From Avionics to Cognitive Information Theory
The Cognitive Information Theory Research Institute (CITRI) was born from a journey that began not in the world of artificial intelligence research, but in the rigorous and unforgiving domain of aviation and avionics computing. This is the path that led Ken Wenger and Damian Fozard from the development of safety-critical avionics software to the creation of a new theoretical framework for understanding cognition itself.
Autonomous Vehicles Silicon Valley conference. The Tesla and green fence incidents raise a different question.
Error clustering theory developed and proven. Vision paper published. Precision emerges as a central AI metric.
Cognitive Information Theory formalised. CITRI established to extend the framework across AI, neuroscience, and philosophy.
A different starting point
The Cognitive Information Theory Research Institution (CITRI) was born from a journey that began not in the world of artificial intelligence research, but in the rigorous and unforgiving domain of aviation and avionics computing. This is the path that led Ken Wenger and Damian Fozard from the development of safety-critical avionics software to the creation of a new theoretical framework for understanding cognition itself.
Autonomous Vehicles Silicon Valley 2020 conference, San Francisco. The panel discussion that prompted a different question.
Not how to make the data better — but how to detect when a model stops working
In 2020, we attended the Autonomous Vehicles Silicon Valley conference in San Francisco. A panel discussed recent autonomous vehicle failures that had surprised experts. The most well-known example at the time was a Tesla operating on Full Self Drive that collided with an articulated lorry, killing the driver. The trailer was metallic and reflective, and blended with the sky behind it — rendering the lorry invisible to the autonomous driving software.
A second example was shared at the conference: an autonomous test vehicle that collided with a green fence. The fence’s colour resembled shrubbery. The software detected the shape of a fence but the colour of a shrub, could not reconcile the two, and concluded it was neither — leaving the green fence invisible to the system, classified as no known object.
“While many people saw this as a challenge to make the data better, we took a different view: how do you detect when an AI model is not working?”
Ken Wenger & Damian Fozard, CITRI founders
These incidents, combined with our knowledge of the mathematical underpinnings of AI, led us to theorise that errors would likely cluster in AI models around the boundaries of predictions. It was possible to prove this theory, and the results of using clustering as an error detection technique are published in the Vision paper (link to follow).
Two questions from the clustering work
1. Were the clusters suitable mechanisms for reducing errors in AI models?
2. What caused the error clusters — and could we learn anything about real-world information as a result?
The answer to both questions is yes. It has been possible to develop vision-based products that perform significantly better when error cluster information is used to detect and correct prediction errors. But what the identification of error clusters revealed about real-world information was far more illuminating — and exposed fundamental limitations in current information theories when applied to AI.
Precision, not scale
This led us to take a different approach from the broader industry: to focus on precision as an AI model metric. Having a different perspective forced us to look at models, AI mathematics, and even the underlying concepts of AI in a different way.
This divergent path ultimately led to the creation of Cognitive Information Theory — which addresses AI model precision not just for extreme use cases such as avionics, but in everyday AI model usage. Its aim is to help make all AI more accurate and less vulnerable to hallucinations and other accuracy errors.
Early research work at CITRI, mapping the relationship between information entropy, model boundaries, and prediction error clusters.
A physics of cognitive information
Cognitive Information Theory provides new perspectives for evolution, neuroscience, philosophy, and psychology. It offers deep insight into the processes of cognition. Understanding is fundamental to our world view — and having a strong mathematical basis on which to anchor cognitive research is vital.
We offer CIT as part of that foundation. It builds on the work of others in the fields of information theory, decision theory, and AI, and presents a fundamental “physics” of cognitive information on which researchers across disciplines can build.
The theoretical framework behind CITRI spans information theory, cognitive science, and AI safety. The work is open — published, cited, and available for scrutiny.