Once Fact and Evaluation have represented both the attributes of the information itself and the assessment of its meaning in machine-readable form, the next question is “what rules should be applied to that state?” In I2EA, the stage that defines these rules is “Policy.”

Policy defines how an AI system or other system should respond to the results of Fact and Evaluation. For example, when an image is to be used as AI training data, a rule may permit its use only when multiple conditions are satisfied, such as copyright ownership, consent for AI training, and the absence of confidentiality restrictions.

Importantly, Policy does not directly make decisions about individual content. Rather, it defines the rules to be applied in decision-making. An individual Decision that “this image may be used” is distinct from a Policy stating that “images satisfying these conditions may be used.” Policy represents the latter in machine-readable form.

In addition to applicable conditions, a Policy may include an ID identifying the rule, an Action such as permit or deny, a Priority for resolving conflicts among rules, the Authority underlying the rule, an Effective Date, and a Version. This makes it possible to trace not only the resulting outcome, but also which rule was applied, under which authority, at what time, and in which version.

Policy is not necessarily static. Changes in laws and regulations, internal organizational rules, or contractual terms may alter the rules applicable to the same Fact and Evaluation. It is therefore important for Policy itself to retain information such as Authority, Effective Date, and Version, enabling identification of which rule was in effect at a particular point in time.

Multiple Policies may also apply to the same information. National law, regional regulation, industry standards, internal organizational rules, and contracts may apply simultaneously. Rather than integrating these into a single value judgment, I2EA maintains each rule together with its underlying basis in an explicit form.

Accordingly, Policy is the stage in which machine-readable rules define, for the state structured through Fact and Evaluation, what response should be taken and under what conditions.

At this point, information obtained from Reality has been structured as Fact, its meaning has been analyzed through Evaluation, and the applicable rules have been defined through Policy. The next question is who has the authority to apply those rules.