Plans are Not Promises
A viral rationalist essay proposes using financial bonds to make AI development promises concrete and accountable.
A viral essay by Drake Morrison on the rationalist community site LessWrong is reframing how tech professionals think about commitments, especially in the high-stakes world of AI development. The piece, titled 'Plans are Not Promises,' argues that plans are statements of intention and worldview that invite collaboration but come with an implicit caveat for change. Promises, however, carry the weight of expectation and allow others to invest resources in a specific future. When a promise is broken, that investment is lost, damaging trust. The essay uses the analogy of a contract, where breaking it incurs legal consequences to recompense the injured party.
Morrison's key innovation is proposing 'bonds' as a practical tool to make promises concrete. He gives the example of lending a $180 board game: the borrower gives the lender $180 as a bond, which is returned only when the game is. This makes the cost of breaking the promise (losing the game) legible and pre-agreed. The lender can then fully 'invest' in the world where they get the game back, assured of compensation if it fails. This concept is gaining traction as a model for AI governance, where vague corporate 'plans' for safety are often mistaken for public 'promises.' Applying a bond-like mechanism could create clearer accountability for AI developers making commitments about model behavior, deployment timelines, or safety standards.
- Distinguishes plans (malleable intentions) from promises (costly commitments that enable investment).
- Proposes financial 'bonds' to make the cost of breaking a promise legible and agreed upon in advance.
- Framework is being applied to AI safety debates to hold companies accountable for public commitments.
Why It Matters
Provides a concrete mechanism to turn vague AI ethics pledges into accountable promises, building crucial trust.