The Lost Innovators Hypothesis: A Single-Case Documentation of AI-Mediated Theoretical Inquiry by a Non-Credentialed Author

The credentialing structure of formal knowledge production excludes a population of capable individuals—the "lost innovators"—who possess sustained interest in researchable questions and latent intellectual capacity but lack the tools, training, and access to translate those questions into formal inquiry. Recent empirical literature on generative artificial intelligence documents productivity gains concentrated disproportionately among lower-skilled workers (skill compression), but does not address whether AI mediation also expands the demographic base of contributors to formal knowledge production—the distinct claim of the Lost Innovators Hypothesis. This paper documents one case in the autoethnographic tradition: the development of a theoretical position by a non-credentialed individual with documented learning differences, using multi-system AI mediation (ChatGPT, Claude, Perplexity, Grok) between approximately 2024 and 2026. The case provides existence proof that the production-and-defense form of the phenomenon can occur; it does not establish frequency or scale, nor that the output constitutes validated formal knowledge production (which awaits external peer review). Three contributions follow: transparent documentation of one instance, including specific AI failure patterns (vague institutional attribution and substantive source mischaracterization) that depart from the conventional "AI hallucinates citations" framing; a methodological decomposition of AI-mediated knowledge work into four functionally distinct modes (generative drafting, advisory critique, external validity audit, research assistance); and an empirical research program with four concrete study designs, explicit falsifying conditions, and policy implications for AI access equity, educational integration, accessibility, and verification literacy. The paper is hypothesis-generating rather than hypothesis-testing. Changes in version 3.2 (conformance version; no changes to argument, evidence, or citations): (1) The byline is brought into conformance with the author's other published work and the Earned Trust standard v1.6.3 — William Stafford, ADN, LI-AIast4 — on the cover and in the AI-use disclosure, with the house cover format (version, deposit DOI, and licence stated on the cover). (2) Language referring to the author's age and to specific personal dates is made age-neutral throughout (e.g., "formulated as a teenager," "carried across the decades"), with no change to the case documentation's substance. (3) One factual description is corrected: the machine the subject encountered as a teenager is now identified as an IBM 5100 portable computer rather than "one of the early IBM personal computers."