A large bibliometric mapping shows human–AI interaction and collaboration dominate academic attention while conversational AI and ethics are more specialized, and the authors offer a four-type framework—symbiotic, augmented, assisted, substituted—to clarify human–AI relationships and guide future organizational research.

Main Finding

The field of human–AI relationships is conceptually organized around five thematic clusters—Human–AI Interactions, Human–AI Collaboration, Teaming and Augmentation, Conversational AI, and Ethics & Responsibility—with the first three forming the structural core of the literature. From this empirical mapping the authors propose a four-category conceptual framework (symbiotic, augmented, assisted, substituted intelligence) that classifies human–AI relationships by AI autonomy and human involvement. The contribution is primarily scholarly: clarifying conceptual structure and pointing to research opportunities rather than prescribing organizational practice.

Key Points

• Dataset and scope
  • Bibliometric co-word analysis of 4,093 peer‑reviewed documents indexed in Scopus.
• Thematic clusters (with centrality scores indicating structural importance)
  • Human–AI Interactions — Centrality: 1595 (foundational)
  • Human–AI Collaboration — Centrality: 1150 (foundational)
  • Teaming and Augmentation — Centrality: 1131 (foundational)
  • Conversational AI — Centrality: 655 (specialized)
  • Ethics and Responsibility — Centrality: 431 (specialized)
• Analytical approach
  • Strategic diagram used to assess relevance and maturity of themes.
  • Co-word mapping identifies intellectual structure rather than testing causal claims.
• Conceptual framework
  • Four relationship types distinguished by AI autonomy and human involvement:
    • Symbiotic intelligence
    • Augmented intelligence
    • Assisted intelligence
    • Substituted intelligence
  • Framework intended as a taxonomic and theoretical tool for future empirical and theoretical work.
• Scope and purpose
  • Emphasis on scholarly consolidation and conceptual clarity.
  • Ethical alignment and interdisciplinary dialogue are highlighted as critical cross-cutting needs.
  • Not prescriptive guidance for practitioners; organizational implications are secondary and inferential.

Data & Methods

• Data source: 4,093 peer‑reviewed publications from Scopus.
• Method: bibliometric co‑word analysis to detect frequently co-occurring terms and map the intellectual landscape.
• Visualization/assessment: strategic diagram to evaluate each thematic cluster’s centrality (relevance to the field) and density (maturity/cohesion).
• Outputs: identification of five major thematic clusters and their relative positions in the field; proposal of a conceptual taxonomy linking AI autonomy and human agency.
• Limitations (implicit in method): descriptive mapping (not causal), dependent on Scopus coverage and keyword selection, may underrepresent non‑indexed or gray literature and practice-oriented sources.

Implications for AI Economics

• Labor markets and task allocation
  • The four-category framework (symbiotic → substituted) provides a useful taxonomy for modeling task-level complementarity versus substitution between labor and AI, refining predictions about which jobs or tasks are vulnerable to automation versus those that will see productivity complements.
  • Research can map tasks and occupations to framework categories to estimate differential impacts on wages, employment, and skill premiums.
• Productivity and firm-level returns
  • Foundational clusters (Interaction, Collaboration, Teaming/Augmentation) suggest major channels through which AI affects firm productivity: improved decision-making, coordinated human–AI workflows, and augmentation of worker capabilities. Empirical work should estimate returns to AI investments conditional on the type of human–AI relationship.
• Human capital and skill demand
  • The prominence of augmentation and collaboration themes implies increased demand for skills that enable effective human–AI teaming (e.g., oversight, interpretation, domain expertise). This motivates models of investment in complementary skills and retraining policies.
• Market structure and organizational design
  • Different relationship types imply different adoption thresholds, fixed costs, and returns to scale. Substituted-intelligence applications may concentrate market power (higher returns to incumbent platforms), while symbiotic/augmented forms may decentralize capabilities across firms and workers.
• Policy, regulation, and externalities
  • The Ethics & Responsibility cluster (specialized but present) signals regulatory and welfare-relevant concerns—liability, fairness, transparency—that shape adoption incentives and social returns. Economic analysis should incorporate regulatory constraints and potential negative externalities (e.g., biased outcomes, privacy harms) into welfare calculations.
• Measurement and empirical agenda
  • Bibliometric findings motivate a task-based empirical agenda: link textual/organizational indicators (e.g., job postings, software adoption, process changes) to the four relationship categories to quantify effects on productivity, firm performance, and distributional outcomes.
  • Natural experiments, firm-level microdata, and matched employer‑employee panels can be used to identify causal impacts across relationship types.
• Interdisciplinary coordination
  • The paper’s call for interdisciplinary dialogue matters for economic modeling: integrating insights from HCI, organizational behavior, and ethics will improve behavioral assumptions, constraint specifications, and policy prescriptions in economic models of AI.
• Research priorities for AI economics
  • Estimate complementary vs. substitutive impacts at the task level.
  • Model investment in human capital under different human–AI relationship regimes.
  • Evaluate distributional consequences and design policies (training subsidies, liability rules, platform regulation) sensitive to relationship type.
  • Incorporate ethical/regulatory risk into firm adoption and innovation models.

Overall, the study supplies a structured conceptual scaffold that AI economists can use to classify technologies and mechanisms before estimating their economic effects—helping to target empirical work, refine theoretical models, and ground policy analysis in a clearer taxonomy of human–AI relationships.