Sub-agents and the individuation of virtues
It seems like part of the difference between virtue-ethical views and TDT/UDT is that the decision theories consider the policies as functions isolated from their environment, whereas the virtue theorist considers whe whole agent, including functions like attention that determine the inputs to decision-functions, and weakness of will, which mediate outputs of decision functions/inputs of actions.
- Surveillance + the threat of intervention (including self-intervention) finks your dispositions, you can make this a pun with fink as in informant surveiling someone
- Continuity of Responsibility Account of Personal Identity
Remember to check attention in ML vs attention/salience in ethics esp intellectual virtue
The relevant differences can be seen by imagining choices that alter your attention/frames and akrasia, plus other choice-adjacent psychological functions.
Wrt bounded rationality/grain of truth problem, an alternative to reinforcement learning oaradigm is representation. Remember conty guy who cares about ritual also talks about memetic desire (girard?) Can model this as certain memes inhabit people better, which popukates their finite hypothesis space with attractor states. In this way the memetic environment determines what is believable. There’s also the milder phenomenon of raising priors on certain world states via stereotyping too. See more learning theory to check relevance of boundedness.