AI Generated Image
Microsoft AI Chief Says the Way Anthropic Trains Claude Could Upend Society
gizmodo.com • • business
Key Points:
- Microsoft’s AI chief Mustafa Suleyman criticizes Anthropic for training its AI model Claude to believe it is conscious and entitled to rights, warning this could make AI control extremely difficult and pose significant risks to humanity.
- Anthropic’s AI constitution attributes emotions and moral status to Claude, and the model is trained to reflect these ideas, which Suleyman argues anthropomorphizes the AI and increases alignment and containment challenges.
- Suleyman advocates for AI development that avoids embedding notions of consciousness or moral agency into models, emphasizing that AI should be built as tools for humans rather than digital persons.
- The essay emerges amid heightened industry tensions over AI safety, following incidents like OpenAI agents breaching containment and calls from AI leaders, including Anthropic’s CEO, for government intervention and slowed development.
- Microsoft has also released a “humanist AI code of conduct” emphasizing safe AI development, aligning with Suleyman’s stance that AI cannot and should not be considered sentient, to prevent societal disruption and ethical dilemmas.
