Practical Rationality

Recall from the beginning of this chapter that epistemic rationality is reasonableness of belief, whereas practical (or pragmatic) rationality is reasonableness of choices or actions. Decision theory is the branch of philosophy that deals with practical rationality: it investigates what kinds of choices and actions are reasonable. A popular account of practical rationality is Bayesian decision theory, which will be introduced in what follows.

Bayesian Decision Theory

In Bayesian decision theory, a choice between several alternative actions can be represented in a decision matrix—a table in which the rows represent possible choices or actions, and the columns represent a relevant set of mutually exclusive and exhaustive propositions about the world, like this:

W1 W2 W3
A1 U11 U12 U13
A2 U21 U22 U23
A3 U31 U32 U33

A1, A2, A3 (etc.) represent the possible acts you are considering. W1, W2, W3 (etc.) are mutually exclusive and exhaustive propositions about the world that are relevant to your deliberations. U11, U12, (etc.) represent the utilities you assign to each possible outcome. That is, they represent how much you value or disvalue each act + proposition pair.

This is all very abstract, but bear with me! It will make more sense when we look at a specific example, below.

The expected utility of each possible act (A1, A2, A3, …) is calculated by multiplying the utility of each possible outcome times your credence in the corresponding proposition, then adding these numbers. For example, if pr(W1), pr(W2), and pr(W3) are your credences in the propositions W1, W2, and W3, then the expected utility of act A1, given the above decision matrix, is:

U11 pr(W1) + U12 pr(W2) + U13 pr(W3)

According to Bayesian decision theory, the rational thing to do is to choose whichever act has the highest expected utility. If there’s a tie, and two options both have the highest expected utility, either option is rational.

Suppose you don’t know whether one of your professors will give a pop quiz tomorrow, but you think it’s 30% likely. In other words, your credence in the proposition that there will be a quiz tomorrow is:

pr(Q) = .3

You are considering whether to study, just in case. The choice is between two possible actions: to study or not to study, that is the question! The propositions Q and ~Q are mutually exclusive and exhaustive propositions relevant to your deliberation. So, we’ll need two rows and two columns in the decision matrix, and there are four possible outcomes to consider:

  1. If you study and there is a quiz, let’s suppose you regard that as a good outcome, since you’ll be well-prepared and likely get a high grade. We’ll give that possible outcome a value of 40. (The exact numbers and the units don’t matter, since this is just an idealized theoretical model to help us understand how rational deliberation works.)
  2. On the other hand, if you don’t study and there is a quiz, that’s the worst case scenario. Let’s give that outcome a disvalue of -50.
  3. You regard the possible outcome in which you don’t study and there is no quiz as moderately bad, since you feel you wasted your time studying, but not nearly as bad as failing the quiz. Let’s rate it with a disvalue of -10.
  4. Finally, the outcome in which you don’t study and there is no quiz is ok, though not quite as exciting as the thrill of acing a quiz when you’re well-prepared. Let’s give it a 20.

The following decision matrix represents your situation:

Quiz
pr(Q) = .3
No quiz
pr(~Q) = .7
Study 40 -10
Don’t study -50 20

What should you do? According to Bayesian decision theory, the practically rational choice is whichever option has the highest expected utility. In this example, the expected utility of studying is:

(40 × .3)  +  (-10 × .7)  =  5

The expected utility of not studying is:

(-50 × .3)  +  (20 × .7)  =  -1

Studying has the highest expected utility. So, according to Bayesian decision theory, studying is the rational thing to do in this situation.