What if The Machines Worked Too Well: AI, Productivity, and the Demand Void
Executive summary
- This white paper argues that modern AI can create a macroeconomic “demand void”: production capacity rises while household incomes decline, breaking the circular flow that converts productivity into sustainable demand; this is a shift from the classic productivity paradox to a demand-deficiency paradox under deep automation.
- Short-run implementation frictions (the “J-curve”) help explain muted measured gains, but the deeper structural risk is that labour-displacing efficiency erodes the consumer base that firms ultimately depend on, amplifying recessionary pressures despite technological success.
- The paper outlines a mechanism for demand destruction, connects it to underconsumption theory, and proposes policy responses aimed at maintaining purchasing power when machines, not households, capture productivity.
Background: from productivity paradox to demand void
- The classic “productivity paradox” (Solow: “You can see the computer age everywhere but in the productivity statistics”) framed a gap between IT diffusion and measured productivity, highlighting implementation, measurement, and reorganization lags rather than inherent technological failure.
- AI-era dynamics invert the concern: organizations can realize process-level gains, yet macro-outcomes degrade if displaced income depresses aggregate demand, turning firm-level efficiency into system-level fragility via reduced consumption capacity.
- Recent modelling and empirical work on AI adoption underscores that productivity effects are path-dependent and sensitive to reorganization costs and labour distribution, reinforcing why early-phase underperformance (a “J-curve”) can coexist with long-run structural risks to demand.
The J-curve of AI adoption
- Implementation of general-purpose technologies often entails near-term declines as firms restructure processes, re-skill, and integrate workflows, which delays realized TFP gains; AI is no exception per recent estimation frameworks addressing “AI’s Solow paradox”.
- Simulation and process-mining studies show how short-run frictions suppress measured productivity even as technical potential grows, bridging the gap between micro-level case improvements and macro-level stagnation in early adoption phases.
- This J-curve explains “underwhelming at times” impacts on productivity statistics but does not eliminate the later-stage risk that gains are captured in ways that reduce wage income and weaken effective demand.
The demand-destruction mechanism
- Mechanism: when AI substitutes for labour in tasks that previously funded household consumption, unit costs fall but wage income also falls; unless returns are recycled to households at scale, consumer demand weakens relative to productive capacity, generating demand-deficient unemployment dynamics.
- In macro terms, if labour income shares declines faster than new income channels to households, the propensity to consume drops, raising the risk of negative multipliers and output gaps, even as firms individually optimize costs and throughput.
- General equilibrium risk: technology-driven savings at the firm level can become “paradox of thrift” effects at scale when investment opportunities do not expand commensurately and households cut spending, reinforcing underconsumption pressures through feedback loops.
Why “machines working too well” can destabilize capitalism
- Market economies rely on the circular flow—wages fund consumption that sustains output; when production shifts from wage-funded labour to capital-intensive, low-marginal-cost AI, the flow can decouple from household incomes unless countervailed by transfers or new income channels.
- If AI accelerates the substitution of labour faster than it catalyses new, broad-based income streams, consumer demand can lag productive potential, raising secular stagnation pressure despite high productive frontier—an “efficiency without participation” problem.
- Absent mechanisms that translate productivity into household purchasing power, the macro system risks chronic demand deficiency rather than a healthy equilibrium at higher output and lower prices.
Historical anchors and theory
- Prior IT waves exhibited measurement lags and reorganization costs, but ultimately integrated with labour in ways that broadened income; AI differs by targeting a wider range of cognitive tasks, intensifying displacement risk unless complemented by new demand channels.
- Underconsumption and demand-deficiency traditions in macroeconomics illuminate how aggregate demand can become insufficient even when productive capacity expands, especially if income concentrates in agents with low marginal propensity to consume.
- Recent macro-innovation modelling emphasizes the importance of labour share, savings behaviour, and substitution elasticities for growth paths and poverty traps, consistent with the risk that rapid labour substitution can induce non-linear demand shortfalls.
Measurement notes: reconciling micro wins with macro strain
- Firm-level case studies can register impressive cycle-time reductions and cost savings, yet national accounts capture realized productivity only after reorganization and diffusion, while simultaneously logging income and spending effects; this can yield mixed or negative short-run macro signals.
- Process-mining approaches show heterogeneous gains by function and sequencing, suggesting that sectoral aggregation can temporarily mute or offset measured productivity until complementarities and redesign are complete.
- Meanwhile, distributional changes can depress consumption even as technical efficiency rises, producing the paradox of “strong micro, weak macro” when incomes do not recycle to high-MPC households.
Policy implications: converting productivity into demand
- Wage-linked redistribution: tie a portion of AI-driven surplus to broad-based household income (e.g., negative income tax, wage subsidies, or profit-sharing designs) to preserve effective demand as labour’s task share shrinks.
- Investment in new demand channels: scale public and mission-driven investment that creates human-cantered services and infrastructure with high employment multipliers to counterbalance substitution effects and elevate aggregate spending.
- Tax-benefit alignment: shift fiscal frameworks toward capturing a share of automation rents (via neutral bases on economic rents rather than inputs) and recycling to households to maintain the circular flow without distorting adoption incentives.
- Transition insurance and re-skilling: accept the J-curve by funding time-bound support and reallocation services, while recognizing that not all displaced labour will be absorbed without income supports in high-substitution regimes.
- Competition and diffusion: prevent excessive concentration of AI gains that bottlenecks wage transmission; broaden access to AI complements for SMEs to foster income diversity and expand the set of demand-generating producers.
Research agenda
- Dynamic general equilibrium with endogenous adoption: quantify how labour share trajectories, substitution elasticities, and profit distribution interact with AI diffusion to predict demand gaps and optimal transfer schemes.
- Micro-to-macro transmission mapping: integrate process-mining productivity estimates with household income-expenditure data to forecast regional demand effects under different adoption speeds and surplus-sharing regimes.
- Measurement of J-curve phases: develop leading indicators that distinguish early reorganization drag from structural demand deficiency to guide countercyclical and structural responses in real time.
Conclusion
- AI can “work too well” for the macroeconomy if its efficiency gains sever the link between productivity and household purchasing power; short-run adoption frictions create J-curves, but the deeper hazard is a persistent demand void born of labour displacement without compensating income mechanisms.
- The central task of policy is to convert technical surplus into sustained effective demand—so that productivity does not become self-defeating, and abundance translates into widely shared prosperity rather than chronic underconsumption.
- http://arxiv.org/pdf/2210.01252.pdf
- https://arxiv.org/pdf/2403.17405.pdf
- https://scholarworks.lib.csusb.edu/ciima/vol10/iss4/2
- https://www.irrodl.org/index.php/irrodl/article/download/211/1625
- https://www.frontiersin.org/articles/10.3389/fams.2019.00001/pdf
- https://arxiv.org/pdf/1206.2494.pdf
- http://downloads.hindawi.com/journals/ddns/2018/6831508.pdf
- https://www.jstor.org/stable/10.2307/136426?origin=crossref
- http://www.tandfonline.com/doi/full/10.1080/01603477.2000.11490280
- https://www.semanticscholar.org/paper/4c73c16e1f1cf0eba5590c9e9e56077f919fad3d
- https://onlinelibrary.wiley.com/doi/10.1111/1468-0106.00180
- https://www.semanticscholar.org/paper/2b033c7c960e68bd2b2618240c32234daff38b00
- https://www.cambridge.org/core/product/identifier/S0034433800138540/type/journal_article
- http://ieeexplore.ieee.org/document/1406820/
- https://www.semanticscholar.org/paper/a5d2aa39a15a95f2dffade7e680d205489d1e324
- https://www.semanticscholar.org/paper/0b9c9edab3e1c5be828ce91a79f40a829cc1d65c
- https://arxiv.org/pdf/2310.09098.pdf
- http://www.efnahagsmal.is/article/download/a.2008.6.2.4/pdf
- https://zenodo.org/record/1768800/files/article.pdf
- https://zenodo.org/record/1448672/files/article.pdf
- Prihlásiť sa na účely uverejňovania komentárov