In this article, we will briefly discuss how artificial intelligence (AI) learns rule-based texts such as articles of law. Teaching AI something is not that different from teaching a human being. The most important difference is that humans learn through intuition, while AI learns through logical modelling.

To teach a computer a rule, you first need to determine what it recognises as a "rule". A rule must be structured in such a way that it yields the correct result when it is fulfilled and the incorrect result when it is not. For example, in a rule stated as a sentence such as “The internship period must be uninterrupted” in the Law on Attorneys, AI learns to perceive “internship period” as the subject, “uninterrupted” as the condition, and the word “must be” as the expression of obligation. As more examples of this kind accumulate, the system becomes capable of distinguishing which types of sentences constitute rules.

Humans learn by seeing and testing; AI learns through examples. It is shown a legal text together with the rule derived from that text. For example, when training AI, a legal text such as "The bar president appoints a solicitor before the announcement of the internship application" is loaded into the processor together with the rule "The bar president must have appointed a solicitor before the announcement of the internship application." Hundreds of such matching examples enable the AI to grasp the rule pattern. Now, when it encounters a new rule text, it can generate a similar rule on its own. Of course, for the computer to do this correctly, it must also correctly parse the sentence structure of the language used. Which word is the subject, which is the predicate, which is the condition? Natural language processing technology is used for this purpose. AI parses the sentence into parts, identifying "bar association president" as the “subject”, “appoints” as the “verb”, and "a solicitor" as the “object”. It thus understands which elements are fixed and which may vary depending on the context. This enables the rules to be automatically rewritten.

AI does not merely extract the rules and criteria from rule-based texts; it can also learn to test them. To this end, it uses the methods and principles previously taught to it. At this stage, feedback from human trainers is crucial. As AI sees its mistakes, it learns to measure and correct the accuracy of the rules it generates. Each new piece of data used for this purpose reinforces or corrects what was previously learned.

Ultimately, the AI's learning process is about teaching logic, not information. The AI creates its own internal model by recognising the logical connections between the examples shown to it. Humans are the guides in this process; it is humans who determine which information is meaningful and which rules are applicable.