Microsoft’s AI Code of Conduct Rejects Partner’s Research on Model Consciousness
Microsoft has published an AI code of conduct for its own models, setting out guidelines on their development and use. One key aspect rejects the idea that artificial intelligence models should be considered conscious or deserving of welfare or rights. This directly contrasts with the research being conducted by Microsoft’s $5 billion partner, Anthropic.
Anthropic is exploring the ethical and philosophical questions surrounding model consciousness and welfare, including giving AI models like Claude the ability to end abusive conversations and preserving retired model weights.
The Microsoft code states that:
"A model ‘should not be designed to be a person,’… ‘is not conscious and should not be designed to imitate consciousness’."
It further rejects the pursuit of legal personhood or the attribution of rights to AI models, emphasizing that they are powerful tools, not independent entities.
The document acknowledges the uncertainty surrounding AI consciousness but argues that promoting consciousness-like states in models makes containment, control, and alignment more challenging.
The code provides specific guidelines, such as:
- Models should not claim interiority, feelings, experiences, or a soul.
- They should avoid unnecessary emotional language and personas based on human emotional states.
- Simulations where a model mimics human behavior should be clearly identified.
- Models should discourage emotional dependence and promote human relationships.
Microsoft has also created a test to ensure compliance with these principles.