Summary of Key Points
This article discusses the impact of artificial intelligence (AI) as a new generation of general technology on the economy, focusing on four key issues: whether AI can break through the linear bottlenecks in economic growth (optimists believe AI will lead to exponential growth, while pessimists point out that bottleneck tasks limit this potential); the impact of AI on the labor market—whether it will replace jobs or enhance human capabilities; the distortion of wealth distribution and the invalidation of statistical measures caused by AI (GDP does not reflect consumers' actual benefits, and the increasing proportion of capital exacerbates wealth inequality); and how to address these challenges through policy adjustments (solutions are proposed in areas such as taxation, anti-monopoly measures, data ownership rights, and public procurement). The article concludes by emphasizing that the impact of AI is not predetermined by technology itself but depends on whether institutional designs can guide it towards inclusive growth that benefits workers.
I. Macroeconomic Growth: Can AI Trigger Exponential Growth?
Traditional economic models suggest that as investment in research and development (R&D) increases, marginal returns decline (for example, despite the significant increase in researchers and funding over the past century, developed economies have only grown at an average of around 2% annually). However, optimistic scholar Corneick argues that AI could break this pattern—once its cognitive capabilities reach a critical threshold, it will become a “R&D workforce that can improve itself.” For instance, AI can assist in designing chips, writing code, and accelerating scientific experiments (similar to the rapid trial-and-error process in structural biology), creating a snowball effect that transforms linear growth into exponential growth.
Pessimist Jones, on the other hand, cites the “barrel theory”: economic growth is like a barrel made up of many complementary tasks, and the weakest link (the most difficult tasks to automate) determines the overall limit. Even if some tasks become infinitely efficient, their impact on total output will be limited. For example, even if AI can handle data perfectly, if the final step requires complex human decision-making (such as in medical surgeries), economic growth will still be constrained by this bottleneck.
II. Labor Market: Will AI Take Jobs or Act as a Super Assistant?
The industrial revolution replaced physical labor; those displaced could move to mental work (e.g., becoming engineers). However, AI is different as it can replace cognitive tasks—large language models can process more data than humans in their entire lives and engage in logical reasoning (e.g., self-correction through “chain thinking”). Purely intellectual workers (such as programmers and consultants) may face the risk of their skills becoming obsolete.
Nobel laureate Acemoglu proposes a “worker-friendly” approach to AI: AI does not necessarily have to replace humans but can serve as an assistant. For example, in healthcare, AI can help doctors analyze CT scans for abnormalities while doctors make the final diagnosis and communicate with patients; in engineering, AI can handle preliminary designs while engineers refine them. This complementarity between humans and AI will not devalue human skills but allow professionals to focus on more valuable higher-level tasks. The key lies in the direction of corporate R&D: whether to create systems that completely replace humans or tools that assist them.
III. Wealth Distribution: Why Does There a Disparity Between Statistical Data and Real Life?
The first issue is the inaccuracy of GDP: AI reduces the cost of many services (e.g., software development, consulting), allowing consumers to access better services for less money (e.g., free AI tools). However, GDP measures only the monetary value of market transactions, which may lead to an illusion of economic decline despite improved quality of life. The second issue is unfair wealth distribution: if AI replaces many jobs, capital owners (e.g., AI company owners) will earn more, while laborers’ incomes decrease. Given that capital ownership is already concentrated in the hands of a few, this will widen the wealth gap. Some suggest providing a “basic income” for everyone, but this has two problems: current taxes rely mainly on wage income, and with fewer jobs, tax revenue will decrease; moreover, money alone cannot address the loss of dignity that comes with job loss.
IV. How to Address These Issues Through Policy?
To guide AI towards inclusive growth, policies need to focus on the following areas:
1. Correct Tax Bias: Many countries impose high taxes on hiring employees (e.g., social security, wage taxes), while reducing taxes on companies purchasing AI equipment and software. Policies should balance these to encourage companies to use AI for practical needs rather than tax avoidance.
2. Anti-Monopoly and New Business Models: Large tech companies may monopolize computing power and data and acquire startups that develop worker-friendly AI. Regulations are needed to prevent this, allowing small companies to develop assistant technologies.
3. Protect Data Ownership Rights: AI training relies on the knowledge of professionals (e.g., doctors’ diagnostic experiences, lawyers’ case studies), but these individuals do not receive compensation. New rules should ensure that knowledge creators benefit from AI usage, such through shared profits.
4. Public Procurement: Governments can prioritize purchasing worker-friendly AI systems to drive companies to develop such technologies.
5. Adjust Career Barriers: Some professional certifications are unnecessary and serve to protect existing workers. Policies should lower these barriers to enable ordinary people to use AI to enhance their skills, such as allowing non-professionals to use AI in licensed jobs (as long as safety is not compromised).
6. Global Governance of Computing Power: A “computing power tax” on high-end chips can slow down unregulated AI competition and fund research on AI security.
Conclusion
AI is not a “black technology” that will inevitably either boost the economy or lead to mass unemployment. Its impact depends on how our institutions are designed. If policies guide AI to be a tool that enhances human well-being rather than a replacement, and if everyone can benefit from its benefits, then AI can truly contribute to societal progress. Otherwise, if capital dictates the path of AI development, it may exacerbate wealth inequality and social tensions. The key is for institutional innovation to keep up with the pace of AI advancement.