Active inference
The model predicts what its sensors register, then adapts to the difference. Sensing, Adapting, and Acting are the same computation.
The hard part
Generalizable Model
One model that works across multiple scenarios.
Confidence Driven
Models are intrinsically motivated to minimize both surprise and uncertainty.
Goal Driven
Outcomes encoded as probability distributions in the model.
Edge based
Up to 1 kHz on local compute. No cloud in the control path, nothing leaving the site.
“The technology that underwrites ThoughtForge is beautifully simple and almost magically powerful. Its power is inherited from a deep commitment to first principles — originating in cybernetics — and evincing the fundamentals of self-organisation: by engineering fast, free energy minimising dynamics in silico, ThoughtForge can create artefacts that respond optimally in the blink of an eye. This remarkable capacity rests on exploring configuration spaces dynamically within milliseconds. It is a fundamentally different approach – to context-sensitive, optimal behaviour — that opens up new avenues in control theory and a myriad of applications."
