In 1987, Nobel laureate Robert Solow famously quipped: 'You can see the computer age everywhere but in the productivity statistics.' For decades, economists struggled to explain why massive investments in IT didn't immediately lead to economic growth.
The J-Curve of Technology
Today, we face a similar puzzle. Nvidia's stock is vertical, every startup is an 'AI company', and yet global productivity growth remains sluggish. Stanford's Erik Brynjolfsson argues this is a classic J-Curve effect: powerful technologies require massive complementary investments in organizational capital before they pay off.
Electricity didn't boost manufacturing until factories were redesigned from a central steam engine layout to a distributed electric motor layout. Similarly, AI won't boost GDP until we redesign the firm itself.
The Noise Hypothesis
But there is a darker theory. What if Generative AI is actually lowering productivity in the short term?
Consider the 'Bullshit Asymmetry Principle': It takes an order of magnitude more energy to refute bullshit than to produce it. LLMs have made the marginal cost of producing 'average' content near zero. We are flooding the world with emails, code, and reports that look plausible but require human verification.
If every employee uses ChatGPT to write a 10-page memo, and every manager uses ChatGPT to summarize it, have we created value? Or have we just built a perpetual motion machine of token-generation?
Measurement Error?
Finally, it's possible our metrics are broken. GDP measures the value of goods and services sold. It does not capture Consumer Surplus. If AI makes coding 50% faster, software prices might fall (deflationary technology), making the sector look smaller in nominal GDP terms, even though real utility has skyrocketed.
Conclusion
We are likely in the 'implementation lag' phase. The real productivity boom won't come from writing emails faster. It will come when AI unlocks scientific discovery (AlphaFold) or optimizes physical supply chains—areas where 'hallucinations' are bugs, not features.