NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·1h ago
Data Science
The Gradient·Feb 18·12 min read
After Orthogonality: Virtue-Ethical Agency and AI Alignment
TL;DR
Current AI alignment approaches assume agents need explicit goals, but this creates mismatches with how humans actually reason through interconnected practices rather than isolated objectives. The author proposes eudaimonic rationality—a deliberation structure based on virtue ethics and human flourishing—as an alternative framework for AI alignment.
✦ Why It Matters
Engineers can design AI systems with deliberation structures matching human reasoning, improving transparency, corrigibility, and safety without brittle goal specifications.
Key Takeaways
- Align AI actions with human practices for better collaboration.
- Evaluate AI behavior based on eudaimonic principles.
- Focus on interconnected actions rather than isolated goals.
- Address AI safety challenges through eudaimonic rationality.
How It Works
Eudaimonic rationality integrates actions into a network of practices, where each action contributes to the overall excellence of the practice, similar to how notes create a melody.