, later reprinted as the 1983 chapter; that delivery date/venue detail is [unverified-quant — needs primary] and does not affect the dated, confirmed 1983 published text above.

This closes one half of the caveat in observation-simon-intuition-framework-diagnoses-perceptron-sterility-conjecture and the question at question-did-simon-comment-on-perceptron-controversy-or-connectionism. See also moc-backpropagation-origins.