Epistemic lensing
The Framework
Epistemic lensing is the systematic deformation of belief-updating induced by the channel between world and agent. The central insight is the distinction between ignorance (a deficit of signal) and distortion (a reshaping of the inferential path). A society suffering from ignorance needs more information. A society suffering from distortion needs differently structured channels.
Five Distortion Operators
Any mediating channel can be decomposed into five elementary operations:
Attenuation removes signal, increasing uncertainty without directional bias.
Selection passes some signals and blocks others, creating partial world-models.
Warping reframes content, producing directional bias in the posterior.
Amplification overweights certain cues, inflating their salience beyond evidential strength.
Recursion feeds channel output back into itself, creating path dependence and hysteresis.
Four Metrics
Information loss measures how much world-relevant information survives mediation.
Posterior divergence measures how far the mediated posterior bends from the benchmark.
Inferential curvature compares the sensitivity of belief-update to evidence.
Hysteresis measures residual distortion after corrective evidence arrives.
Key Finding
Ignorance and distortion are qualitatively different. Attenuation produces high information loss but low posterior divergence and zero hysteresis. Warping and recursion produce lower information loss but high posterior divergence and high hysteresis. The agent is confident, wrong, and resistant to correction.
Model Changelog
- Initial 9-parameter toy model: 6 channel operators + 3 agent parameters
- Posterior trajectory visualization over 200 timesteps with correction at t=150
- Five metrics: information loss, posterior divergence, inferential curvature, hysteresis, calibration error
- Four presets mapping to paper scenarios: attenuation, selection+warping, amplification, recursion
- Parameter sweep and sensitivity analysis across all 9 parameters