S6 Ep20: Power and progress

Episode Summary

Executive Summary: Daron Acemoglu argues that today’s AI boom is being shaped less by neutral progress than by power: a small, homogeneous elite is steering technology toward automation, worker replacement, and greater inequality. He calls for democratic control, stronger worker voice, better taxation of capital, and even breaking up big tech to redirect innovation toward machine usefulness that complements people rather than displaces them.

Main Topics: Techno-optimism vs. power (Priority: 5/5): Acemoglu challenges the idea that technological advancement automatically benefits everyone, arguing that who controls innovation determines who gains and who loses. Automation, inequality, and labor markets (Priority: 5/5): He says AI and automation often raise average productivity while failing to improve workers’ productivity or bargaining power, so wages and shared prosperity do not necessarily rise. Machine usefulness over machine intelligence (Priority: 5/5): The book advocates designing technology to augment human capabilities—like calculators or Wikipedia—rather than pursuing autonomous intelligence aimed at replacing humans. Vision oligarchy and democratic control (Priority: 4/5): Acemoglu warns that a small elite in Silicon Valley has outsized influence over technological direction and that broader democratic participation is needed to shape innovation. Work, dignity, and social contribution (Priority: 4/5): He defends the importance of work not only as income but as a source of belonging, contribution, and social worth, warning that a workless society could become dystopian. Policy responses: taxes, regulation, and antitrust (Priority: 4/5): He discusses leveling taxes between capital and labor, supporting a temporary pause on frontier AI development, and considering breakup of big tech as tools to rebalance power. Rethinking economics and political economy (Priority: 3/5): The conversation closes with a call to bring power, institutions, and politics back into economics teaching and research, while noting the field is already moving in that direction.

Key Arguments: Technological progress is not automatically shared; outcomes depend on institutions, bargaining power, and social choices. AI’s current trajectory is overly centered on automation and replacement rather than complementing human workers. Average productivity can rise even when workers’ productivity and wages do not, especially if technology is used to eliminate labor. Big tech’s influence is not just economic but ideological; a narrow group sets the agenda for future technology. Democratic control of technology is necessary because engineers and CEOs should not unilaterally decide society’s technological future. Old-style unions may not return, but worker and societal voice can be rebuilt through new organizational forms and bargaining over data, work design, and technology. Current tax systems often subsidize capital and digital automation relative to labor, encouraging substitution away from workers. Breaking up major tech firms could reduce their ability to shape politics, markets, and policy agendas. Economics should treat power and institutions as central, not treat technology as a neutral productivity shifter. The goal should be machine usefulness: technology that helps humans do more, rather than making humans obsolete.

Data Points: Book writing timeline: More than 3 years - Acemoglu says he and Simon Johnson began writing Power and Progress over three years before the interview. Perspective horizon: 1,000 years - The book subtitle frames the argument as a long struggle over technology and prosperity across a millennium. Historical comparison: 300 years ago - He contrasts current living standards with life 300 years ago to note that progress has happened, but not automatically. AI pause proposal: 6 months - He references the open letter calling for a six-month halt to training more powerful AI systems. Patents in the United States: 5 to 6 fold increase over 40 years - Used to illustrate that innovation has accelerated even though broad productivity growth remains weak. Estimated automation share: 3% to 4% - He suggests the U.S. economy may automate only a small fraction of tasks and occupations, limiting aggregate productivity gains from pure automation. Productivity growth: As low as it has ever been - He argues that despite many new technologies and patents, U.S. productivity growth has been historically weak.

Pivotal Quotes: "We should want machine usefulness, machines being useful to humans." — Daron Acemoglu: He contrasts his preferred goal for AI with the Silicon Valley ideal of autonomous machine intelligence. "The extent to which productivity gains are shared between capital and labor are not equal. They are societal choices." — Daron Acemoglu: He explains that distributional outcomes from technology depend on institutions and power, not inevitability. "If I am at all techno-optimist, I'd be looking out for... that machine that could program really well, because as soon as I have one, I have as many as I need." — Tim Phillips: Phillips summarizes the corporate incentive to automate rather than complement human labor.

Implications: The episode urges listeners to see AI as a political and economic choice, not a neutral force. For policymakers and firms, the challenge is to design rules and incentives that favor human augmentation, labor voice, and democratic oversight over unchecked automation.