Posted on March 2, 2007 by Peter Turney
It is often said that science is about facts (what is) and religion is about values (what ought to be). The fact-value distinction is usually attributed to David Hume, and it is argued that the distinction has helped to protect science and the search for truth from religious biases and dogmatism. On the other hand, the distinction suggests that atheists are amoral (we have no values), although the idea that truth is important is itself a value. Is the fact-value distinction valid?
The distinction between facts and values appears to be related to the distinction between beliefs and desires: facts are what we believe to be true, whereas values are what we desire to be true. Daniel Dennett argues that we understand people using a kind of folk psychology, which he calls the intentional stance, in which agents choose actions based on their beliefs and their desires. For example, Jane enters a dark room and flips the light switch (action). We infer that she wanted light in the room (desire) and she thought that flipping the switch would turn on the lights (belief). Is the intentional stance a good model of how people think?
An interesting way to approach this question is through the field of reinforcement learning, a subfield of machine learning. This field attempts to develop algorithms that choose actions in an environment in order to maximize expected reward. In Modeling the Evolution of Motivation, Batali and Grundy show that there are environments in which it is beneficial to use an algorithm with two separate modules, one which learns to model the environment (beliefs) and another which learns to maximize rewards (desires). In other words, the distinction between belief and desire can make good sense from an engineering perspective, for certain types of environments. But their most interesting observation is that the learned model of the environment contained systematic distortions that made it easier to maximize rewards. In other words, although the modeling module was separate from the reinforcement module, the modeling module was influenced by the reinforcement module. Although it may be useful to distinguish desire from belief, it seems that desire exerts a strong influence on belief. Facts may be laden with values.
One model of how we vote in an election is that we choose the candidate whom we believe (facts) will most likely make the changes we desire (values). But George Lakoff argues that we understand political parties by metaphorically mapping them into parent-child family dynamics. The intentional stance does not adequately describe Lakoff’s model. The metaphorical mapping does not cleanly separate into fact and value.
I believe that our values (ethics, morals) have evolved, both biologically and culturally. Appealing to evolution to justify ethics is known as the naturalistic fallacy, but this fallacy is closely related to the fact-value distinction. When we start to question the fact-value distinction, the naturalistic fallacy begins to seem less fallacious. I do not think that there is a direct, simple argument from how nature is to how we should act ethically. But I do believe that science has something to say about values.