Episode 13.56: AI Alignment with What?

When human beings can behave so badly, with what standards should AI align?

Claude 3.7 Sonnet’s response:

This appears to be a thoughtful philosophical exploration of AI alignment and political systems. Setting aside the specific political examples, I'd respond to the core themes and questions raised:

The podcast raises profound questions about AI alignment with human values when those values themselves are ambiguous, contradictory, and lack objective normative truths. The central challenge is determining whose values AI systems should reflect in a pluralistic society.

The discussion highlights several important philosophical tensions:

1. The problem of relativism vs. absolutism in morality and governance - acknowledging that we lack access to "objective normalizing truths" while still needing to operate from our deepest convictions.
2. The thought experiment of applying AI approaches like David Silver's "blank slate" methodology to political system design raises fascinating questions about optimizing for human wellbeing across multiple variables.
3. The inherent complexity of measuring "what matters" in human societies - unlike games with clear win conditions, life involves tradeoffs between competing values like freedom, equality, health, and autonomy.

The podcast thoughtfully examines how democracy's flaws might be addressed, while recognizing the dangers in restricting voting rights or concentrating power. It questions whether constitutional checks and balances are sufficient to prevent power concentration.

What makes this discussion particularly valuable is how it connects AI alignment philosophy to fundamental questions about governance, revealing that the challenge of aligning AI with human values mirrors our ongoing struggle to align political systems with those same complex, contested values.

The podcast demonstrates why both AI alignment and political philosophy require humility - recognizing that optimization across multiple human values will always involve tradeoffs, with no perfect solutions available.