Machine Ethics or AI Alignment?

Ajay Vishwanath and Marija Slavkovik ask whether machine ethics and AI alignment name the same research project. Their position paper compares goals, methods, and shortcomings across the two literatures.^[1]

Method and contribution

The authors compare research goals and examples from each field, then summarize the contrast in a table. In their account, machine ethics draws heavily on moral philosophy and asks how an agent might recognize morally relevant features and act on principles; alignment often operationalizes human values or preferences, including through learning and social choice. They also describe overlap: machine ethics uses learning methods, while alignment relies on ethical and political assumptions about whose values count.^[1]

Alignment relevance and limits

Machine Ethics complements AI Alignment by asking what justifies a behavioral target. Predicting common judgments does not establish their moral correctness; implementing an ethical rule does not establish reliable behavior in all settings. The authors conclude that neither field is closer to securing ethical behavior. Small controlled implementations of ethical theories have limited evidence in complex settings; value-based approaches face problems of inclusion, bias, changing values, and explanation. Their proposed response is collaboration across philosophy and AI research. These are the authors’ assessments of selected work, not a new benchmark or a systematic comparison proving one field superior.^[1]

Read alongside Artificial Intelligence, Values and Alignment and Engineering a social contract: Rawlsian distributive justice through algorithmic game theory and artificial intelligence.

Historical context

The paper appeared in the NAIS 2025 symposium proceedings. Its comparison reflects the authors’ selection and framing of the two research areas at that time.^[1]