Schmidhuber's 'Annotated History of Modern AI and Deep Learning' dates the field's origin to the 1676 chain rule and circa-1800 neural nets — reading the present category back onto the deep past

Jürgen Schmidhuber's Annotated History of Modern AI and Deep Learning (arXiv:2212.11279) is the vault's most-cited single source for correcting misattributed credit in backpropagation's genealogy (claim-credit-assignment-spans-sociology-and-backprop, claim-ivakhnenko-gmdh-first-deep-characterization, claim-dreyfus-1973-lineage-link-uncorroborated). The same document is itself an instance of the historiographical fallacy it fights. Its timeline periodizes "modern AI and deep learning" back to "the chain rule (1676), the first NNs (circa 1800), the first practical AI (1914)" — projecting a 2010s-vintage category, "deep learning," onto seventeenth-, nineteenth-, and early-twentieth-century mathematics and devices that were not built, framed, or understood by their creators as steps toward that category. This is Whig history in Butterfield's exact sense (claim-butterfield-1931-coined-whig-history-as-present-referenced-past): studying the mathematical past with reference to a present label.

The self-referential twist is the note's value. Schmidhuber's project runs the field's most sustained anti-Whig campaign at the level of individual credit — insisting Linnainmaa, Ivakhnenko, and others get named rather than folded into famous successors — while being simultaneously Whiggish at the level of periodization, reading a unified "deep learning" lineage into work that long predates the concept. Correcting who gets credit and correcting whether the category applies at all are different axes, and this document does the first rigorously while doing the second casually. See moc-backpropagation-origins.