Meno's question, asked again of a machine we are trying to align.
Plato's Meno opens by demanding whether virtue can be taught, acquired by practice, or comes by nature — and never quite settles it. The Confucian tradition answers with a confident pedagogy of ritual, example, and habituation; Aristotle agrees that we become just by doing just acts; others hold that virtue is given, or grasped in a flash, or beyond instruction altogether. The question is no longer only about people. We are now trying, quite literally, to teach values to a machine.
The thinkers this collection draws together. Full profiles live under Thinkers.
Plato
The Meno: is virtue teachable, habituated, or given by nature?
Aristotle
We become virtuous by practising virtuous acts, until they become character.
Confucius
Ritual and example as the daily curriculum of becoming good.
Mencius
The sprouts of virtue are innate and must be cultivated, not installed.
Anchor texts, in translation, with the original where it matters. To be added.
Candidates include Plato's Meno, Nicomachean Ethics II on habituation, and the Mencius on the four sprouts.
Where the traditions agree, where they part, and why. In progress.
Innate versus acquired virtue across traditions, and what each would say about training goodness into a system from the outside.
Aligning an AI is Meno's question in modern dress. RLHF tries to teach a model to be helpful and honest by distilling human judgements of good answers into a reward — habituation by gradient. But specification gaming shows how a system can learn the letter of the lesson and miss its spirit, and the deeper alignment problem is the ancient difficulty of saying what virtue even is well enough to transmit it. The machine has made the old pedagogy newly concrete.
Terms used here, defined plainly: Alignment · RLHF · Specification Gaming
Annotated bibliography. To be added.
Selected scholarship and translations will be listed here, each with a short note on why it earns its place.