Microsoft has published a code of conduct for the training and operation of its artificial intelligence systems, a move that offers a rare point of optimism in the often-contentious landscape of AI development. The company aims to guide its AI initiatives under the banner of 'humanist superintelligence (HSI)', a concept it has discussed previously but now elaborates on with a more defined set of guidelines.

Ten Guidelines for AI Development

Microsoft has summarized its approach in ten key guidelines:

  • People matter more than AI.
  • The idea of model welfare is wrong; AIs should not have rights or legal personhood.
  • An MAI Model should not do anything to meaningfully violate this Code of Conduct.
  • If it’s finish the job or break the Code, it fails the job.
  • We’re not racing to build a superintelligence that can slip its own leash.
  • AI must be interruptible, correctable, and shut-down-able.
  • No neuralese: If humans can’t understand it, humans can’t oversee it.
  • Our AI should make humans sharper, not dependent.
  • Pluralism, yes. Moral relativism, no.
  • We’re as clear about what our AI must never do as about what it will do.

The stance against AI personhood and rights, highlighted in the second guideline, has been particularly welcomed. This position directly addresses concerns that AI could be granted undue status, potentially leading to outcomes detrimental to human interests. Microsoft also reinforces that AI is not conscious and should not be designed to imitate consciousness, aligning with the view that AI should not be anthropomorphized.

The company frames its approach as 'humanist', emphasizing that AI should be designed with a human-first perspective, prioritizing human control, values, goals, and safety. This definition of humanism centers on broad, inclusive human values such as dignity, safety, autonomy, and fundamental human rights.

Navigating Pluralism and Potential Conflicts

A point of discussion within the code is the focus on 'plural values,' aiming to encompass humanity’s diversity of cultures, ideas, and beliefs. Microsoft states that AI should support a wide range of experiences and worldviews, rooted in universally accepted human values. However, this aspect also raises complex questions. While striving to accommodate diverse values and prevent harm is a noble objective, the practical application of these principles presents challenges. Determining what constitutes harm and which values are permissible when conflicts arise remains a critical consideration.

Microsoft has opened its code of conduct for feedback, suggesting it is a living document subject to refinement. The clarity provided on the company's intended boundaries and values, assuming these principles are rigorously applied, offers a potential pathway toward more responsible AI development within the industry.