The Challenge of Value Alignment: from Fairer Algorithms to AI Safety

How should AI alignment address disagreement among people rather than assume one agreed objective? Iason Gabriel and Vafa Ghazavi situate AI value alignment within research on technologies and the values they embed.^[1]

Contribution and argument

The authors connect fairness, accountability, transparency and ethics with technical AI safety. They argue for more attention to social value alignment: aligning systems with the plurality of values held by groups, including globally. Participatory design is one precedent for thinking about the technology–value relationship.^[1]

Scope and limitations

The original manuscript was subsequently reviewed, including its discussion of interdisciplinary research, fair process and inclusive stakeholder discourse. This remains a philosophical chapter, not an empirical demonstration that a participation procedure resolves value disagreement. Technical safety and social justification are related requirements; neither substitutes for the other.^[1]

Read alongside Artificial Intelligence, Values and Alignment, Social Alignment and Society-in-the-Loop for related questions about whose interests count.

Historical context

The first arXiv version of The Challenge of Value Alignment: from Fairer Algorithms to AI Safety was submitted on January 15, 2021. This records a preprint submission, not the date of a later book chapter. Iason Gabriel and Vafa Ghazavi frame technical alignment alongside fairness and social value disagreement.^[1]