Failing to Understand the Exponential, Again
Source: https://schrittwieser.com/blog/exponential Author: Julian Schrittwieser (AlphaGo/AlphaZero team) Date: 2025-09-29
Summary
Julian Schrittwieser (part of the AlphaGo team) argues that humans systematically fail to comprehend exponential growth, and this failure is particularly costly when thinking about AI progress.
Key Claims
- Human cognition is linear: we project recent rates of progress forward linearly, systematically underestimating exponential growth.
- AI progress is not linear: capability jumps happen at specific thresholds (emergent abilities), which are impossible to predict from linear extrapolation.
- The “this time is different” trap: each time AI seems to plateau, skeptics conclude progress is over. This has been wrong repeatedly.
- Counterpoint: exponential growth doesn’t always continue. There are S-curves and real limits. The question is where they occur.
Concepts
- Scaling & Compute — exponential progress vs. S-curves
- Autonomous Research — recursive self-improvement could create super-exponential growth