Back
Research on AI hallucinations
webmisinforeview.hks.harvard.edu·misinforeview.hks.harvard.edu/article/new-sources-of-inac...
Published in Harvard Kennedy School's Misinformation Review, this piece is relevant to AI safety researchers and policymakers concerned with how hallucinations in LLMs intersect with broader information integrity and deployment safety concerns.
Metadata
Importance: 52/100journal articleanalysis
Summary
This article proposes a conceptual framework for categorizing and studying AI hallucinations, distinguishing them from other forms of misinformation and inaccuracy. It analyzes the unique epistemological challenges posed by generative AI systems that produce plausible but factually incorrect outputs. The framework aims to guide researchers and policymakers in understanding and mitigating AI-generated misinformation.
Key Points
- •Introduces a structured taxonomy distinguishing AI hallucinations from traditional misinformation, disinformation, and human error.
- •Argues that AI hallucinations represent a novel category of inaccuracy with distinct causes rooted in model architecture and training data.
- •Highlights the epistemological challenge of hallucinations being difficult for users to detect due to their fluency and surface-level plausibility.
- •Provides a conceptual foundation intended to guide empirical research and policy responses to AI-generated inaccuracies.
- •Situates AI hallucinations within the broader misinformation ecosystem, noting implications for public trust and information integrity.
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| Epistemic Collapse | Risk | 49.0 |
Cached Content Preview
HTTP 200Fetched Mar 20, 202645 KB
[Skip to main content](https://misinforeview.hks.harvard.edu/article/new-sources-of-inaccuracy-a-conceptual-framework-for-studying-ai-hallucinations/#main-content)
In February 2025, Google’s AI Overview fooled itself and its users when it cited an April Fool’s satire about “microscopic bees powering computers” as factual in search results (Kidman, 2025). Google did not intend to mislead, yet the system produced a confident falsehood. Such cases mark a shift from misinformation caused by human mistakes to errors generated by probabilistic AI systems with no understanding of accuracy or intent to deceive. With the working definition of misinformation as any content that contradicts the best available evidence, I argue that such “AI hallucinations” represent a distinct form of misinformation requiring new frameworks of interpretations and interventions.
###### By
[Anqi Shao](https://misinforeview.hks.harvard.edu/article/author/anqi-shao "Posts by Anqi Shao")
Department of Life Sciences Communication, University of Wisconsin-Madison, USA
###### Article Metrics
7
CrossRef Citations
527
PDF Downloads
8020
Page Views
image by [Michi-nordlicht](https://pixabay.com/users/michi-nordlicht-5407710/ "") on [pixabay](https://pixabay.com/)
###### Topics
- [Artificial Intelligence](https://misinforeview.hks.harvard.edu/explore/?fwp_topic=artificial-intelligence "Artificial Intelligence")
* * *
## Introduction
AI hallucinations are inaccurate outputs generated by AI tools, such as ChatGPT, Gemini, and Claude, that appear plausible but contain fabricated or inaccurate information (Augenstein et al., 2024). These inaccuracies can emerge from AI systems without deliberate human intent to deceive. Unlike traditional (human) communicators, AI lacks in intent or epistemic awareness that would allow it to recognize or prevent the generation of hallucinated content. Platform guardrails (e.g., internal filters) and retrieval-augmented generation (RAG) systems (e.g., the “search the web” function from many AI tools) can improve factual accuracy, yet hallucinations persist because these models generate language by predicting the next most likely word based on statistical patterns in training data. This characteristic makes AI hallucinations different from human-driven misinformation, which current research attributes to cognitive bias, motivated reasoning, or attempts to deceive.
AI hallucinations are at the boundary of what scholars define as misinformation (Schäfer, 2023). Are they just technical errors, or do they act like human-generated misinformation in influencing public decision-making? Existing interventions such as fact-checking (Krause et al., 2020), accuracy nudges (Pennycook et al., 2020), or alignment strategies (Gabriel, 2020) often assume the presence of
... (truncated, 45 KB total)Resource ID:
b0bf272733103485 | Stable ID: ZDViMjUxZj