Effective Equilibrium · velocity with vigilance
The Acceleration Paradox
How Effective Equilibrium resolves the conflict between speed and safety.
Not paralysis. Not recklessness. A control theory for civilization.
Anthropic now warns that AI is beginning to improve itself faster than we can review it, and asks the field's open question: “what should we do?” This book is the answer that is neither pause nor race. See how ›
The one idea
Every civilization runs two clocks: a clock of capability and a clock of control. Put the gap in one number. Keep R below 1, or you have already lost control, whatever the dashboard says.
Measure your own control ratio
Use any unit, as long as H and L use the same one. Runs entirely in your browser; nothing is stored.
Grow the denominator
Five gears, one number
The equilibrium law: danger is how fast it changes divided by how fast we can correct. You lower it by slowing the numerator, or by growing the denominator with five levers.
Governance
Who can say stop, how fast, and on what authority.
Equity
Who bears the risk and who holds the upside. Misalignment hides failure.
Aligned incentives
The reward for caution must beat the reward for speed.
Resilience
What survives the failure you did not predict: rollback, kill switches, margin.
Steering
Can you change direction after launch, or only watch.
EA vs e/acc vs Effective Equilibrium
Three answers to one tension
EA
“pause”
- Slow or halt to avoid catastrophe.
- Watches the clock of capability and treats it as the threat.
- Forfeits the upside; a pause is hard to coordinate.
e/acc
“race”
- Accelerate; progress solves its own problems.
- Watches the same clock and treats it as the prize.
- Assumes correction keeps pace. It rarely does.
Effective Equilibrium
“velocity with vigilance”
- Go as fast as your ability to correct allows.
- Watches the gap between both clocks, where danger lives.
- Grows the denominator. Keeps the upside and the brakes.
From doctrine to infrastructure
A worked answer to “what should we do?”
When a frontier lab like Anthropic asks how to stay in control of recursive self-improvement, its own answer points to one thing: a slowdown has to be verifiable, or no rival will trust it enough to use it. That is a denominator move. The book's flagship proposal builds exactly that.
Governance
Machine-readable commitments on a neutral platform no single lab or state controls.
Steering
A dormant slowdown you can actually switch on, adopted in peacetime for everyday reasons.
Resilience
Verification that proves bounds, not contents, so trust survives even between rivals.
The Verifiable Compute Commons makes a coordinated slowdown credible: each lab verifies the others are in compliance without exposing models or data, so a defector is seen, not merely asked to stop. It holds R = L / H below 1 by making correction switchable instead of hypothetical.
Free, private, client-side
The Equilibrium Toolkit
Score any project, product, policy, or model on the five gears and get a verdict: accelerate, proceed with conditions, redesign, or stop. The book's veto rules are built in. Nothing is sent anywhere, and nothing is stored.
Free downloads
Take the tools with you
The Equilibrium Workbook
Worksheets to run the Scorecard and the half-life test on your own work.
Get the book
Read it three ways
Paperback (6×9, color interior, 253 pp), hardcover, and Kindle. The revised and expanded edition.
Already read it? Leave a review. It does more for a book than any ad.