by Brian Tomasik
First written: 19 Mar. 2015; last update: 19 Mar. 2015
Sometimes people object to an action by pointing out that if everyone did the same action, bad consequences would follow. This reasoning resembles Kant's categorical imperative. Is such reasoning valid? And how do we interpret it concretely? I argue that the categorical imperative can be interpreted in varying ways depending on the types of maxims that are taken as being followed. The clearest defense of a flavor of categorical imperative comes from non-causal decision theories.
"But if everyone did that..."
Suppose you live in California, and you're deciding whether to vote in the upcoming election. Your district is so liberal that all its elections will result in landslide Democratic victories whether you vote or not. In general, the cost to buy a vote through get-out-the-vote (GOTV) drives may be ~$50-500, but your vote is worth almost nothing because of where you live. So instead of voting, you decide to stay home, get some extra work done, and thereby earn slightly more money that you can donate to GOTV in swing states.
A common objection to this kind of reasoning is: "But if everyone did that, no one would actually vote for the Democrats, so Democrats would lose!" It's very unclear whether and in what circumstances this kind of "if everyone did that" objection is valid.
If everyone did what?
The objection seems to take inspiration from Kant's categorical imperative:
Act only according to that maxim whereby you can at the same time will that it should become a universal law without contradiction.a
A main puzzle is to specify exactly what maxim we're attempting to universalize. Presumably someone objecting to a policy of earning instead of voting in safe districts takes the maxim to be: "Donate to GOTV campaigns instead of voting yourself." If every Democrat followed this maxim, Democrats would indeed lose no matter how big their GOTV budgets.
But suppose the maxim is instead: "When the marginal expected votes from donating exceed those from voting yourself, earn to donate instead of voting directly." Then some people, especially higher earners, will calculate that they can do better by donating. Eventually, GOTV efforts will become saturated with cash, at which point it will become more cost-effective to vote directly. By following this maxim, Democrats would optimize their chances of victory.
So which maxim should a person take himself to be following when he chooses a given action? Scott Alexander makes this indeterminacy even more manifest by the following example:
Let’s say Paula the Policewoman is arresting Robby the Robber (she caught him by noticing his name was Robby in a world where everyone’s name sounds like their most salient characteristic). No doubt she thinks she is following the maxim “Police officers should arrest robbers”. But what about other maxims that lead to the same action?
1. Police officers should arrest people
2. Everyone should arrest robbers
3. Paula should arrest Robby
4. Paula should arrest other people
5. Everyone should arrest Robby
6. Everyone should arrest EVERYONE ELSE IN THE WORLD
Consequentialism as a categorical imperative?
One particular maxim that someone could follow would be: Choose the action whose calculated expected consequences seem best. This would be act-consequentialism. Act-consequentialists would be glad to see this maxim become universal. So it naively appears as though consequentialism is at least consistent with the categorical imperative. RM Hare elaborates on this point in "Could Kant Have Been a Utilitarian?"
Hare wonders whether Kant's categorical imperative assumes that the maxims in question must be extremely simple:
[Kant's] argument against suicide is again very weak. I could certainly without contradiction will universally that those who would otherwise have to endure intolerable pain should kill themselves. This could indeed become a universal law of nature, and I could act as if it were to become so through my will. Kant thinks [his] is a good argument only because he thinks (perhaps owing to his rigorist upbringing) that maxims have to be very simple. If we have a choice between the simple maxims 'Always preserve human life' and 'Destroy human life whenever you please', we shall probably opt for the former. But there are many less simple maxims in between these extremes which most of us would will in preference to either of them: for example 'Preserve people's lives when that is in their interests' (and perhaps we would wish to add other qualifications). As we have seen (8.1) moral principles do not have to be as simple and general as Kant seems to have thought, and they can still be universal all the same [...].
Categorical imperative as decision theory
The categorical imperative makes most sense to me when interpreted through the lens of decision theory. In particular, compare Kant's formulation of the categorical imperative with this summary of timeless decision theory: Choose the output to your cognitive algorithm whereby you can at the same time will that it should become the universal output of all instances of that cognitive algorithm. This clears up the fuzziness about exactly what maxim our action is supposed to be following, since the "maxim" is whatever algorithm we're executing when making a given decision.b In fact, there are many algorithms that go into a given choice, so presumably we should act as though we're determining all of them at once. I don't know exactly how to make this work, but now we can see that it's just a technical problem in the realm of decision theory.
Determining in practice which sorts of algorithm outputs are fixed by various choices is a nontrivial question. So we still haven't fully answered the "But if everyone did that..." objection, but at least we can now see where to begin when sorting it out.
The ideas in this post seem fairly well known on LessWrong. As an example of a similar idea, see "UDT agents as deontologists". In 2006, Gary Drescher mentioned the connection between the categorical imperative and non-causal decision theory in chapter 7.2.1 of Good and Real.
- Kant's interpretation of the "without contradiction" clause seems fishy to me. For example, Wikipedia says:
According to his reasoning, we first have a perfect duty not to act by maxims that result in logical contradictions when we attempt to universalize them. The moral proposition A: "It is permissible to steal" would result in a contradiction upon universalisation. The notion of stealing presupposes the existence of property, but were A universalized, then there could be no property, and so the proposition has logically negated itself.
Compare to the following reasoning, which I made up:
The moral proposition B: "It is required to cease from bullying" would result in a contradiction upon universalisation. The notion of ceasing bullying presupposes the existence of bullying, but were B universalized, then there could be no bullying, and so the proposition has logically negated itself.
- Actually, deciding what algorithm we're implementing remains a fuzzy question, but it seems at least much more concrete than the question of what general maxim a given action corresponds to. (back)