A Different Kind of History of AI

Manuel Wörsdörfer’s A Different Kind of History of AI is a historical and philosophical argument for evaluating AI development alongside its funding, institutions, and ethical consequences. The January 2026 author manuscript asks what earlier debates can contribute to present research and governance. Its long original title quotes the proposal for the Dartmouth summer research project.^[1]

How the argument works

The paper combines writings by AI pioneers and critics with later historical and ethical scholarship. Thirteen thematic lessons, arranged largely chronologically, each move from a proposed lesson through examples to a takeaway. This is a selective narrative and normative synthesis, rather than a systematic historical survey, an experiment, or a quantitative test of what causes AI progress.^[1]

Three strands show its distinctive emphasis. Section 2.6 places public research funding and institutional infrastructure beside algorithms, computing hardware, data, and talent. It uses DARPA-supported laboratories, Shakey, and expert-system programs to challenge an account centered on private enterprise and deregulation. Section 2.7 compares ambitious early promises with later disappointment and funding contractions. Those examples support Wörsdörfer’s warning about expectations; they do not establish a general law that hype causes research downturns.^[1]

Section 2.12 connects questions of machine intelligence to responsibility for automation and delegated judgment. Its discussion brings Turing, Wiener, Weizenbaum, and other thinkers into a broader account of AI Alignment, discrimination, privacy, labor, and governance. The author recommends considering ethical and legal constraints from the outset. This is his interpretation of what the history should teach, rather than evidence that earlier ethical arguments determined later technical methods.^[1]

Reading the history critically

The compressed chronology has a concrete error: section 2.1 says McCarthy first coined “artificial intelligence” during the 1956 workshop, but the original proposal dated August 31, 1955 already uses the term in its title and opening.^[2] Individual dates, quotations, and claims of influence therefore need checking against their original records. Some footnotes in the available text extraction are also corrupted.^[1]

ResearchGate identifies an author upload on January 1, 2026. SSRN separately lists January 1 as the date written and January 28 as posting; those metadata do not establish that the two hosted texts are identical editions.^[1]

The conclusion favors human–AI collaboration, domain-specific problem solving, and continuing ethical scrutiny. The paper’s useful contribution is to make the aims and institutional conditions of AI research part of the historical question. For a comparison grounded in particular original arguments about teaching, purposes, and control, continue to Origins of Alignment: Problems Before the Name.^[1]