Mind · Strategy
ProximateOptions
Inputs that are generated but currently rejected are cheap, latent options sitting in the gap between the generator's and the acceptor's languages
Definition
Mechanism
In a computational channel, a generator language produces inputs (memorable passwords, a writer's drafts, a population's mutations, a team's proposals) and an acceptor language decides which are valid (a strength rule, an editor, an environment, a manager). The rejected inputs, which drive the ComputationalEntropyRate up, are not just mistakes. They were produced by a process that already holds information, so they are proximate: small changes to the acceptor (or the generator) would turn them into valid outputs. The channel's friction therefore stores options. Narrowing the gap can be done from either side, accepting more of what is generated or generating more of what is accepted, and which side to move depends on the environment. Choosing which acceptor to improve against is a meta-option, and having several acceptors available raises MetaAccuracy.
Invariants
- Rejected inputs from an informed generator carry information and are cheap to make valid
- The options sit in the difference between two languages, generator and acceptor
- Zero friction (every input accepted) means no stored options
Risks
Failure modes
Driving rejection to zero and losing the stored options
Treating all rejected inputs as random noise
Always moving the acceptor toward the generator or always the reverse, regardless of environment
Relationships
Builds on
References
Infrastructure
Bits submitted per bit accepted in a channel that computes, where rejection is meaningful work rather than noise
Success at hitting any one of several available targets, each measured in its own code
Used by
References
An implementation that is at once one working model and a partial implementation of several others, so that some functions can be completed or removed with a small change