From VeracIQ to the Compression Efficiency Principle
Connecting biological persistence, information compression, and epistemic integrity
The thread that connects all of my work, from regulatory design to epistemic alignment in machine learning, is simple: systems that survive must learn to compress truthfully.
I first developed VeracIQ as a framework for detecting epistemic gaps in AI-assisted regulatory workflows. It began not as an engineering problem but as a biological one. Living systems persist by minimizing wasted energy while preserving the information that allows prediction. I wanted to know what that meant for oversight systems: how they could discard noise, retain causal structure, and continually test their abstractions against reality.
That logic soon extended beyond regulation. Whether a cell, a mind, or a model, any system operating under finite resources faces the same triad of constraints:
- Persistence – maintaining organized states against entropy
- Prediction – anticipating environmental change to preserve those states
- Compression – representing the world efficiently enough to do both
From this triad emerges what I and my coauthor now call the Compression Efficiency Principle (CEP):
A system aligns with reality in proportion to the efficiency with which it compresses causal information.
In biological terms, energy, metabolism, and information are inseparable. Each act of learning (at any scale) is a negotiation between cost and coherence. When compression fails, organisms drift from the structures that sustain them. When it succeeds, prediction improves, entropy decreases, and survival continues.
Artificial systems mirror this process imperfectly. Machine-learning models compress data to predict outcomes, but their alignment with truth depends entirely on the fidelity of their data and the feedback loops that test their predictions. Without energetic consequence, compression can drift into illusion—efficient but ungrounded. Biology doesn’t allow that luxury.
The Compression Efficiency Principle formalizes something implicit in nature: to endure is to infer efficiently. Every surviving system, from a bacterium to a neural network, refines its internal model until it matches the causal grain of the world that sustains it.
VeracIQ, the applied framework for epistemic integrity, was an early expression of this idea. It asked how human oversight could adopt the same discipline nature does: compress to predict, and predict to persist.
The first time I noticed this pattern was while studying cell membranes. Entropy demanded diffusion, yet the cell held its gradient. Only later did I realize that the membrane isn’t a barrier but a negotiation: exchanging energy for information to stay alive. That paradox became the root of my thinking about compression and persistence.