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.
William Stafford (Tue,) studied this question.