Core claim: AGI (defined as capable of recursive self-improvement) won’t arrive soon because intelligence amplification is fundamentally rate-limited by irreducible bottlenecks.
The bottlenecks:
1.Human language - All human knowledge and values are encoded in inherently imprecise language that requires constant negotiation of meaning
2.Physical reality - Chemistry takes time, materials must be fabricated, chaos/quantum effects make perfect simulation impossible
3.Evaluation loops - Determining “improvement” requires either human judgment or real-world validation, both slow
4.Economic control - Someone must pay for and take responsibility for AI actions, keeping humans in the loop
The killer insight: These bottlenecks apply recursively - tools to escape the bottlenecks face the same bottlenecks. There’s no “escape velocity” where an AI can bypass these constraints through better tools, because tool creation and validation hit the same walls.
Implication: AGI development will proceed at human institutional/physical timescales (decades to centuries), not software timescales (months to years).
English
