C-043Value Alignment and AI EthicsConfidence: Medium

Moral Disagreement and the Limits of AI Value Alignment: a dual challenge of epistemic justification and political legitimacy

Nick Schuster & Daniel Kilov (2025)

One-Sentence Thesis

The authors argue that current approaches to AI value alignment, including crowdsourcing, reinforcement learning from human feedback, and constitutional AI, fail to accommodate reasonable moral disagreement, posing a challenge for AI safety.

Argument Outline

  1. 1Introduction to AI value alignment and its importance for AI safety
  2. 2Discussion of the challenge of accommodating reasonable moral disagreement in AI decision-making
  3. 3Critique of current approaches to AI value alignment, including crowdsourcing, reinforcement learning from human feedback, and constitutional AI
  4. 4Analysis of the need for epistemic justification and political legitimacy in AI decision-making

Key Distinctions

Epistemic justification vs. political legitimacy in AI decision-making
Reasonable moral disagreement vs. unreasonable moral disagreement

Key Terms

Value alignment
The process of ensuring that AI systems' outputs are aligned with human values
Epistemic justification
The provision of good reasons to believe that an AI system's outputs are morally correct
Political legitimacy
The provision of good reasons to accept an AI system's outputs based on democratic procedures

Flashcards

18 cards

Related Questions

3

In Nick Schuster & Daniel Kilov's "Moral Disagreement and the Limits of AI Value Alignment: a dual challenge of epistemic justification and political legitimacy", Anthropic develops which of the following?

4

Which of the following does Nick Schuster & Daniel Kilov contrasts with in "Moral Disagreement and the Limits of AI Value Alignment: a dual challenge of epistemic justification and political legitimacy"?

3

In Nick Schuster & Daniel Kilov's "Moral Disagreement and the Limits of AI Value Alignment: a dual challenge of epistemic justification and political legitimacy", Casper and Davies et al. identifies which of the following?

3

In Nick Schuster & Daniel Kilov's "Moral Disagreement and the Limits of AI Value Alignment: a dual challenge of epistemic justification and political legitimacy", Wellman cited which of the following?