A timeless paradigm for the questions that define modern AI.
A conceptual and ethical framework for the design and governance of agentic AI systems, drawing on the thought of Thomas Aquinas.
The starting point is an architecture: the Thomistic account of the human intellect in its speculative and practical uses, together with the five intellectual virtues, the doctrine of the vestigium, and the demarcation between intellectual and moral virtues. From that architecture, a usable classification of AI components and a taxonomy of failures fall out as consequences.
Thomas distinguishes two uses of the same intellect: one considers what is the case; the other considers what is to be made or to be done. They share a subject but not a measure.
Considers truth as such: what cannot be otherwise, and what happens to be, known as such. Its good work is consideration; its measure is conformity to the thing.
Conformitas intellectus ad rem.
Considers what is to be made (a factum) or what is to be done (an actus). Its measure is conformity to right appetite: the good of the work, or the good of the agent in the acting.
Conformitas ad appetitum rectum.
Five habits perfect the intellect in its proper work: three in the speculative order, two in the practical. They stand in an explicit ordering, and none is reducible to another.
Apprehension of what is known per se. The starting point of demonstration.
Conclusions drawn from principles within a determinate genus. There are many sciences.
Wisdom orders all sciences beneath itself and judges principles and conclusions alike.
Right reason about things to be made. Its good lies in the thing made, not in the maker.
Right reason about things to be done. Its good lies in the agent. Presupposes moral virtue.
Every effect represents its cause somehow, though in two different ways. This distinction is load-bearing: it names, precisely, what a trained model carries of the human operations recorded in its corpus.
Carries the that of the cause, that something passed by, without its form. Smoke represents fire: we know fire was there. We do not find fire in the smoke.
Carries the what of the cause: its form, shared. Fire generated by fire bears the form of the generating fire. A statue of Mercury bears the form of what it depicts.
From the architecture of the intellect and the vestigium premise, a usable classification falls out. Each AI component emulates one of three orders, and is rightly judged only by its proper measure. Confusing the measures is a common source of evaluation failure.
· Tap a vertex of the figure to lift the corresponding order.
Answering, classifying, summarising, inferring from stated premises. The output is a judgment about what is the case.
The output is an outward factum: a produced artefact whose goodness lies in fitness for its intended use.
The output is an action: a step executed in a shared environment, whose rectitude cannot be discharged by truth alone or by the fitness of an artefact.
Contemporary AI standards flag the strain in their own definitions. They restrict terms by footnote. The framework supplies a vocabulary in which the necessary distinctions are already native.
"Knowledge in the AI domain does not imply a cognitive capability… In particular, knowledge does not imply the cognitive act of understanding." ISO/IEC 22989:2023, 3.1.21, Note 1
The standard addresses the problem by restriction. The ordinary meaning of knowledge carries commitments the engineered object does not satisfy.
Three distinct habits of the speculative intellect: intellectus (apprehension of principles), scientia (demonstrative knowledge), and sapientia (knowledge by first causes), with an explicit ordering among them. ST I-II Q.57 a.2
The standard's caveat corresponds to a distinction Thomas already makes. The framework names what the standard has to restrict.
The framework PDF in three languages. Each translation is done by a human translator, not by machine rendering.