Paper Number
ICIS2026-2932
Paper Type
Completed
Abstract
Generative AI hallucinations are often framed as organizational liabilities that require suppression. We argue this framing, while appropriate for factual fabrications, is theoretically incomplete for speculative recombinations, hallucinations that cross semantic boundaries in generatively useful ways. Drawing on Blind Variation and Selective Retention (BVSR) theory and serendipity research, we theorize speculative recombinations as productive noise: a novel form of machine-generated blind variation uniquely positioned to trigger true serendipity, the redirection of organizational goals rather than merely their advancement. Four propositions specify how hallucinations expand the organizational idea set (P1), structurally undermine selective retention through evaluative ambiguity (P2), trigger true serendipity when a prepared mind operates within enabling organizational conditions (P3), and amplify innovation outcome variance (P4). We contribute an extension of BVSR to machine-generated variation, a sharpened pseudo vs. true serendipity distinction for IS innovation research, and governance principles for responsibly harnessing hallucinations as strategic innovation resources.
Recommended Citation
Saeedi, Milad; Austin, Robert D.; and Su, Ning, "Productive Noise: Theorizing GenAI Hallucinations and Serendipitous Innovation" (2026). ICIS 2026 Proceedings. 12.
https://aisel.aisnet.org/icis2026/innov/innov/12
Productive Noise: Theorizing GenAI Hallucinations and Serendipitous Innovation
Generative AI hallucinations are often framed as organizational liabilities that require suppression. We argue this framing, while appropriate for factual fabrications, is theoretically incomplete for speculative recombinations, hallucinations that cross semantic boundaries in generatively useful ways. Drawing on Blind Variation and Selective Retention (BVSR) theory and serendipity research, we theorize speculative recombinations as productive noise: a novel form of machine-generated blind variation uniquely positioned to trigger true serendipity, the redirection of organizational goals rather than merely their advancement. Four propositions specify how hallucinations expand the organizational idea set (P1), structurally undermine selective retention through evaluative ambiguity (P2), trigger true serendipity when a prepared mind operates within enabling organizational conditions (P3), and amplify innovation outcome variance (P4). We contribute an extension of BVSR to machine-generated variation, a sharpened pseudo vs. true serendipity distinction for IS innovation research, and governance principles for responsibly harnessing hallucinations as strategic innovation resources.
Comments
15_Innovation_and_entrepreneurship