Lexical Priority and the Problem of Risk

Here, I refute lexical priority theories in ethics.*

[ *Based on: “Lexical Priority and the Problem of Risk,” Pacific Philosophical Quarterly 91 (2010): 332-51. ]

1. What’s Lexical Priority?

Basically, lexical priority theories say that there is some kind of reason that is infinitely more important than another kind. There are many examples of lexical priority theories, e.g.,

1. Some deontologists think that there are “absolute rights”, which are more important than any amount of utility. So, e.g., they might say you can’t intentionally kill an innocent person, even if it would save the rest of the world from complete destruction. These people don’t deny that utility is worth promoting; they just think that rights are always more important than any amount of utility.

2. John Stuart Mill supposedly thought that there are different “qualities” of pleasure, and that any amount of a higher-quality pleasure is better than any amount of lower-quality pleasure. E.g., one person getting to experience the pleasure of understanding a Fake Nous blog post is more important than a trillion people enjoying pure sensory pleasure for eternity.

3. Most people reject the Repugnant Conclusion (RC) when they hear of it.

   Note: The RC says that if you consider a world full of happy people, it’s always possible to imagine a better world consisting of a sufficiently large number of people whose lives are just slightly worth living.

   The denial of the RC might stem from a lexical priority view (though there are other ways to interpret it): i.e., perhaps people think that a very happy life is infinitely more valuable than a barely-worth-living life, so no number of barely-worth-living lives can outweigh a very happy life.

4. John Rawls’ theory of justice proposes that, for a just state, personal liberties have lexical priority over producing & distributing wealth. I.e., we can’t justify infringing personal liberties to produce any amount of wealth. He also proposed that benefitting the poorest class of society should have lexical priority over benefitting wealthier people. (So you can’t take a penny from the poor even if that would produce $1 trillion of wealth for middle-class people.)

Example 2 above shows that Lexical Priority theories need not be deontological; Mill (according to the orthodox interpretation) held a consequentialist lexical priority theory. Btw, they need not be moral theories either; they could be theories about prudential reasons (as Mill’s view also shows) or practical reasons more generally.

2. The Risk Problem

So let’s say you have a “higher reason”, R, which outweighs any amount of some good G. I.e., R says that you should not do A, but A would produce a large amount of G. The Lexical Priority theorist says you can’t do A no matter how much G it would produce. (E.g., “you can’t kill an innocent person, no matter how many other lives it would save!”)

Cool. Now, what happens if you have some action, B, which also produces a large amount of G, where B only has some probability of going against R? For instance, it might only have some risk of killing an innocent person. Can you take a nonzero risk of killing an innocent person, in order to save the rest of the world from destruction?

I see 3 answers you might give to this:

i. Zero Risk Tolerance: Nope, action B is also absolutely prohibited, no matter how low the risk is, as long as it is nonzero.

ii. 100% Risk Tolerance: The absolute prohibition only kicks in when you are 100% certain of violating R (e.g., it’s certain that the innocent person would be killed).

iii. The Threshold View: There is some threshold probability, t (strictly between 0 and 1), above which the absolute prohibition kicks in. Below t, the risk can be justified by sufficiently large benefits; above t, it can’t.

3. Those Are All Bad Answers

3.1. Against Zero

Suppose we have zero risk tolerance. Problem: there is always a nonzero risk of almost anything. This is going to give the lexical priority theorist an absurdly impractical theory. Examples:

a. Absolute justice: Elizabeth Anscombe once said that it is always unjust to knowingly punish an innocent person, and that if you think such an injustice could be justified by sufficiently good consequences, then you have a corrupt mind and she doesn’t want to talk to you.

However, whenever you punish anyone, there is always a nonzero probability that you’ll be punishing an innocent person. So if we embrace Zero Risk Tolerance, then we can never punish anyone, so we have to shut down the criminal justice system.

Of course, Liz said “knowingly”. You might claim that we never knowingly punish the innocent. But this is false; by having a criminal justice system at all, we know that we’re going to be sometimes punishing the innocent (we just don’t know which people are the innocents we are punishing).

b. Absolute right to life: Suppose the right to life is absolute. Now, if I had a deadly disease, and I knew for certain that by going near you I would transmit that disease to you, which would kill you, then it would be wrong for me to go near you. It would, indeed, be murder and hence unjustifiable by any amount of good produced.

So, if we embrace the Zero Risk Tolerance, we must also say that if there is a nonzero probability that I have a deadly disease which I will transmit to you by going near you, then it is also absolutely prohibited for me to go near you. But there is always a nonzero risk of this. So no one may go near anyone.

More generally, no one can do anything.

3.2. Against 100%

The 100% Risk Tolerance view doesn’t have absurd consequences. The problem is just that it is pointless, because in real life, virtually nothing has a 100% probability. Hence, the lexically superior reasons would never apply in real life, which kind of defeats the point of holding a lexical priority theory.

E.g., you’re never certain that you’re going to be killing an innocent person, so an absolute deontological prohibition on killing the innocent would never apply to any real actions.

3.3. Against Thresholds

Maybe this is the most common sense view: you can only take risks that are below a certain threshold.

Problem: You could have two actions, B1 and B2, which are each below the threshold, but the combined action of doing both B1 and B2 (“B1+B2”), is above the threshold. Suppose that each of B1, B2, and B1+B2 would produce a huge amount of good. Then you can do B1, and you can do B2, but you can’t do B1+B2.

Or perhaps: whether you can do B1 and B2 depends on how you individuate actions: if you think of B1+B2 as a single action, then it’s impermissible, but if you think of it as two actions, then it’s permissible. That seems wrong; the permissibility of some behavior can’t depend on how we individuate actions.

Example: Back to punishing the innocent. Let’s say you can punish someone as long as they are at least 95% likely to be guilty. If they have a >5% chance of being innocent, then it’s absolutely impermissible to punish them, regardless of the consequences. Now suppose we have two defendants, A and B, who are each 96% likely to be guilty. We punish both of them, producing large benefits. Each had only a 4% chance of being innocent, so it’s fine. But wait. Between the two convictions, there is a 7.84% probability that we punished at least one innocent person (we just don’t know who, but that seems irrelevant). That’s over the threshold, so we committed a horrible wrong (?).

Again, the criminal justice system as a whole is virtually certain to sometimes punish the innocent, as long as we have any standard of proof short of absolute certainty (i.e., as long as we ever punish anyone). So the “action” of having a criminal justice system would be prohibited.

• Possible silly solution: “It’s okay because different people do the punishing for different defendants. You can’t aggregate actions by different agents; hence, there is no action that goes over the risk threshold.”

  This view has the implication that if you have multiple different defendants, it’s super-important that you get different juries, different judges, etc., to run the trials. E.g., if the same jury convicted both A and B, then there’s a horrible injustice, whereas if it was two different juries, then everything is fine. I take it that that’s obviously wrong.

4. Conclusion

You should reject lexical priority theories because they have no reasonable answer to how to deal with risk.

The argument applies to nearly all lexical priority theories (unless for some reason it is impossible to have a mere risk of acting against the lexically superior reasons).

It does not apply, though, to weakly deontological theories. E.g., if you think it’s permissible to violate rights when sufficiently large benefits are produced, then you should also say that it’s okay to take a risk of violating rights when sufficiently large expected benefits are produced. That won’t create any incoherence (at least, not of the type discussed here).