A new macro growth model finds AI boosts output only when human–AI synergy keeps pace: without investment in organisational and human-capital 'synergy', AI accumulation risks labor displacement and stagnant per‑capita gains.

Main Finding

The paper develops the Human-AI Complementarity Production Function (HACPF), a Cobb–Douglas–style growth framework that treats AI stock (AI) and an explicitly modeled Human–AI Synergy Capacity (C) as independent multiplicative inputs alongside physical capital (K) and labor (L). The core result is that whether AI substitutes for or complements human labor depends crucially on the level and responsiveness of C: high/expanding synergy capacity makes AI augment human productivity and supports higher per‑capita and possibly endogenous long‑run growth; low or lagging synergy leads to labor displacement and adverse distributional outcomes.

Key Points

• Model specification: Y = A * K^α * L^β * AI^γ * C^δ, with 0<α,β,γ,δ<1 and Sum = α+β+γ+δ determining returns to scale.
• Synergy modeled endogenously: C = θ * L^ω * AI^ψ * E^φ, where E denotes investments (education, institutions), ω and ψ capture L–AI responsiveness, and θ institutional alignment.
• Marginal products incorporate synergy feedbacks:
  • dY/dL = β(Y/L) + δ (Y/C) (dC/dL)
  • dY/dAI = γ(Y/AI) + δ (Y/C) (dC/dAI) These terms can make AI a net complement if the synergy channel is strong enough.
• Augmentation vs replacement hinge on δ (synergy elasticity) and cross‑derivatives of C:
  • Small δ or negative cross‑derivatives → AI substitutes for labor (displacement).
  • Large δ and positive cross‑derivatives → AI augments labor (wage gains for synergistic workers).
• Growth decomposition: g_Y = g_A + α g_K + β g_L + γ g_AI + δ g_C. Persistent accumulation of AI and C can sustain endogenous long‑run growth.
• Policy and firm strategy implications emphasize investing in human capital, institutional redesign, organizational capital, and user‑centric AI deployment to raise C.
• Limitations: aggregate homogeneity, static/simple C specification, no explicit income‑distribution utility optimization, and no multi‑sectoral/dynamic friction modeling.

Data & Methods

• Nature of research: theoretical/analytical model; no empirical dataset presented.
• Methods:
  • Extended Cobb–Douglas production function to include AI and C.
  • Explicit functional form for synergy (C) linking labor, AI, and educational/institutional investment.
  • Analytical derivation of marginal products, first‑order partial derivatives, and conditions for complementarity vs substitution.
  • Logarithmic differentiation to derive growth accounting and steady‑state implications.
  • Parameter assumptions: elasticities in (0,1); Sum may be ≤, =, or >1 (allowing constant or increasing returns).
• Analytical focus on qualitative comparative statics (how changes in AI and C affect marginal products, wages, and growth) rather than calibration or econometric estimation.
• Stated extensions: multi‑sector models, dynamic differential equations for C (adjustment lags, transitional unemployment), and incorporation of inequality in utility optimization.

Implications for AI Economics

• Theoretically centers "synergy capacity" (C) as the decisive channel that determines AI’s macroeconomic role. Models and empirical work should treat human–AI interaction capacity as a distinct factor, not just embodied technological capital or exogenous TFP.
• Policy emphasis:
  • Shift from subsidies for AI hardware/software alone toward education, retraining, institutional reforms, and organizational investments that raise C (digital literacy, collaboration skills, change management).
  • Anticipate distributional effects: without C growth, AI accumulation can depress labor shares and raise inequality; with C growth, gains are more likely to be broadly shared.
• Firm/managerial implications:
  • AI deployments must be accompanied by redesign of workflows, user training, and human‑centered interfaces; otherwise AI-led productivity gains will be bottlenecked by low C.
• Empirical agenda suggestions:
  • Construct measurable proxies for AI stock (investment flows, model counts, compute capacity) and for C (training intensity, adoption metrics, organizational practices, worker digital skills).
  • Estimate elasticities (α,β,γ,δ) and interaction effects; test whether interaction terms (AI × C or AI × human capital) predict labor productivity and wages.
  • Use cross‑industry or within‑firm panel variation to identify complementarities and transitional dynamics (e.g., event studies around AI adoption combined with workforce training programs).
• Modeling extensions to pursue:
  • Multi‑sector/task‑based HACPF to capture heterogeneity between routine and non‑routine activities.
  • Dynamic microfoundations for C with adjustment costs and frictions (transitional unemployment, retraining lags).
  • Explicit treatment of distributional preferences to link HACPF to welfare and inequality outcomes.

Concise takeaway: AI’s macroeconomic impact depends less on raw AI stock than on the capacity of institutions, education, and organizations to convert AI into human‑AI synergy; modeling and policy that ignore C risk mispredicting both growth and distributional outcomes.