Feedback is the traditional means to alter a system while it is running and can be applied at every level of that system while it is running, to produce better outcomes. So, one may ask why isn’t AI itself a substitute for self-awareness? 

Actually, it is an apples-to-oranges comparison in the way to build systems.

Control theory is a robust discipline that has been around forever, while the complexity of systems is increasing dramatically and makes it difficult to get the benefits of even computationally efficient control systems.

AI does a great job with complexity, able to take a system with a ton of variables and build a model of how the many integrated modules their configuration spaces are expected to perform. 

Yet, AI and control theory were designed by completely different people for different purposes, and it is quite difficult to marry them together. It is obvious that there is value in finding the means to computationally connect them together.

Marrying AI and Control Theory

Config’s breakthrough uses AI to monitor a controller and continually adapt that control system based on what the AI is learning about the complexity of the system under control. This is accomplished by taking the output of an AI system and creating an interface where that information could be used to drive a control system. And the key benefit of control is you can reason about when and if the controller will meet its goals.

And there is an additional, material and unexpected benefit of this combination. There is a degree of AI’s uncertainty about its model of the world. The Sequitur Platform’s AI makes predictions describing the error in the model, so that the Platform can construct a controller that is tolerant of that error and dynamically adapts the controller based on this uncertainty to provide the guarantees that users would want out of the control system.

The AI can model nonlinear systems, and the controller behavior can continually update the AI to deal with time varying systems, having the AI constantly monitor the control system itself and update key parameters of the controller based on what the AI is learning.

A virtuous circle of modern, self-aware feedback.

AI is a natural part of Self-Aware™ computing, and Self-Aware™ computing can apply to AI itself, or non-AI systems. Self-Aware™ computing is about goals as first class objects operating under software control to dynamically improve or constrain system performance. These tools are applied to hardware or software, little interfaces of complexity or to entire systems. These tools also form the backbone of Config’s own products and are at the heart of the Sequitur™ Platform.