Believe it or not, I don’t actually like to be all doom and gloom about new technologies like AI—it’s just that’s what the world keeps handing me. So I’ll be damned if I don’t take the time to report on something actually positive about it when I can, and I’m hesitantly optimistic that Microsoft’s newly-published code of conduct for AI training and operation is one such opportunity.

The overall goal is the same as that which Microsoft has proclaimed before: “humanist superintelligence (HSI).” Now, though, it’s expanded much more on what this means. Though the company notes the current code is a work in progress and is opening its doors for feedback.

In a summary list explaining the code of conduct, Microsoft lists ten guidelines:

1. “People matter more than AI. The whole document in 5 words.
2. The idea of model welfare is wrong. AIs should not have rights or legal personhood.
3. An MAI Model should not do anything to meaningfully violate this Code of Conduct.
4. If it’s finish the job or break the Code, it fails the job.
5. We’re not racing to build a superintelligence that can slip its own leash.
6. Interruptible, correctable, shut-down-able. If it isn’t, we don’t ship it.
7. No neuralese. If humans can’t understand it, humans can’t oversee it.
8. Our AI should make you sharper, not dependent.
9. Pluralism, yes. Moral relativism, no.
10. We’re as clear about what our AI must never do as about what it will do.”

All of this seems great to me, and there are some things that I’m especially happy to see. Specifically, point number two: “The idea of model welfare is wrong. AIs should not have rights or legal personhood.”

This isn’t armchair philosophy: there are actual people (potentially influential ones at that) out there arguing that AI should be ascribed moral rights and so on. This is dangerous because if we start ascribing rights to AI then we may put them on an even footing with us and potentially start thinking we have to do things for ‘them’ that are to our detriment. Such as keeping the whole AI industry afloat despite any negative effects to real people that are actually conscious.

Speaking of which, Microsoft also states that AI “is not conscious and should not be designed to imitate consciousness.” Bingo. I couldn’t agree more.

I have some hang-ups about ‘humanism’ as it is usually defined and practiced today, given that in addition to the baseline human-focused perspective it is also often situated alongside superfluous and misguided doctrines, such as in scientism. Here, though, Microsoft at least seems to be referring to humanism in the simple sense of ensuring that AI is designed to be human-first, putting human control, values, goals, and safety at the forefront.

One potential cause of confusion that I noticed, however, concerns the focus on “plural values”:

“Humanity’s greatest strength is its variety and diversity of cultures, ideas, and values. We believe that AI should support a wide range of experiences, worldviews, and beliefs. Humanist AI is built to encompass this diversity of traditions, contexts, and forms of flourishing. Pluralism does not mean neutrality toward harm or that anything goes. Humanism is rooted in broad, inclusive human values such as human dignity, safety, autonomy, and fundamental human rights.”

Sounds good, right? But as someone with a background in philosophy who is privy to the kinds of debates that can occur when spending time thoroughly interrogating ethics—debates that likely occur even in the boards of AI ethics committees and so on—I think we need to be aware what a principled problem this poses.

It’s not straightforwardly clear that it is possible to both encompass a plurality of values and simultaneously guard against harm. Microsoft makes clear that it’s talking about “broad, inclusive human values” such as fundamental human rights, but what about when the rubber meets the road in specific cases where values conflict?

Here is probably not the place, nor is now the time, to discuss specific and likely contentious examples, but the point is that we are always evaluating things from a particular perspective, and as such will always be rejecting competing values. It’s a noble goal to attempt to account for diverse values while preventing harm and rejecting moral relativism, but exactly what this means in practice—what constitutes harm and which values are permissible—will be precisely the question.

I suppose in the rest of this code of practice we at least now have some idea of where those particular value flags are being planted. Assuming isn’t all just talk, of course.