Skip to content

AI Virtue: What is "Good" Knowledge in the Age of Artificial Intelligence?

🕒 Published (v1): 2026-07-02 06:46 UTC · Source: Arxiv · link

Ask a follow-up

Open an assistant pre-loaded with this paper's context.

💬 Ask ChatGPT✦ Ask Claude

TL;DR

A digital humanities essay applying virtue epistemology to a corpus of 553 AI-related journal articles (2024), mapping the "epistemic values" (truth, accuracy, creativity, etc.) that scholars use to judge AI knowledge. It argues that current AI discourse is fragmented across preset domains and scales without an adequate integrative framework, and proposes "generativity" as a forward-looking epistemic value suited to the AI era.

Problem

Scholarly and public discourse about AI evaluates it by jumping discontinuously between preset domains (healthcare, education, business) and scales (individual user, organization, society), all anchored to pre-AI "knowledge work" assumptions. No sufficiently expansive, integrated framework exists to appraise AI's epistemic worth across these dimensions.

Method

Corpus study of 553 Web of Science journal articles published in 2024 with "artificial intelligence" or "AI" in their titles: 227 "highly cited" articles from WoS Core Collection and 326 from the Arts & Humanities Citation Index (AHCI). Both corpora were topic-modeled using MALLET, optimized at 22 topics each. Representative articles per topic were close-read. The author then applies virtue epistemology—the philosophical study of "epistemic virtues" (qualities of good knowers: open-mindedness, intellectual courage) and "epistemic values" (qualities of knowledge itself: truth, accuracy, coherence, creativity)—as an analytical lens on this discourse.

Key Contributions

  • Empirical mapping of epistemic values circulating in 2024 AI scholarship via dual-corpus topic modeling (WoS Core + WoS AHCI).
  • Diagnosis that current AI discourse operates through abrupt "preset" domain/scale jumps rather than an integrated sociotechnical framework.
  • Proposal of virtue epistemology as an epistemological (rather than ontological) framework for evaluating AI's knowledge-worth.
  • Identification of "generativity" as a candidate future-oriented epistemic value less locked into pre-AI norms.
  • Case study of "creativity" as an epistemic value undergoing redefinition under AI.
  • Critique of individual-scale (first-person voice) public-intellectual AI discourse for failing to bridge to collective/societal scales.

Results

  • The 22-topic model of WoS Core is dominated by healthcare, business, education, and regulatory/policy topics (e.g., "AI in Healthcare and Clinical Practice," "Medical Imaging and Diagnostics," "AI in Workplace and HR Management," "AI in Education and Academic Writing," "AI and Policy").
  • The 22-topic model of WoS AHCI identifies humanities/arts topics including writing education, literary studies, philosophy, music, and religious studies.
  • Close reading across both corpora shows discourse consistently "jumps" between individual-user and organizational/societal frames without conceptual connective tissue.
  • No quantitative benchmark results are reported; this is qualitative/interpretive scholarship.

Limitations

  • Corpus limited to Web of Science; Scopus unavailable; arXiv preprints excluded, meaning cutting-edge technical AI research is underrepresented.
  • Restricted to 2024 English-language published articles with "AI" in the title; 553 articles is a manageable but constrained scale.
  • No canonical or complete list of epistemic values exists to validate the framework against.
  • The proposed "generativity" framework is programmatic rather than operationalized; the essay does not resolve the domain/scale integration problem it diagnoses.
  • The 40-topic AHCI model is noted as finer-grained but less optimal for coherence.

Relevance to Agentic AI / LLM Agents

The paper directly engages agentic AI as a sociotechnical phenomenon, citing Claude Code's multi-agent teaming ("one session acts as the team lead, coordinating work, assigning tasks"), swarm parallelism, and third-party agent-orchestration frameworks as evidence that the unit of AI agency is shifting from individual LLM to coordinated agent ensembles. It raises the "medium is the message" concern that multi-agent protocols will reshape human team behavior and organizational structure—pointing toward what Menshikov et al. call "artificial sociality." For researchers tracking agentic AI, this paper provides a humanistic/sociological lens on the systemic and epistemic implications of moving from single-model to multi-agent deployments, situating technical developments in a broader framework of organizational and knowledge-work transformation.