The Counterargument That Must Be Made
Everything argued so far rests on a claim that deserves direct examination: that practical wisdom — phronesis — is a distinctively human capacity that AI cannot replicate.
Some serious people disagree. Their arguments are not foolish. They deserve engagement.
The strongest version of the counterargument runs like this: phronesis, at its core, is pattern recognition applied to complex situations. It is the ability to read a context, integrate multiple signals, weigh competing considerations, and produce an appropriate response. And pattern recognition at scale is precisely what large language models do extraordinarily well. If phronesis is sophisticated pattern recognition, then perhaps it is not categorically beyond AI — perhaps it is simply pattern recognition that requires more patterns, better training data, and richer contextual inputs than current models possess.
The Case From Reinforcement Learning
Researchers working on reinforcement learning from human feedback (RLHF) are, in effect, trying to teach AI systems to optimise for human approval of their outputs — not just for task completion, but for the quality of their judgment as assessed by people with relevant experience.
The results have been striking. Models trained with sophisticated feedback mechanisms do appear to develop something like calibrated judgment: they become better at recognising when they are uncertain, better at identifying edge cases where simple rules break down, better at producing outputs that experienced practitioners would endorse.
Constitutional AI — the approach developed by Anthropic and others — goes further, attempting to instil not just patterns of good output but something closer to principles that generate good output: a set of values that the model applies across novel situations rather than a lookup table of approved responses. These are genuine advances. They should not be dismissed.
The Case From Embodied Cognition
A different line of argument comes from philosophy of mind and cognitive science. Researchers studying embodied cognition have argued that human judgment is not purely cognitive — it is grounded in bodily experience, in the felt sense of situations, in the emotional and physical responses that signal meaning before conceptual understanding arrives.
On this view, phronesis is not just sophisticated pattern recognition. It is pattern recognition grounded in a kind of experience that currently has no analogue in AI systems: the experience of having stakes in the outcome, of being the one who will live with the consequences, of feeling the weight of a decision in a way that is not merely computational.
An AI system can be told that a decision has high stakes. It cannot feel the stakes. And the feeling of stakes is not decorative — it is part of what makes human practical wisdom reliable in ways that rule-following cannot replicate.
What Machines Can Imitate and What They Cannot
The most intellectually honest position is probably this: AI systems are becoming capable of imitating practical wisdom in increasingly narrow and well-specified domains. A model fine-tuned on tens of thousands of expert decisions in a particular field may produce outputs that, within that field, are indistinguishable from the outputs of a practitioner with genuine phronesis.
But imitation is not the thing itself. Machines may imitate practical wisdom. They may even outperform humans in narrow judgments within well-defined domains. But practical wisdom is not merely selecting the optimal action. It is bearing responsibility for the consequences of that action — being the one who will answer for it, who will learn from it, who will carry it forward into the next decision.
That is not a technical limitation that will be resolved by the next generation of models. It is a philosophical distinction about what judgment is and what it is for.
The human in the loop is not a stopgap until AI gets good enough. It is an irreducible feature of what it means to act with practical wisdom in a world where consequences are real.
Practical exercises
- Find an area in your work where AI now matches or exceeds your judgment on routine cases. Ask: what are the edge cases where you still outperform it, and why? That gap is where your phronesis is most visible.
- Reflect on a professional decision where you were wrong and learned from it. Write down what the experience taught you. Ask: could a system that was never wrong in the same way learn the same lesson?
- Have a conversation with a colleague about where they think AI will acquire genuine judgment versus where it will only simulate it. That disagreement is philosophically important.