Generative AI promises productivity and innovation in information systems but adoption is hindered by hallucinations, bias and an absence of governance; IS research should move from documenting impacts to actively shaping human–AI co-evolution and adaptive regulatory and organizational practices.

Main Finding

Generative AI (GenAI) offers substantial transformative potential for productivity and innovation, but its diffusion and economic benefits are materially constrained by interrelated technical, social-ethical, and governance challenges. A persistent misalignment exists between rapidly evolving technical capabilities and slower-changing social institutions, creating adoption frictions and externalities that make purposeful socio-technical co-evolution — guided by information systems (IS) research and adaptive governance — essential to realize GenAI’s economic promise.

Key Points

• Transformative potential
  • GenAI can increase productivity, enable new products/services, and augment creativity and decision-making across firms and sectors.
• Technical constraints
  • Core limitations include hallucinations (incorrect but plausible outputs), performance drift across contexts, brittleness, and unpredictable failure modes.
• Societal and ethical risks
  • Risks include bias and discrimination, misuse (fraud, deception), erosion of human skills through overreliance, and unequal access that may widen disparities.
• Governance vacuum
  • Weak or lagging frameworks around privacy, accountability, intellectual property, liability, and standards create legal and strategic uncertainty for adopters.
• Socio-technical misalignment
  • The technical subsystem (rapid model innovation, probabilistic outputs) is outpacing the social subsystem (procedures, norms, institutions), producing coordination failures and adoption frictions.
• Role of IS research
  • IS should move from descriptive impact studies toward shaping co-evolution: design of hybrid human–AI ensembles, situated validation practices, design principles for probabilistic systems, and adaptive governance mechanisms.
• Research agenda highlights
  • Emphasize hybrid human–AI work systems, context-sensitive validation and monitoring, design for uncertainty, and institutional innovations for adaptive regulation and accountability.

Data & Methods

• Approach: Systematic literature review and synthesis focused on secondary studies and existing research agendas.
• Corpus: 28 papers published since 2023.
• Analytical lens: Socio-technical framework to interpret findings and identify cross-cutting themes across technical, social-ethical, and governance domains.
• Synthesis methods: Thematic aggregation of reported risks, adoption barriers, and proposed research directions; interpretation emphasizing joint optimization of technical and social systems.
• Limitations (inferred from scope)
  • Restricted to secondary studies and research agendas (no primary empirical microdata); limited timeframe (papers since 2023) may bias toward early, rapidly evolving viewpoints; possible publication and selection biases in the reviewed corpus.

Implications for AI Economics

• Productivity and TFP measurement
  • Estimating GenAI’s contribution to productivity and total factor productivity requires models that account for probabilistic outputs, quality adjustments, and complementarities with human labor. Standard productivity metrics may misstate effects without task-level measurement and verification.
• Adoption frictions and investment dynamics
  • Technical unreliability and governance uncertainty raise the effective cost and risk of adoption, creating option value to delay investment. This can slow diffusion and dampen aggregate gains even when technologies are welfare-improving in expectation.
• Labor markets and skill-biased change
  • GenAI affects task allocation: complementarities with cognitive and supervisory skills may raise returns to certain human skills while displacing routine tasks. Risk of skill erosion and reallocation costs suggests transitional unemployment and distributional impacts.
• Contracting, pricing, and market design
  • Probabilistic outputs complicate contracting (performance-based pay, warranties), pricing of AI services, and the design of verification/certification markets. New contracts and certification regimes (e.g., liability/insurance products) will be important to internalize risk.
• Intellectual property and innovation incentives
  • Unclear IP regimes for model outputs and training data create disputes over value capture and may distort incentives for data sharing, model innovation, and downstream entrepreneurship. Policy choices will shape market structure and concentration dynamics.
• Market power and concentration risks
  • High fixed costs, data advantages, and network effects can lead to winner-take-all dynamics. Governance and antitrust policy choices will materially affect competitive structure and the distribution of economic gains.
• Externalities and public goods
  • Negative externalities (misinformation, privacy harms, systemic risk) and positive externalities (public-good improvements) argue for public policy interventions (standards, liability rules, subsidies for validation/benchmarks).
• Research directions for AI economics
  • Develop structural models of adoption that incorporate technical unreliability, governance uncertainty, and human oversight costs.
  • Micro-econometric measurement of task-level productivity and labor reallocation in firms using GenAI, including methods for quality-adjusted output when outputs are probabilistic.
  • Contract theory applied to probabilistic AI outputs: optimal contracts, liability regimes, and insurance market design.
  • Models of innovation incentives under different IP and data-governance regimes to assess effects on competition and welfare.
  • General-equilibrium and distributional models capturing long-run effects on wages, inequality, and capital–labor shares under alternative policy scenarios.
  • Empirical work on certification, verification markets, and regulatory interventions (e.g., mandated testing, disclosure rules) to evaluate effects on adoption, safety, and welfare.
• Policy implications
  • Proactive, adaptive governance (standards, accountability frameworks, liability regimes) can reduce adoption uncertainty and externalities, accelerating safe diffusion and improving social welfare.
  • Policies that support worker retraining, certification for human–AI supervisory roles, and incentives for robust validation infrastructure can improve complementarities and reduce transitional costs.
  • Antitrust, IP, and data-governance choices will shape the distribution of gains; economic analysis should inform policy balancing dynamic innovation incentives with competitive markets and equity.

Overall, the review suggests that realizing GenAI’s economic potential depends as much on institutional design, governance, and workforce strategies as on model improvements. For AI economists, this points to research combining firm- and task-level measurement, contract and mechanism design for probabilistic systems, and evaluation of regulatory regimes that align private incentives with social welfare.