Sigmoid Curves Don't Guarantee AI Progress Will Slow Down Soon
Original: The sigmoids won't save you
Why This Matters
Challenges common assumptions about AI progress plateauing and highlights prediction biases
Scott Alexander argues that while exponential growth eventually becomes sigmoid, this doesn't mean AI capabilities will plateau soon. Historical examples like solar power deployment and UN birth rate projections show experts consistently mispredict when exponential trends will flatten.
The article examines the common argument that AI progress must slow because 'all exponentials eventually become sigmoids.' Alexander presents a 'Sigmoid Misidentification Hall of Fame' showing repeated prediction failures: UN birthrate projections consistently overestimated when declining rates would flatten, World Energy Organization repeatedly underestimated solar power deployment growth, and a 2026 Wharton paper incorrectly predicted AI capabilities would plateau. The author notes that while sigmoid curves are inevitable due to physical limits, they don't necessarily occur when analysts expect. Historical examples like airspeed records show technology progresses through multiple generations before hitting fundamental limits. The key insight is that understanding the underlying process generating a trend is more reliable than assuming exponential growth will soon become sigmoid.