Summary of Key Points
This article discusses the paradox of rapid advancements in AI technology versus a lack of significant increase in global GDP growth. It examines the potential impacts of AI on the economy through the perspectives of three economists:
- The Skeptics argue that the impact of AI on the macroeconomy is minimal.
- The Optimists predict that if AI develops to the point where it can replace all human jobs, it could lead to explosive GDP growth but also a collapse in average wages.
- A Third Group focuses on the trade-offs between survival risks and extended life expectancy brought about by AI, rather than just economic growth.
The article highlights the core disagreements among experts regarding three key issues:
1. How many tasks can be automated? Skeptics believe only 4.6% of tasks can be automated, while optimists argue that all tasks could theoretically be replaced.
2. Where are the boundaries for difficult tasks to automate? Skeptics think AI cannot handle complex tasks like management and decision-making, whereas optimists believe AI will eventually overcome these limitations.
3. Can AI self-iterate (improve on its own)? This is a critical question that differentiates between skeptical and optimistic views.
Why hasn’t AI accelerated GDP growth? – Like computers in the past, it’s “invisible”
The article begins by noting that despite significant improvements in AI capabilities since the launch of ChatGPT over three years ago, GDP growth in China and the United States has slowed. This is similar to what Nobel laureate Paul Solow observed in 1987: computers were widely used, but their impact on productivity was not reflected in macroeconomic data for nearly a decade. AI may currently be in a “latent phase” – it has changed our lives (e.g., using AI for writing content and research), but it has not yet penetrated all industries (such as core agricultural and manufacturing processes) to drive GDP growth.
The Skeptics: AI’s impact on the economy is negligible
A leading skeptic is Nobel laureate David Acemoglu, who predicts that AI’s cumulative contribution to total factor productivity (efficiency improvements from technological progress) will be at most 0.66% over the next decade, or an average of just 0.06% per year, which is virtually negligible. His reasoning includes:
- Task breakdown: He breaks down all jobs into numerous small tasks and determines which can be automated.
- Proportion calculation: Only 20% of tasks are affected by AI, and of these, only 23% can be economically replaced by AI, resulting in a total automation rate of 4.6%.
- Limitations of AI: Currently, AI is capable of handling easily learnable tasks (e.g., coding, customer service), but it will struggle with more complex, unstructured tasks (e.g., strategic decision-making), potentially reducing its overall impact even further.
The Optimists: AI could lead to explosive GDP growth but wage collapse
Colin Nellis from the University of Virginia suggests that if AI becomes capable of replacing all human jobs (AGI), the outcome depends on the “race between automation and capital accumulation.” If AI replaces jobs faster than capital creates new ones, wages will plummet. He outlines four scenarios:
- Base case: Tasks remain complex, and AI’s impact is limited.
- Aggressive AGI: AI replaces all tasks within five years, leading to rapid GDP growth but immediate wage collapse.
- Moderate AGI: Automation takes 20 years, with a subsequent wage decline.
- Long-tail scenario: While most tasks are automated, complex ones still require human intervention, leading to initial wage reductions followed by stabilization.
Nellis argues that the higher efficiency of AI could lead to faster GDP growth, but this could also result in rapid wage declines for ordinary workers.
The Third Group: Economic growth is not the focus; survival risks and life expectancy are more important
Jones from Stanford University emphasizes that while experts debate the economic benefits of AI, they overlook its potential dangers. She points out that AI is a double-edged sword:
- Survival risks: AI could lead to human extinction due to conflicts over values or resource depletion.
- Life expectancy gains: AI could potentially cure diseases and extend life expectancy from 100 to 200 years.
Using models, Jones calculates that if AI only increases GDP growth by 2% to 10%, the acceptable risk of extinction is 4%. However, if it extends life expectancy, this risk would rise to 25%. She argues that the value of AI lies not just in economic growth but also in improving health and longevity, suggesting humans might be willing to accept higher risks for a longer lifespan.
The core disagreements among experts
The three key issues are:
1. Automation potential: Skeptics estimate 4.6% automation, while optimists believe all tasks could be automated.
2. Limits of difficult tasks: Skeptics doubt AI’s ability to handle complex tasks, while optimists expect it to overcome these limitations.
3. AI’s self-iteration: This is a critical factor that differentiates between skeptical and optimistic views, with Jones highlighting the potential for significant survival risks.
In conclusion, while AI has already changed our lives (like mobile phones), its impact on GDP growth is currently limited. However, if its development accelerates (similar to computers in the 1980s), it could have a transformative effect on the economy, similar to the steam engine in the past. Whether this will lead to a civilization-level leap (comparable to the “fire” that sparked industrialization) is still uncertain and difficult to predict.