Partial access to generative AI can improve team communication: when only some teammates use GenAI they become more dominant and create useful role differentiation, boosting short-term team communication, while giving everyone equal access removes that advantage.

Main Finding

Unequal access to generative AI (only some team members using AI) led the AI users to adopt more dominant communication styles, producing larger within-team dominant-style differences. Those greater style differences improved team communication effectiveness and, in turn, raised short-term team performance relative to both (a) the no-AI baseline and (b) the full-access condition where all members used AI.

Key Points

• Research question: How do different AI-integration structures (no access, unequal access, full access) affect individual dominance, team-level style differences, communication effectiveness, and team performance?
• Core mechanism: AI access → capability enhancement + information asymmetry → more dominant communication by AI users → larger team dominant-style differences → improved communication effectiveness → higher team performance (serial mediation).
• Empirical result (summary): Unequal access produced the largest dominant-style differences and the highest communication effectiveness compared with both no-access and full-access teams.
• Theoretical tension resolved: Contrary to a simple pro-equality intuition, partial/unequal AI provision can yield short-term coordination and communication benefits in ad hoc two-person teams.
• Boundary/context: Findings are from short-term, ad hoc, two-person laboratory teams (student sample, confederates used), so generalizability to larger, long-lived, or organizational teams is limited.

Data & Methods

• Design: Randomized controlled laboratory experiment with two-phase tasks (control then treatment).
  • All teams completed a control task with no AI access first; then teams were randomized into treatment conditions.
• Sample: 60 two-person teams formed by 60 participants (students) paired with five trained confederates (i.e., one participant + one confederate per team). Sample was 75% female; ~98% had prior GenAI experience.
• Conditions:
  • No-access (control phase for all teams).
  • Unequal access (treatment): exactly one team member allowed to use GenAI — either the participant or the confederate (combined n = 40 teams).
  • Full access: both team members allowed to use GenAI (n = 20 teams).
• Task: Collaborative press-release writing under time constraint (700 characters; electric bicycle in control, AR glasses in treatment); each task ≤45 minutes.
• AI tool: Kimi 3.0 (a Chinese LLM) was the allowed GenAI for teams assigned AI access.
• Confederate protocol: Five confederates trained to adopt neutral communication styles to hold partner behavior constant across teams.
• Measurements (as reported / implied):
  • Individual communication dominance coded/measured from interaction.
  • Team-level dominant-style difference computed from individual measures.
  • Team communication effectiveness assessed (reported improvement under unequal access).
  • Team performance measured via task outputs (and monetary bonus linkage described).
• Ethics and compliance: IRB-approved experimental protocol; on-site recordings used to ensure adherence.

Limitations of methods (noted by authors or inferred) - Use of confederates controls partner variance but limits naturalistic interaction. - Short-term, two-person, student-based lab setting constrains external validity to larger or long-lived organizational teams. - Details on exact measurement scales and statistical estimates were not included in the provided excerpt.

Implications for AI Economics

Practical and organizational implications - Strategic, partial deployment of AI tools can be welfare-improving for short-term, goal-directed ad hoc teams: concentrating AI access can raise communication effectiveness and short-run productivity by creating clearer leadership/knowledge-integration roles. - Blanket full-access rollouts are not always strictly welfare-superior for every task/context. Firms may consider targeted access (or role-based provisioning) when rapid coordination and clear direction are priorities. - However, unequal access creates status and information asymmetries. While beneficial in the short term for two-person ad hoc tasks, these asymmetries may produce longer-term costs (morale, skill erosion, perceived unfairness) that firms and policymakers need to weigh.

Market and policy-level implications - Platform and pricing strategies: AI providers and platforms could offer tiered access or role-based products (e.g., single-seat “assistant” licenses for team leads) that align with organizational coordination needs. - Labor-market effects: Differential AI access within teams can change intra-team bargaining power and may contribute to wage dispersion or changes in task allocation if advantages persist. Monitoring who gains access may be important for equity. - Regulation and fairness: Policymakers should consider not only access quantity but distributional effects within organizations. Interventions (training, rotation of access, transparency) may be needed where unequal access risks exclusionary effects.

Research implications and open questions for AI economics - External validity: Do the communication and productivity gains from unequal access scale to larger teams, repeated interactions, or different task types (e.g., deliberative vs. execution tasks)? - Dynamics over time: Will short-run gains from unequal access persist, or will full access become superior for long-lived teams because of learning, diffusion, and reduced coordination friction? - Welfare trade-offs: How do short-term efficiency gains compare to distributional harms (skill-atrophy, exclusion) over the medium and long term? Quantify trade-offs in field settings. - Organizational design variants: Explore optimal policies (rotating access, role-based access, staged rollouts, training) to capture coordination benefits while mitigating inequality costs. - Economic modeling: Incorporate asymmetric AI access into models of team production, bargaining, and human capital accumulation to predict aggregate labor-market impacts.

Bottom line In short-term ad hoc two-person teams, unequal AI access can increase dominance differences and improve communication effectiveness and performance. For AI economics, this suggests that selective allocation of AI resources can be an instrument for improving coordination and productivity, but designers and policymakers should weigh short-run gains against potential long-run distributional and skill-development costs.