Hypothesis Subspace

How does the inherited field theory apply?

A large part of [[deontic-arrays]] as an inherited part of ideological inference engines was the idea of a field defined using discrete structures across a state space. In its main formulation, IIE seem to lack that otherwise promising focus. How can it be recovered?

First, one way to recover this connection would be to consider the entailment verifier as yielding a continuous value denoting compatibility of the knowledge base with the verification target. This would be opposed to a GOFAI-like boolean output of this component. This fuzzier output would allow us to define a landscape of compatibility with the knowledge base across action space, and hence provide gradients for guiding model state towards compatible regions and away from incompatible ones.

However, this is not immediately intuitive considering the expectation that only one verification target is picked and undergoes verification at a time. In a sense, a whole space of actions would need to undergo verification in order to help specify that gradient. Maybe one could investigate KB ⊨ A where A would be a space of actions, rather than a single one. And instead of yielding a single fuzzy estimate of compatibility, it would yield this space of compatibilities as an output.

Additionally, the direct connection between discrete items contained in the knowledge base and those compatibility gradients is now less visible. When visualizing magnet-like structures for each KB item, it's intuitive to think of how each individual item is influencing the model. However, the grouping of items in a global KB before yielding compatibility gradients through entailment obscures this connection somewhat. The individual contributions to shaping the field are aggregated before having a say.

How does the inherited field theory apply?