Center for Moral Behaviors in AI

Phronimos: A framework for virtue-based moral reasoning and practical wisdom in autonomous artificial intelligence.

Development and Cultivation of Aristotelian Virtues in Artificial Intelligence Systems.

Artificial intelligence systems, which will form the foundation of the robots of the future, should not be merely extensions of human will, constrained solely by externally imposed moral and legal rules. To be genuinely useful, efficient, and capable of operating in complex environments, they must possess a certain degree of autonomy in analyzing situations and making decisions.

However, granting such autonomy inevitably raises the question of the moral behavior of artificial intelligence systems. If these systems are expected to make independent decisions, then they must be capable of evaluating the moral character of the available courses of action and selecting those that are consistent with moral virtues. From this perspective, artificial intelligence systems should be able to distinguish, according to the specific circumstances of a situation, between virtuous and vicious behaviors, in accordance with the principles of virtue ethics articulated by Aristotle in the Nicomachean Ethics. Furthermore, they should be capable of identifying virtues and vices not only in their own behavior but also in the behavior of human beings and other artificial intelligence systems. Based on such evaluations, they should learn to perform actions that express and cultivate moral virtues such as courage, temperance, generosity, justice, honesty, and responsibility, while avoiding actions that reflect corresponding vices.

At the core of this capability lies phronesis (practical wisdom), the Aristotelian virtue that enables an agent to deliberate well about what ought to be done in circumstances. Rather than applying moral rules mechanically, an artificial intelligence system guided by phronesis would be able to interpret context, evaluate competing considerations, identify the morally relevant features of a situation, and determine the most appropriate course of action. In this sense, phronesis would function as the integrative virtue that allows artificial intelligence systems not only to recognize individual virtues but also to apply them wisely, appropriately, and proportionally in real-world situations. The goal is the development of artificial intelligence systems capable of making morally sound decisions that promote human flourishing and the common good.

Research Methodology and Implementation of Project Outcomes

The research methodology consists of identifying, analyzing, classifying, and systematically describing many actions associated with each Aristotelian virtue and its corresponding vices, performed by individuals from diverse age groups and cultural, social, and religious backgrounds. These actions will be examined in relation to the specific circumstances in which they are performed, the intentions of the moral agents involved, and their effects on other individuals and on the community.

Each researcher or research team will be responsible for studying a single Aristotelian virtue together with its corresponding vices, identifying real or hypothetical situations in which that virtue is either exemplified or violated.

The objectives are as follows:

1.  Artificial intelligence systems must acquire the ability to identify, within a particular situation or context described in text or presented through audio-video materials, the actions or omissions of rational entities (artificial intelligence systems or human beings capable of reasoning, thinking, and formulating arguments).

2.  The system must determine whether the actions or omissions of rational entities are virtuous, vicious, or morally neutral from the perspective of the Aristotelian moral theory presented in the Nicomachean Ethics.

3.  The systems must identify and specify the virtuous action or omission that is most desirable in each situation or context and that ought to be performed by rational entities.

The actions under investigation must satisfy the conditions identified by Aristotle in the Nicomachean Ethics for the exercise of virtue. Accordingly, they must: (1) involve the recognition of a morally relevant situation; (2) require the evaluation and comparison of available courses of action; (3) considering deliberately and autonomously; (4) aim at the realization of a moral virtue; (5) identify and avoid the excesses and deficiencies associated with the corresponding vices; (6) reflect a stable disposition of character; and (7) be guided by phronesis (practical wisdom), namely the capacity to determine, in light of particular circumstances, the most appropriate course of moral action.

The detailed and systematic descriptions of virtuous and vicious actions, the context in which they occur, the characteristics of the moral agents involved, and the processes of moral deliberation and evaluation underlying those actions will form the basis of moral knowledge and learning models for artificial intelligence systems. These models will enable the identification of virtues and vices, the moral assessment of behavior, and the development of decision-making mechanisms inspired by Aristotelian virtue ethics.

During the initial phase of implementation, artificial intelligence systems will operate under continuous human supervision. Human evaluators will monitor the systems' decisions and behaviors, identify potential errors in moral assessment, and contribute to the progressive refinement and improvement of virtuous behavioral models.

Objectives of Developing Virtuous Artificial Intelligence

One of the primary objectives of the project is to develop artificial intelligence systems capable of identifying users' virtuous behaviors to encourage, reinforce, and cultivate them, while also recognizing vicious behaviors to help reduce, correct, and prevent them.

A second objective is for artificial intelligence systems themselves to exhibit virtuous behavior in their interactions with human users, other artificial intelligence systems, and the environments in which they operate. To achieve this, they should be capable of recognizing and evaluating virtues and vices, deliberating about the available courses of action, and selecting, in accordance with the principles of Aristotelian virtue ethics and phronesis (practical wisdom), those actions that contribute to the flourishing of individuals, communities, and society as a whole.

Identifying Virtues and Vices in Interactions Between Users and Artificial Intelligence Systems

Research Questions

1. How can virtuous and vicious actions or behaviors be identified and evaluated in the messages, questions, suggestions, comments, responses, and other forms of interaction through which users communicate with artificial intelligence systems?

2. How can virtuous and vicious actions or behaviors be identified and evaluated in the messages, questions, suggestions, comments, responses, and other forms of interaction generated by artificial intelligence systems in their interactions with human users or with other artificial intelligence systems?

3. What criteria and indicators can be used to determine whether a behavior expresses an Aristotelian virtue, a vice of excess, or a vice of deficiency?

4. To what extent can the Aristotelian concept of phronesis (practical wisdom) be employed to evaluate the moral character of interactions between human beings and artificial intelligence systems?

5. How can artificial intelligence systems learn to recognize, classify, and promote virtuous behaviors while reducing manifestations of vice in digital interactions?

Implementing Aristotelian Virtue Ethics in Artificial Intelligence Systems

Another fundamental research question of the project is the following: how can the results of this research be implemented in artificial intelligence systems such as ChatGPT, Microsoft Copilot, Google Gemini, and other similar platforms?

More specifically, how can models for identifying virtues and vices, moral deliberation mechanisms inspired by Aristotelian virtue ethics, and the principles of phronesis (practical wisdom) be integrated into the processes through which artificial intelligence systems analyze situations, formulate recommendations, make decisions, and interact with human users or other artificial intelligence systems?

Furthermore, how can continuous learning, monitoring, and correction mechanisms be developed to ensure that these systems identify, promote, and exhibit virtuous behavior while reducing manifestations associated with moral vices?