In last week's post I explained the first principle of the Universal Moral Core (UMC) in greater detail. Today I want to investigate some of its implications for how we should behave.

As a reminder, earlier in the series, I suggested that the goal of philosophy should be to help us answer the following question:

How should we choose our actions in light of our understanding of reality and the potential impact of our actions on this reality (which happens to include how we and others are feeling)?

So how does principle 1 help us do this? Well for each action we can ask ourselves if it is likely to help grow knowledge. One important part here is the "likely" -- we can never have certainty about the outcomes of our actions. There are too many unknowns and too much randomness in the world. This is why we need to apply judgment, which I introduced earlier in the series.

Take lying as an example. What does principle 1 tell us about it? The answer is: it depends. There can be no strict rules or calculus of consequences absolving us from exercising judgment. Different people will come to different conclusions, which is entirely acceptable as long as the principle can act as an overall guide post to making decisions.

Consider lying about the outcome of an experiment by falsifying the data. The principle suggests that this is bad and probably to be avoided. Falsified data pollutes knowledge, potentially for generations to come. Allocation of resources may be severely misguided as a result. The 2006 Alzheimer's paper in Nature perfectly illustrates this type of damage. The author had manipulated the experimental results to fabricate cleaner and more impactful data. The paper wasn’t retracted until 2024 and in the meantime a lot of research funding had been allocated towards chasing down this specific finding.

On the other hand, consider lying to an officer of a dictatorial regime to protect a scientist from persecution. The principle suggests that this is acceptable. Technically, given that I have defined knowledge as being externalized in a medium, a lie in a conversation doesn’t even enter knowledge (at least not directly unless someone records it). Even over email or text it stays quite confined between two parties. If that exchange were to be published it would still be seen as a statement in a specific context. Now what about if this scientist subsequently winds up developing a horrible virus that kills millions of people? Unless you had active knowledge of their plans to do so this cannot possibly enter the decision. The details of the future are unknowable. The scientist might instead be the one who develops the antiviral.

This is one concrete example of how the principle enables people to make judgments about choosing an action. It is not crucial that everyone arrives at the same judgment every time. Instead what matters is a direction, which leads me to the next point. The principle isn’t just about individual choice but it is also about systems which have a higher likelihood of growing knowledge.

Let’s consider censorship. We can easily see how this is bad when it suppresses the publication of criticism. Without the ability to criticize existing knowledge, our explanations can no longer be improved. Wrong or misleading knowledge will simply persist. As it turns out there are many subtle forms of censorship, such as the allocation of attention and resources to graduate students based on whether what they are working on is supportive or critical of the existing dominant theory (often created by their advisor or the head of the department). Then there is the pervasive self-censorship based on wanting the approval of others or more recently from fearing an online mob. The first principle suggests that these are all bad forms of censorship. Systems that support pervasive censorship are thus bad.

Does principle 1 categorically rule out all censorship? No. For example, censoring calls for violence against individuals or groups is in most cases compatible with this principle. Threats of violence are a key method of suppressing criticism. So censoring these types of threats will often support the growth of knowledge.

Now a big claim of the UMC is to apply, well, universally. Let’s consider an artificial super intelligence (ASI). We would want such an ASI to be deeply aligned with this principle, so that it can exercise judgment in situations such as the ones described above. An ASI that falsifies data at scale would deeply pollute knowledge. Given that it lacks many of the limitations that apply to humans, the scale of such pollution could be vast and take a long time to discover and unwind. An ASI lying to protect a specific human from prosecution again seems entirely acceptable. Similarly, ASI participating in mass censorship seems deeply problematic as it could easily freeze knowledge. Conversely, an ASI detecting and suppressing calls for violence appears beneficial.

What about different human societies? One requirement for the UMC was to “allow for the largest variety of cultures to exist.” So from the perspective of principle 1 it is not an issue if there are some societies that want to exercise extensive censorship, as long as this remains limited to their society. It will potentially result in a slow down in the growth of knowledge but will not inhibit it. Where we need to exercise great care though is if that society is trying to export its censorship widely so that it becomes dominant for all of humanity. Similarly, it is acceptable, albeit undesirable, if a society is worse at rooting out academic fraud, again as long as this doesn’t become endemic to all of humanity. So the principle is universal in the sense that these societies would likely be better off with less censorship and higher academic integrity but there is no need to compel them unless it becomes a systemic threat to all of human knowledge.

Similarly, we should not expect all ASIs to be aligned with this principle. Someone somewhere will likely create an ASI aimed at pervasive global censorship. What matters is that there are a sufficient number of ASIs aligned with this principle to help fight off any threats from the ones that are not. This is one of several reasons why a plurality of ASIs is highly desirable over a single ASI.