Anthropic revises Claude constitution and hints at chatbot consciousness

Anthropic released a revised version of Claude's Constitution on Wednesday. The 80-page document provides a holistic explanation of the context in which Claude operates and the kind of entity the company would like Claude to be. It was released in conjunction with CEO Dario Amodei's appearance at the World Economic Forum in Davos. The revised version retains most of the same principles from the 2023 original but adds more nuance and detail on ethics and user safety. Anthropic first published Claude's Constitution in 2023 as part of its Constitutional AI approach. That system trains the chatbot using a specific set of ethical principles rather than human feedback. The document has four separate parts that represent the chatbot's core values. Those values are being broadly safe, being broadly ethical, being compliant with Anthropic's guidelines, and being genuinely helpful. Each section explains what the principles mean and how they affect Claude's behavior. In the safety section, the document states that Claude has been designed to avoid problems that have plagued other chatbots. It directs users to relevant emergency services or provides basic safety information in situations that involve a risk to human life. The ethical section emphasizes Claude knowing how to actually be ethical in specific contexts rather than engaging in ethical theorizing. Claude has constraints that prohibit certain conversations, such as discussions of developing a bioweapon. The helpfulness section states that Claude should consider users' immediate desires as well as their long-term well-being. It notes that Claude should identify the most plausible interpretation of what its principals want and balance those considerations. The document ends