Crossing the Chasm from LLM Hallucinations to Invention via Abduction
Proceedings of the AAAI Symposium Series · 18.05.2026 · 8 (1) · 648-656
DOI: 10.1609/aaaiss.v8i1.42601 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Symposium Series, online erschienen am 18.05.2026. Verfasst von Boris Galitsky.
Abstract (Originalsprache)
Large Language Models (LLMs) are increasingly employed as collaborative partners in creative and problem-solving dia-logues, yet their usefulness is constrained by hallucinations—plausible but unsupported or inconsistent statements that are typically treated as reliability failures. This paper argues that, in inventive human–LLM interactions, hallucinations can also function as productive cognitive perturbations that expand the hypothesis space and seed innovation. We propose a neuro-symbolic framework that reframes hallucinated outputs as low-prior abductive hypotheses, which are then evaluated and trans-formed through constraint satisfaction, counter-abduction, and human oversight. Using a curated Hall2Invent dataset and a suite of evaluation metrics, we show that abductive and con-straint-based reasoning substantially improves hallucination identification, enables systematic repair of flawed explanations, and increases the yield of feasible and non-trivial inventions across engineering and systems domains. Our results demon-strate that symbolic reasoning not only reduces harmful reason-ing hallucinations but also preserves and channels the creative potential of LLMs. We conclude that trustworthy creative AI should not aim to eliminate hallucinations outright, but to gov-ern them through structured reasoning processes that bridge the gap between error and invention.
Zitieren
APA
Galitsky, B. (2026). Crossing the Chasm from LLM Hallucinations to Invention via Abduction. Proceedings of the AAAI Symposium Series, 8(1), 648-656. https://doi.org/10.1609/aaaiss.v8i1.42601
Vancouver
Galitsky B. Crossing the Chasm from LLM Hallucinations to Invention via Abduction. Proceedings of the AAAI Symposium Series. 2026;8(1):648-656. doi:10.1609/aaaiss.v8i1.42601
BibTeX
@article{Galitsky2026,
author = {Galitsky, Boris},
title = {Crossing the Chasm from LLM Hallucinations to Invention via Abduction},
journal = {Proceedings of the AAAI Symposium Series},
year = {2026},
volume = {8},
number = {1},
pages = {648--656},
doi = {10.1609/aaaiss.v8i1.42601}
}
EUVIMED