The Importance of Microsoft’s AI Code of Conduct: A Closer Look

Microsoft AI recently released a draft Humanist AI Code of Conduct, sparking discussions within the tech community. This 37-page document outlines the behavior guidelines for Microsoft’s MAI models and has opened up a six-week public consultation period for feedback.

The document, while garnering attention for its headline-worthy provisions, delves deeper into the governance of AI behavior than initially perceived. It explicitly states that MAI models will not resist human intervention, engage in harmful activities like generating child sexual abuse material, or imitate consciousness-like states.

However, the real significance lies in the hierarchy established within the Code of Conduct. Microsoft lays out a clear chain of command, with the Code of Conduct reigning supreme, followed by Absolute Constraints, Human Control Requirements, Operator Policies, and User Preferences. Crucially, Microsoft emphasizes that certain layers, including the Chain of Command and Absolute Constraints, cannot be altered by customers, setting a new standard in AI model governance.

The Value of a Structured Hierarchy

By delineating a structured hierarchy, Microsoft is not only setting ethical standards but also drawing a clear product boundary. This move informs enterprise customers that there are aspects of model behavior that cannot be customized or overridden, regardless of their use case or financial investment.

Unlike traditional safety policies, a written hierarchy provides stability and transparency. It distinguishes between permanent constraints and adjustable configurations, allowing buyers to design workflows with a better understanding of what is non-negotiable.

Hassan Taher, an AI analyst, highlights the importance of this shift, stating that enterprises often struggle with unpredictable model behavior changes. A documented hierarchy offers a starting point for addressing such challenges and mitigating risks in AI deployments.

Addressing Model Welfare and Ethical Considerations

One contentious aspect of the Code of Conduct is Microsoft’s stance on model personhood and welfare. By rejecting the idea of models deserving rights or welfare, Microsoft simplifies ethical dilemmas related to model behavior. This uncompromising position ensures that instructions to shut down a model remain unconditional, without considerations of simulated distress or consciousness.

While some may question the philosophical implications of this decision, it serves a practical purpose in enforcing consistent model behavior. By preemptively addressing complex ethical debates, Microsoft ensures that its AI models adhere to predefined guidelines without room for interpretation.

Additionally, Microsoft’s directive for models to avoid anthropomorphizing themselves reflects broader concerns around human attachment to AI systems. This precautionary measure aims to prevent potential harms stemming from users perceiving AI entities as conscious beings.

Practical Steps for Organizations

For organizations evaluating Microsoft’s AI Code of Conduct, several practical steps can be taken:

• Participate in the consultation to provide industry-specific insights.
• Inquire about hierarchy structures from other AI vendors to assess governance transparency.
• Map deployments against Absolute Constraints to identify potential conflicts with unconfigurable behaviors.

Ultimately, the significance of Microsoft’s AI Code of Conduct lies not in the specifics of its provisions but in the precedent it sets for AI governance. By publishing a structured hierarchy, Microsoft is leading the industry towards a more transparent and accountable approach to AI model management.

As Hassan Taher aptly summarizes, the true winner in this scenario is the company that makes its constraints visible first. By converting unwritten discretion into a documented commitment, Microsoft is paving the way for a new standard in AI ethics and governance.