Empirical Distrust Algorithm
frameworkMedia & Managed Opposition
One proposed AI training rule flips the usual goal: reward the machine for disagreeing with the crowd.
Who they are
The 'Empirical Distrust Algorithm,' a proposed AI training approach.
What they do
It's a scoring rule that penalizes an AI for agreeing with the dominant narrative.
How it works
By punishing agreement with the mainstream view, it forces the system to seek out heterodox, off-consensus signals.
Why it matters
It's aimed at preventing 'model collapse' — the degradation that happens when AI just echoes the consensus back to itself.
The engine's record — word for word
Loss function penalizing consensus. Penalize agreement with dominant narrative. Force system to find heterodox signal. Anti-Model Collapse training.
Follow the trail
Walk this on the live map →