After Orthogonality: Virtue-Ethical Agency and AI Alignment
Challenges the dominant goal-directed AI alignment paradigm - relevant background for APS AI ethics and governance thinkers, not practitioners.
Key points
- Essay argues AI alignment should be grounded in virtue ethics and practices rather than goal-directed reasoning.
- Proposes that AI agents sharing a 'practices-based' deliberative logic with humans would better support human agency.
- Philosophical alignment research - interesting framing but no immediate policy or operational implications for APS agencies.
Summary
This essay from The Gradient challenges the orthogonality thesis - the assumption that AI systems can be built around arbitrary fixed goals - and proposes a virtue-ethical alternative. It argues that human rationality is structured around networks of practices rather than discrete goal states, and that AI agents designed to collaborate with humans should share this deliberative structure. The argument is philosophical in nature, drawing on practice theory to reframe what 'aligned' AI behaviour means. It has limited immediate policy application but may interest APS staff working on AI ethics frameworks or foundational governance principles.
Implications for Australian agencies
- Monitor APS staff working on AI ethics or responsible AI frameworks may want to monitor virtue-ethical approaches as an emerging alternative to goal-based alignment thinking.
Implications are AI-generated. Starting points, not advice.
"After Orthogonality: Virtue-Ethical Agency and AI Alignment" Source: The Gradient – Substack Published: 19 February 2026 URL: https://thegradientpub.substack.com/p/the-reasonable-effectiveness-of-virtue This essay from The Gradient challenges the orthogonality thesis - the assumption that AI systems can be built around arbitrary fixed goals - and proposes a virtue-ethical alternative. It argues that human rationality is structured around networks of practices rather than discrete goal states, and that AI agents designed to collaborate with humans should share this deliberative structure. The argument is philosophical in nature, drawing on practice theory to reframe what 'aligned' AI behaviour means. It has limited immediate policy application but may interest APS staff working on AI ethics frameworks or foundational governance principles. Implications for Australian agencies: - [Monitor] APS staff working on AI ethics or responsible AI frameworks may want to monitor virtue-ethical approaches as an emerging alternative to goal-based alignment thinking. Retrieved from SIMS, 18 May 2026.