Science
How Senua AI applies it.
The theory is public. What Senua AI adds is the engineering: learning causal states from live streams, at scale, on small processors, while the system runs.
| From the theory | In the engine |
|---|---|
| Causal states (1989, 2001) | Every category the engine learns is a causal state. From Wikidata's relations alone, with no labels, it discovered entity types matching the true types at 81% purity. |
| A hierarchy of machines (1994) | Layers of structure, each built over the states of the one below. |
| Inference with confidence (2014) | Every belief carries a confidence and its rate of change; the engine abstains when confidence is low. |
| The epsilon-transducer (2015) | The engine's layer for reasoning about intervention, partly built. |
Where it sits in the wider field
The engine's objective, to reduce the gap between what it predicts and what it observes, belongs to the same family as the free-energy account of the brain proposed by Karl Friston. Its causation is predictive, in the line of Crutchfield and Shalizi. That is a different question from the interventional causation set out by Judea Pearl, which needs experiments to answer; the transducer is how the engine moves toward it.
What remains open
Prediction, structure and belief rest on proved mathematics, and the engine implements them. Turning them into general agency is an open question for everyone working in the field, and we make no claim to have settled it.