The Macroeconomics of Artificial Intelligence
Source details
- Canonical URL
- The Macroeconomics of Artificial Intelligence
Other formats
Bibliographic details
- Authors: Erik Brynjolfsson, GABRIEL UNGER
- Published: December 2, 2023
Overview
- Focus: implications of AI on productivity growth, the labor market, and industrial concentration.
- Central claim: AI’s economic impact is not predetermined; collective technological and policy choices today will shape divergent futures.
- Good policy requirements listed in the article:
- Creative policy experiments
- A set of positive goals for what society wants from AI, not just negative outcomes to be avoided
- Understanding that the technological possibilities of AI are deeply uncertain and rapidly evolving and that society must be flexible in evolving with them
Productivity growth
- Context:
- The US economy has been stuck with disturbingly low productivity growth for most of the past 50 years, except for a brief resurgence in the late 1990s and early 2000s.
- Productivity—output per unit of input—largely determines the wealth of nations and living standards.
- Low-productivity future (path of least resistance):
- AI adoption may remain slow and confined to large firms.
- AI could be a narrow labor-saving technology (a “so-so technology”) rather than enabling novel or powerful worker complements.
- Displaced workers might end up in less productive jobs, muting aggregate productivity gains.
- Economic gains may appear only decades after initial technological promise (Solow paradox).
- Organizational and managerial failures could blunt AI’s benefits.
- Legal and regulatory hurdles may arise around intellectual property, potentially creating a “patent thicket” that prevents models from being trained on data without clear rights.
- National regulators may impose strict regulations or bans that slow development and dissemination.
- High-productivity future:
- AI could be applied to a substantial share of tasks done by most workers (Eloundou and others 2023), massively boosting productivity in those tasks.
- AI might complement workers, freeing time for nonroutine, creative, and inventive tasks.
- AI could capture tacit knowledge from individuals and organizations using vast newly digitized data.
- Result: a permanently higher growth rate, broader applicability when integrated with robots, and AI-backed research enabling radical advances in medicine, math, science, and recursive self-improvement.
Income inequality
- Context:
- Increase in income inequality between individual workers over the past 40 years is a major concern.
- Past computerization and IT automated routine middle-income jobs, polarizing the labor force.
- Higher-inequality future:
- AI substitutes directly for many kinds of human labor, driving down wages for many workers.
- Generative AI produces words, images, and sounds—tasks once considered nonroutine and creative—expanding job threats.
- Entire industries may be upended; displacement concentrated into low-paying service jobs (hospital orderlies, nannies, doormen).
- Outcome: greater polarization into a small high-skilled elite and a large underclass of poorly paid service workers.
- This is not a future of mass unemployment, but of increased inequality.
- Lower-inequality future:
- AI assists least experienced or least knowledgeable workers to become more productive (example: software coders using Copilot).
- Study cited: among 5,000 workers in complex customer assistance jobs, those given an AI assistant had the greatest productivity gains among the least skilled or newest workers (Brynjolfsson, Li, and Raymond 2023).
- If employers share productivity gains with workers, income distribution could become more equal.
- AI may remove tedious routine work and complement creative tasks, improving work experience, reducing turnover, and increasing customer satisfaction.
Industrial concentration
- Context:
- Since the early 1980s, industrial concentration has risen dramatically in the United States and many other advanced economies; large “superstar” firms are more capital-intensive and technologically sophisticated.
- Higher-concentration future:
- Only the largest firms intensively use AI, becoming more productive, profitable, and larger.
- AI model development is expensive in raw computational power and massive datasets; examples:
- GPT-4 initial development cost: more than $100 million to train
- GPT-4 operating cost: about $700,000 a day to run
- Typical cost of developing a large AI model may soon be in the billions of dollars
- Scaling laws relating training costs to improved performance favor firms with the biggest budgets and datasets.
- Large firms may gain further advantages via AI-enabled internal coordination, strengthening the “visible hand” of top management.
- Lower-concentration future:
- Open-source AI models (examples in the article: Meta’s LLaMA, Berkeley’s Koala) become widely available, creating a vibrant open-source ecosystem.
- Small businesses gain access to industry-leading production technologies they could not otherwise afford.
- Internal memo (May 2023) cited: “open-source models are faster, more customizable, more private, and pound-for-pound more capable” than proprietary models; small open-source models can be repeated quickly and trained more cheaply.
- AI could encourage decentralized innovation, altering firm boundaries and enabling more innovators to found small firms.
Toward a policy agenda and research priorities
- Core point: the worse futures (low productivity, higher inequality, higher concentration) are the path of least resistance; reaching better futures requires active policy shaping.
- Policy questions to guide framing:
- How can policies encourage AI that complements human labor instead of replacing it?
- What choices will encourage AI development that firms of all sizes can access?
- What kind of open-source ecosystem is required, and how can policymakers support it?
- How should AI labs approach model development, and how should firms approach AI implementation?
- How does society get AI that unleashes radical innovation rather than marginal tweaks?
- Actors with power to affect direction:
- Major corporations, AI/computer science university labs, federal legislators and regulators, local policymakers, voters, and labor unions.
- Research needs:
- There is a deep imbalance between research advancing AI frontiers and research understanding AI’s economic and social consequences.
- Given potential effects measured in trillions of dollars, far greater investment should be made in research on the economics of AI.
- Society needs innovations in economic and policy understanding that match the scale and scope of AI breakthroughs.
Authors
- ERIK BRYNJOLFSSON: Jerry Yang and Akiko Yamazaki Professor at the Stanford Institute for Human-Centered AI; directs the Stanford Digital Economy Lab.
- GABRIEL UNGER: Postdoctoral fellow at the Stanford Digital Economy Lab.
Source: The Macroeconomics of Artificial Intelligence, ERIK BRYNJOLFSSON and GABRIEL UNGER, December 2023.
Content in this bundle
- F&D : THE MACROECONOMICS OF Artificial Intelligence