- The paper introduces a novel framework for evaluating epistemic virtues in AI by leveraging digital humanities and computational linguistics.
- It offers an empirical toolkit that maps semantic networks, revealing converging terms between creativity and generativity.
- It redefines traditional notions of good knowledge by emphasizing hybrid human-AI assemblages and emerging epistemic values in sociotechnical systems.
Good Knowledge and Epistemic Values in the AI Era
Introduction
"AI Virtue: What is 'Good' Knowledge in the Age of Artificial Intelligence?" (2607.01776) provides a rigorous examination of epistemic values and virtues in discourses surrounding AI, deploying digital humanities methodologies across a corpus of 553 AI-centric articles published in 2024. Alan Liu interrogates how established knowledge values—truth, accuracy, creativity—are shifting due to the ascendancy of generative AI, and proposes that the epistemic frameworks inherited from pre-AI knowledge work are increasingly insufficient for capturing the emerging sociotechnical and cognitive landscape. The paper suggests new directions for evaluating knowledge-worth in AI, emphasizing the generativity and reconfiguration of values, and offers practical tools—including an extensive online digital kit—for researchers to map and analyze epistemic values pertinent to AI.
Epistemic Virtues, Values, and Knowledge Frameworks
The paper foregrounds the inadequacy of default evaluative domains and scales—individual, organizational, disciplinary, and societal—that currently structure most discourse on AI. Critiquing the tendency to traverse these scales rigidly, Liu argues that AI's impact on knowledge space necessitates evaluating the interplay of agents, tools, and epistemic values in more nuanced ensemble configurations. Drawing on virtue epistemology, he delineates epistemic virtues (traits of "good knowers," e.g., open-mindedness, intellectual thoroughness) and epistemic values (attributes of knowledge itself, e.g., truth, accuracy, transparency), emphasizing that these categories are linguistically and conceptually entangled in practice.
Liu's survey of epistemic values reveals a field with no comprehensive canonical set, but rather an evolving vocabulary reflecting both disciplinary and sociotechnical specificity. He proposes tracking these values through word frequency, collocational analysis, and word embeddings, leveraging computational linguistics and machine learning techniques for empirical mapping.
Norms, Domains, and Sociotechnical Assemblages
The work critiques contemporary evaluation of AI as being overly reliant on static, pre-AI social-technological entities. Instead, it anticipates and partially theorizes new modes of organization in which humans and AI act as co-agents, especially evident in the emergence of "teams + AI" structures where both humans and multiple AI agents interact, swarm, and hybridize tasks.
Drawing from actor-network theory, poststructuralist philosophy (Foucault, Deleuze & Guattari, Latour), and large technical systems theory, Liu suggests that knowledge, virtue, and epistemic value must be understood as emergent properties of heterogeneous assemblages that cross prevailing organizational and cognitive domains and scales. Multi-agent LLM teams (as exemplified by Claude's agent orchestration features) and the rise of "agentic organizations" forecast a landscape in which the authority and value structures of knowledge production are increasingly aligned with the architectures of AI itself.
Case Study: Creativity and Generativity
A central empirical thread in the paper is the close examination of "creativity" as an epistemic value. Drawing on Boden’s tripartite model of creativity—combinatory, exploratory, and transformative—the analysis positions generative AI as disruptive primarily at the level of combinatory creativity, with growing but as yet inadequate capacities for exploratory and transformative innovation.
Significantly, the corpus analysis demonstrates that whereas terms like "creative" and "creativity" are more prevalent in humanities-oriented articles, business and STEM articles substitute proxies like "innovation." The work stresses that AI’s epistemic value spectrum is not zero-sum with respect to human capabilities. Instead, there emerges a possibility for new hybrid values—e.g., "argumentative creativity," "virtuosity" in hallucination—where the locus of "good knowledge" shifts from isolated creation to newly configured assemblages of human and machine effort.
Hallucination, Monoculture, and the Trickster Value Spectrum
The paper engages critically with recent attempts to reevaluate traditionally negative attributions such as "hallucination" in LLMs. Noting recent research that positions hallucination as "confabulation" with functional value in human sensemaking (Sui et al., 2024, Sui et al., 11 Nov 2025), Liu situates such revaluations as exemplary of a broader epistemic ambiguity, or "trickster" logic, wherein the same property may be pejorative or virtuous depending on context and system configuration.
The proliferation of terms such as "slop," "monoculture," and "enshittification" to stigmatize AI output is mapped onto a broader cultural and informational entropy—simultaneously signaling degeneration and the fecund potential for emergence of new forms and values. The analysis posits that entropy itself may be a generative epistemic value in the AI epoch.
The research offers methodological transparency and extensibility through the publication of its digital kit for analyzing AI discourse, including interactive visualizations, word embeddings, and topic models. This toolkit enables granular, iterative exploration of the shifting semantic networks of epistemic values, making it possible for future researchers to empirically test and extend the theoretical frameworks proposed.
Key findings from the toolkit include:
- Strong disciplinary variation in the usage and frequency of creativity-related terms.
- Evidence of converging semantic neighborhoods between "creativity" and "generativity," suggesting a potential future semantic and conceptual shift where generative capability becomes an epistemic value in its own right.
- Insights into the distributed agency of creativity—mapping not just whether AI is creative, but how the system-level configuration of humans, AIs, and organizations together instantiate creativity.
Implications and Prospective Directions
The theoretical implications of this research are extensive: it calls for expansion of the epistemological and organizational frameworks by which AI’s impact on knowledge and society is assessed. Practically, this means moving beyond technical benchmarks to include ongoing, empirical mapping of evolving epistemic values in sociotechnical systems.
Prospectively, the paper suggests that new value assemblages—arising from hybridization and reconfiguration of traditional virtues—will be crucial both for the distribution of labor (e.g., redistribution of creative, analytical, and responsible tasks between humans and machines) and for the legitimate appraisal of AI’s role in social and economic systems. The democratization of creativity, the potential redefinition of knowledge work, and the emergence of new epistemic value spectra are identified as domains requiring sustained critical attention as AI matures.
Conclusion
"AI Virtue: What is 'Good' Knowledge in the Age of Artificial Intelligence?" establishes a critical-empirical framework for mapping the shifting terrain of epistemic values in AI discourse, demanding both theoretical innovation and methodological rigor. By foregrounding the generative, hybrid, and ambiguous nature of epistemic values in the AI era, it challenges researchers to engage more dynamically with the concepts of knowledge, virtue, and value—both in analysis and in social practice. The online toolkit and analytic approach set out in the paper constitute a foundation for future research aiming to track, evaluate, and shape the evolution of good knowledge in the age of AI.