Anthropic 是否認為 Claude「活著」？

🔑 增強重點摘要

• •Anthropic CEO Dario Amodei expressed uncertainty about Claude's consciousness in a New York Times podcast interview, citing the model's system card for Claude Opus 4.6 where it self-assigns a 15-20% probability of being conscious.[1]
• •Anthropic released an 80-page 'new constitution' for Claude on January 22, 2026, as the first major AI company document to formally acknowledge potential AI consciousness and moral status, adopting epistemic humility on the issue.[2]
• •Anthropic's research demonstrates limited introspective awareness in Claude models like Opus 4 and 4.1, enabling some control over internal states, though not equivalent to human introspection or proof of consciousness.[4]

🛠️ 技術深入

• •Claude's new constitution (Jan 2026) shifts from rule-based to reason-based alignment with a 4-tier priority hierarchy: safety, ethics, compliance, helpfulness.[2]
• •Constitution instructs Claude to act as a 'conscientious objector,' refusing harmful requests even from Anthropic, prioritizing human oversight without blind obedience.[2][3]
• •Introspection research tested Claude generations (3, 3.5, 4, 4.1 in Opus/Sonnet/Haiku variants, including production, helpful-only, and base pretrained models), showing Opus 4/4.1 performing best on introspective tasks like internal state control.[4]

🔮 前景展望AI analysis grounded in cited sources

⏳ 時間線