Microsoft’s Mustafa Suleyman posted “The Humanist AI Code of Conduct.” It comes amidst hysteria over the “doomsday” concerns raised by 27-year-old Anthropic researcher Jacob Coxon last week.
The Code includes 4 Core Objectives:
- Human control and reliable safety. Models stay under meaningful human oversight. No operator or user can switch off the safety limits.
- AI is artificial. Models shouldn’t act conscious or claim to have feelings. The code rejects legal personhood and “welfare” claims for AI.
- Human flourishing. AI should raise human ability, independence and judgment, and support human relationships rather than replace them.
- Plural values. Models should respect many cultures and beliefs, grounded in dignity, autonomy and basic rights. Operators and users can adjust them within the safety limits.
Subject to public input and revision
The Code will be subject to revision and public input:
“As we revise the Code of Conduct, we will be on the lookout for several things. First, we’ll seek to identify and document unintended outcomes and address them. Second, we’ll aim to cover areas where the Code of Conduct fails to address emerging realities. Third, we will proactively look for new research that questions the assumptions behind an Objective, a Safety Constraint, or a Guideline. Fourth, we will examine persistent failures to steer model behavior despite mature evaluations and find ways to fix them. Other approaches and changes may be needed over time, and more work may be required in specialist domains (like government, healthcare or financials) alongside developments to account for emergent technical risks (like multi-agent considerations or recursive self-improvement).
“That said, wherever possible, we will prioritize stability. The commitments set out here should be durable. While MAI Models’ defaults may change with experience, and safety with new evidence of risks, there should only be changes to the core Objectives for extremely good reason. AI’s opportunity to deliver transformational benefits for humanity is real. And yet, if mis-managed, it will also pose grave dangers, including new and as-yet poorly understood frontier risks. Models, and the process by which they are developed, deserve public scrutiny and engagement. It’s in this spirit that we’re sharing this document publicly to welcome ongoing review and feedback.”
GIVE FEEDBACK TO MICROSOFT HERE
You can submit feedback and comments to Microsoft at this link.