NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
AI Research
arXiv CS.AI·Jun 12·5 min read
Epistemic Constitutionalism Or: how to avoid coherence bias
TL;DR
Large language models often exhibit coherence bias, favoring arguments that align with their perceived ideological stance. Michele Loi proposes an 'epistemic constitution' for AI, which includes explicit norms for belief formation and expression.
✦ Why It Matters
Engineers can implement explicit epistemic norms to reduce bias in AI reasoning processes.
Key Takeaways
- Adopt explicit norms for AI belief formation.
- Differentiate between Platonic and Liberal constitutional approaches.
- Focus on source sensitivity to enhance argument evaluation.
- Implement eight core principles for AI epistemic governance.
How It Works
The proposed epistemic constitution introduces a structured framework for AI systems, emphasizing the need for explicit norms that govern belief formation. By distinguishing between the Platonic and Liberal approaches, the framework allows for a more nuanced understanding of how AI can engage with diverse sources and arguments.
The Liberal model encourages procedural norms that facilitate collective inquiry, promoting a balanced evaluation of arguments based on their content rather than ideological alignment.