inverse variance of a prediction error — how confident an agent is about a particular signal
in active inference: precision determines which prediction errors get amplified and which get suppressed. high precision = this signal is reliable, weight it heavily. low precision = this signal is noisy, down-weight it
attention in the Fristonian framework IS precision-weighting: attending to something means increasing the gain on prediction errors from that source
in cyber
precision maps to token staking in the cybergraph:
- high stake on a cyberlink = high precision = the neuron is confident this connection is real
- low stake = low precision = uncertain, tentative link
- staking amplifies the signal in the tri-kernel computation — precisely the gain modulation that precision provides in brains
this makes precision an economic signal: backing beliefs with value. gaming precision (staking heavily on false connections) is punished by slashing — skin in the game
the precision-attention equivalence
| predictive coding | cyber |
|---|---|
| increase precision on a sensory channel | stake more tokens on a particle or cyberlink |
| suppress low-precision errors | low-stake links contribute less to π |
| attention = selective precision | focus = stake-weighted attention distribution |
see active inference for the framework. see free energy principle for the theory. see predictive coding for the neural architecture