Values Are Facts
Humanity has developed many rich and powerful tools for thinking about “facts”. Our estimates of facts are rich and deeply connected, allowing us to learn much about each fact from all the others. Furthermore, we have many kinds of specialists who deal with particular kinds of facts.
However, we also care about “values”, and many say that our ways to think about facts just don’t work for values. They call attention to a “fact-value distinction.” And they develop whole separate ways to discuss values, and whole separate kinds of specialists, with value ways and specialists not much connected to their versions for facts.
That seems a mistake to me, as “values” are just another kind of “facts”. They may be an unusually difficult kinds of facts to think about, and there may be specialized tools appropriate for them. But due to values being facts, we should be able to use their connections to other facts to learn much about them. Our estimates of values should be well connected with and integrated into our tools and views about other related facts. Just as we do with other facts.
In “facts” I include everything we might say about the true arrangement of our physical world. Such as where each particle (e.g.,, electron, quark, or photon) is at each point in time. And its spin, momentum, etc. Not just at the present, but also the past and the future. Everywhere in spacetime, in fact. (And of course the shape of spacetime and the full quantum state of all of this.)
“Facts” also include all our observations and data. We have many standard tools for drawing inferences about our observations from physical arrangements, and vice versa, and also for estimating joint estimates regarding both. This topic area also includes “indexical” facts, about the mappings between observations and physical arrangements.
“Facts” also include all counterfactuals about other facts. That is, what fact would instead have been true if other facts had been different than they actually are, were, or will be. Our best theories of how the world works usually tell us how to predict counterfactuals given actual facts.
Finally, “facts” include fits to abstractions of all of the above. For example, “temperature at a point” is an abstraction, not implied by any exact particular arrangement of particles. But we have good standard ways to estimate that abstraction from such particle arrangements. Other physical abstractions include “molecules”, “planets”, “plasmas”, and “explosions”. And to the extent that we agree on how to fit such abstraction parameter estimates to actual arrangements, we can treat claims about these parameters as facts.
“Creatures” are also abstractions, and so we have many facts about them, including their locations, movements, and actions. For example, “attack” is a creature abstraction that we can usually fit to physical arrangements, allowing us to talk about who will, did, or might attack whom.
Another quite useful creature abstraction is “expected utility”, whereby we describe a set of creature actions via a set of state-dependent utility numbers, and also a set of state-dependent belief numbers, beliefs which are updated over time with the info contained in the observations of that creature. Even for creatures whose actions only approximately satisfy the axions of expected utility theory, we can often find a useful best fit to their actions using this framework. And thus talk about such a creature’s utilities and beliefs that way.
We have many other related decision frameworks that differ from expected utility theory. But almost all of them have analogies to info, beliefs, and utilities. Thus by fitting any of these decision abstractions to the actions of a particular creature, we can talk about the facts of a creature’s utilities, beliefs, and info. In this way there can be facts about each creature’s utilities, and thus about this one kind of “value.”
Some creatures, like humans, often have thoughts and words not only about ordinary facts, but also about their and others’ actions. (Thoughts can also be included in the category actions.) Such thoughts can be about what they or others have done or will do, or about what they might do given counterfactual assumptions. These thoughts can be not just predictions about such actions, but also various attitudes toward those actions, such surprise, identity, wariness, or approval.
Thought attitudes that fit some possible actions more than others can be combined with a decision framework to find the best fit utilities associated with such thoughts about actions. This produces facts about a different kind of creature “values”: the values associated with particular attitudes regarding some possible actions, instead of the values associated with actual actions.
One especially useful abstraction about human attitudes regarding actions is “social norms”, whereby human communities encourage and discourage the actions of they and others, via approval or disapproval. Such norm-based action approval can also be combined with a decision framework to produce a third kind of “value”: normative value. What some community says that a creature “should” think or do.
We thus have three kinds of “values” that can be described as “facts”: the values associated with actual (including future and counterfactual) actions, the values associated with thought attitudes toward actions, and the “normative” values associated with a community’s social norm disapproval of actions.
If you say “no, I don’t mean any of those kinds of values, I mean true real values, but I have ways at all to connect these true values to these other kinds of values which are facts”, well then I’m just not sure what you could possibly mean. If you say “I’m talking about which acts we might agree are actually good”, that looks to me a lot like a particular kind of thought attitude toward such acts.
Being facts, our views regarding all of these kinds of value facts should be integrated with our views regarding all the other connected facts. For example, imagine that we have views about how much actions, or attitudes toward actions, of different kinds of creatures are likely to be correlated with each other, controlling for context. Surely these between-creature correlations would influence our beliefs about the values of any one of them, including ourselves, given beliefs about the others. Similarly, when we have beliefs about correlations between attitudes about actions and the actions themselves, for the same creature, that connection should induce each of these to be influenced by our beliefs about the other.
The key point here is that values are facts, facts connected to each other and to other facts via the many varied and dense connections typical of the connections between all facts. Once we see this, we can realize that we have many useful strategies for inferring values. So, for example, to figure out which of your actions you might approve, you have available to you many other methods beyond directly consulting your intuitions about those specific actions, and asking those intuitions to approve or disapprove.
You can instead ask your intuition to approve or not of many other related actions. Or consider how your intuitions would likely change in counterfactual scenarios. Or consider the intuitions of others re related acts, not just here and now but all around the world and all through history. You can also look at all of your other actual actions, and your counterfactual actions. You can even consider what you know about the architecture of your mind, and about the cultural and biological history that produced it.
Your values are facts, in general facts are connected to many other related facts, and these many dense connections typically allow us to learn much about each fact. So you can learn a great deal about any one value by considering all of these other kinds of related facts, including value facts. Values need not sit in a mysterious separate sacred realm.