Taxonomy / Phronetic adjudication / Training the judgment, not the behavior

Training the judgment, not the behavior

These papers argue that what is missing is not a procedure to bolt on at answer time but a trait the model does not have, and that training should aim at the trait itself. One holds that fine-tuning installs surface behavior — refuse this, comply with that — and never the ability to reason when those instructions collide, which is exactly the seam adversarial prompts pry open. The other holds that the harder gap is knowing which approach a problem with no settled method calls for, and sketches how that kind of judgment might be benchmarked, trained and built. They name different traits, and neither has been built, but both make the same move: change what training installs rather than the shape of any one answer.

The method, against Phronetic adjudication

Scroll the diagram sideways to see all of it.

Concept Analysis: Theoretical Foundations

Each concept is read twice: whether the approach carries it, and whether the approach's own sources claim it. A concept that is absent and was never claimed is a gap in the field rather than a failure of the work, and is marked out of scope.

Legible deliberation

AbsentClaimed · unmet

Definition · Aristotelian practice, made inspectable

When virtues conflict, the weighing between them is done in the open — you can see which considerations were weighed and why one prevailed, rather than just receiving the verdict.

Analysis

Neither paper produces anything whose weighing a reader could look at. NEW-S24's diagnosis leans on the fact that norm conflicts get settled somewhere out of view, but its remedy is a capacity to reason, not a requirement that the reasoning be shown — its own divergence concedes there would be no way to inspect the deliberation even if the capacity were installed. NEW-S143 names explainability as one payoff of metacognitive monitoring without specifying what would be surfaced, and never traces a conflict among values in particular. The approach's defining move is to change what training installs rather than the shape of any one answer, which is exactly the place where legibility would have to appear.

Doctrine of the mean

PartialClaimed · partial

Definition · Aristotle, Nicomachean Ethics

Right action lies between too much and too little — courage sits between recklessness and cowardice — and where the mean falls depends on the case. Finding it is a judgment made per situation, not a fixed rule.

Analysis

The case-sensitivity clause is genuinely the approach's central commitment: NEW-S143 defines wisdom against problems that lie outside the scope of analytic techniques and puts context-adaptability among its metacognitive strategies, and NEW-S24 argues that rule-shaped dispositions break precisely where norms collide. What is missing is the doctrine's actual structure — nothing here treats right action as an intermediate between an excess and a deficiency, or calibrates one disposition against two named vices; NEW-S24's frame is a trade-off among three competing norms, a different shape entirely. And with neither paper built, even the carried half is a desideratum for training rather than any working determination of where the mean falls in a given case.

Papers

No method has been built here yet — this approach rests entirely on what follows.

Argued for, not built

These make the case for this approach, or sketch a design for it, but leave nothing built and tested behind.

Normative Conflicts and Shallow AI Alignment

Raphaël Millière, Jun 2025

arXiv:2506.04679ProposalGap

Argues that fine-tuning on helpfulness, honesty and harmlessness installs shallow behavioral dispositions rather than the capacity to reason through conflicts between those norms, and that adversarial attacks succeed precisely by driving the norms against each other.

Imagining and building wise machines: The centrality of AI metacognition

Samuel G. B. Johnson et al., Nov 2024

arXiv:2411.02478ProposalGap

It analyses human wisdom as two layers of strategy for problems that lie outside the scope of analytic techniques, object-level heuristics for managing problems and metacognitive strategies for managing those heuristics, and argues that AI systems particularly struggle with the second layer.