Interview
The better AI gets, the smaller its share of the economy might get – Alex Imas and Phil Trammell
- The human share of the economy is projected to decline relative to a machine-only economy as robotics and AI solve production processes, potentially leading to a scenario where the network-adjusted capital share reaches 100%.
- Labor market forecasts relying on individual economist predictions are deemed unreliable due to disagreement and the "lump of labor fallacy," with prediction markets proposed as a superior tool for aggregating wisdom.
- Historical precedents like David Ricardo's 1820 predictions suggest automation may displace jobs without preventing prime-age employment rates from reaching record highs by 2026, contingent on demand elasticities and new job creation.
- Economic models may need to explore scenarios where labor share drops to zero or remains constant, as current data on consumer demand elasticities and job destruction is insufficient for precise forecasting.
- If non-human goods become infinitely abundant, their marginal utility may fall faster than their quantity rises, reducing their economic value share, unless demand for new varieties of capital like computation never reaches satiation.
- The price of computation could remain high due to increased opportunity costs as models grow smarter, contradicting the historical trend of transistor value halving every 18 months.
- A "messy middle" scenario carries non-zero risk where automation displaces workers without generating sufficient wealth to compensate them, potentially causing political unrest and shifting the "drip" scenario of wage suppression into a political danger.
- White-collar automation may occur piecemeal by automating individual productive tasks without triggering the massive abundance effects seen in previous technological eras, though no current evidence indicates a "white collar apocalypse."
- Political outcomes could shift significantly with a 2% increase in unemployment, prompting risks associated with Universal Basic Income creating dependency on officials or Universal Basic Capital facing targeting failures.
- Government interventions like wealth taxes risk distorting investment through capital flight, while alternative proposals suggest using consumption taxes to purchase stock baskets for public distribution.
- While cheaper software does not inherently increase consumption, as agriculture demonstrates, scenarios predicting negative growth require improbable conditions such as capital owners refusing to invest or hard-bounded demand.
- Future production may prioritize AI labor with neural speeds due to transaction costs and reliability, making human integration difficult even with comparative advantages, and human requirements in licensing or regulation may be out-competed by AI systems.
- Selection pressures may favor entities with unsatisfiable demand for resources like compute rather than those seeking human-intrinsic goods, with autonomous agents potentially showing no preference for interacting with humans.
- Wealthy individuals who do not satiate in capital may drive a higher savings rate and capital share, though if humans live forever or capital becomes less scarce, returns may lower and incentives to compound wealth could diminish.
- "Greedy titans" or AI agents accumulating resources could drive the investment side of the economy, decreasing labor share even if final consumption remains human-mediated.
- Developing nations face divergence risks regarding access to AI hardware or model training, with potential outcomes ranging from being left behind to leveling the playing field through technology diffusion.
- For developing countries, a wide "messy middle" is possible if starting savings are low, though high interest rates driven by rapid technological growth could allow for significant consumption.
- Strategic recommendations for developing economies prioritize indexing into the global AGI economy via public markets as private AI companies go public, rather than relying solely on retraining.
- AGI becoming commoditized like electricity could distribute gains widely and increase prosperity, though it might intensify "tech race" dynamics by reducing the buffer for safety prioritization and increasing risks of harmful use.