Servaas Storm's INET Paper Debunks AI Bubble Claims Using Russell's Teapot Analogy

Key facts

• Servaas Storm is a Dutch economist and author working on macroeconomics, technological progress, income distribution, economic growth, finance, development, structural change, and climate change.
• The paper uses Bertrand Russell's 1952 Teapot Analogy to argue that unfalsifiable AI claims require proof from those making them.
• Storm claims AGI predictions are structurally flawed because LLMs are predictive pattern matchers, not intelligent systems.
• LLM hallucination rates are 3-8% on closed-world questions and 30-50% on complex reasoning tasks.
• AI labs face escalating training and inference costs, with subscription fees not covering operational costs.
• Anthropic data shows 40% of U.S. occupations have no meaningful AI exposure.
• Economists predict a modest 0.2 percentage point increase in U.S. productivity growth by 2030 from AI.
• The paper estimates $4.5 trillion in AI investments during 2025-2035 will likely yield negative macroeconomic returns.
• 70% of non-management workers surveyed feel anxious or overwhelmed by AI tools.
• AI-generated 'work-slop' is demotivating and drains productivity, with workers wasting eight hours per week cleaning up AI output.
• AI coding agents break previously working code in 75% of long-term maintenance tasks.
• Companies have reduced entry-level hiring, leading to a projected senior developer shortage in 5-7 years.
• Cheap open-weight LLMs outperformed Anthropic's Mythos on basic security reasoning tasks.
• Algorithmic wage discrimination uses surveillance data to set lower wages for indebted workers.
• The paper was originally published on Naked Capitalism.