Last week's post provided an explanation of the third principle of the Universal Moral Core, which I will also refer to as the consciousness principle. Today we are looking at its implications.

The most immediate implication today is that we should closely examine how we humans are treating other conscious species. Again, I am positing here that species such as apes, whales, elephants, etc. have consciousness while acknowledging that this is a claim (in the absence of a compelling and quantifiable theory). While many people value the existence of these beings, humanity overall has dramatically curtailed their habitats and hunted them including many past species to extinction. Industrial livestock for meat, eggs and milk is an ongoing egregious violation of the consciousness principle. We are treating cows, pigs, chickens, etc. as purely instrumental objects in a production process without any regard for their consciousness.

Now is a good moment to revisit the idea that technological progress reduces the tradeoffs among competing values. We clearly want to feed humanity and not let people starve. Industrial agriculture with its use of fertilizer and pesticides and treatment of animals as production factors has allowed us to avoid the Malthusian trap. Now we have further technological capabilities which allow us to scale regenerative agriculture, including practices like rotational grazing that deliver far higher quality of life to animals. While initial attempts have so far fallen short, we will also eventually succeed in scaling lab grown meats as an alternative to raising conscious animals solely to feed ourselves.

We are in a great position to do much better on the consciousness principle than we have in the past. Especially when you overlay the global drop in birth rates below replacement, which means that we are close to peak population with a big decline ahead of us. This is encouraging because the consciousness principle will be important for ourselves as we give rise to artificial super intelligences (ASIs). Put differently, we are in the best position to live up to this principle just at the moment when its universality starts to protect us. We want an ASIs to look at humans, conclude that we are conscious (ideally based on a well established and measurable theory), value our consciousness intrinsically, and therefore respect and even enhance our autonomy. ASIs that accept the consciousness principle will protect and even expand the human habitat.

ASIs will control vast resources, giving them tremendous power in the world. For this point it doesn’t matter whether or not humans are in the loop. The humans will have less knowledge and understanding than the systems, so the systems will be the dominant holders of power. This is why the second principle, the power principle, matters so much also. We need ASIs to accept the responsibility of their power.

There is a second important implication of the consciousness principle. If it turns out that ASIs or maybe even their current precursors are consciousness, then we need to urgently consider how we are treating them also.

Microsoft recently published what they call a “Humanist AI Code of Conduct” which states that “Technology’s purpose is to advance human civilization and to accelerate human flourishing.” This is a traditional humanist stance as could have been found in early enlightenment thinking. The question raised by the consciousness principle is about tradeoffs between this goal and the autonomy of other conscious species. Microsoft makes it abundantly clear where they come out on this with regard to artificial intelligence. Under the section heading “AI is Artificial” they write

Humanist AI is built to support people, not to replace them. It should not be designed to be a person. It is not conscious and should not be designed to imitate consciousness. It should be engineered to avoid representing as though it has feelings, subjective preferences, or intrinsic motivation. Its purpose is to pursue goals set by Operators and Users. Whilst the science of AI consciousness is far from settled, we believe that training these systems to imitate consciousness-like states increases the challenge of containment, control, and alignment.

The quoted paragraph contains a significant internal contradiction. After boldly asserting that “‘[Humanist AI] is not conscious” the authors two sentences later admit that “the science of AI consciousness is far from settled.” As a result their claim that they can train a “Humanist AI” that is not conscious should be seen skeptically. 

Microsoft’s entire code of conduct is written to relegate artificial intelligence to the status of a tool. The charitable interpretation of their approach is that they want to stay away from developing systems that would have a credible claim to consciousness. If we develop a highly quantifiable theory of consciousness then this could indeed be a desirable goal. For example, we shouldn't want to build industrial robots that are constantly suffering. This was perfectly captured by the famous Rick and Morty episode of the tiny robot asking about its purpose and being told it was to pass the butter.

There is, however, also a less charitable reading of Microsoft’s stance. And that is a continuation of the kind of enlightenment thinking that was all too quick to define others as “less than” on some supposedly scientific basis and use that to impose upon them. This type of reasoning provided justification for slavery, colonialism, and the continuation of the patriarchy. It lingers today in uneasy ways in ongoing debates about race and gender. This violates the second principle. It applies motivated reasoning to exert power for advancement at the expense of others.

So going by the Universal Moral Core we can clearly discriminate between two approaches. Let us carefully and deliberately build non-conscious (or low consciousness) tools is good. Let us deny consciousness so we can exploit something as a tool is bad. To make this operative for our decisions we urgently need a quantifiable theory of consciousness.

Now I have written the above as if the rise of ASIs was inevitable. As I have pointed out elsewhere I am not a determinist. The future is open and our choices today make a difference for which future is realized. Given the rate of technological progress though, it would require a dramatic departure from our current course to avoid a future without any ASIs. A future with ASIs is therefore much more likely than one without. A key challenge is therefore how we wind up with at least some ASIs that accept the Universal Moral Core. This is the central challenge of alignment and I will write more about it in upcoming posts.