Blog Editor


  • Copyright 2005-20012 by Adam Kolber
    All rights reserved.

« "Utility: Anticipated, Experienced, and Remembered" | Main | "Keeping Up with Neurolaw: What to Know and Where to Look" »


I pointed out some more reasons why we would not want to really enforce laws smoothly (indeed can not since gossip, employer discrimination, increased likelihood of reoffense will all be effects of even a conviction with a small penalty) in a comment to the first post. Here are a few more thoughts.

First, it's not always clear what it means that the law has smooth penalties. Suppose a speeding ticket is always $500 but the probability that cops will pull you over when speeding is a smooth function of how much you are speeding. Is this a smooth law because the expectation of punishment varies smoothly with behavior or a bumpy one since the penalty is always $500 even for speeding 1 mph. I think our intuitions say that it is the actual penalty that should vary smoothly not just the expected. It's unjust to have a huge prison term because you were unlucky even if you went in with a low expected number of years in prison.

Secondly, while I agree with some of your points I don't think you provide enough justification. For example, one might imagine setting, as a matter of law, the penalty as a function of the risk of death/non-consensual sex/etc.. created by your behavior in excess of some threshold. The reason we can't usefully set such laws is that people are very poor at reasoning about probabilities this way and they are especially bad at ignoring disastrous actual effects when judging how harmful a behavior might be.


Having said all this I also would note that we don't really believe that the level of punishment should turn on how probable it was that the defendants behavior would lead to harm (or how likely it was that his belief the other party consented turned out to be wrong).

For instance, consider someone named David Jones who thinks "In general a woman has to verbally tell you she wants sex free of intoxicants in clear terms in her native language immediately before the act but any woman whose middle name is "Fuck Me David Jones No Matter What I Say." consents to sex even if she verbally disavows this fact." Sure, it's a really weird belief but it's not impossible.

This behavior is extremely unlikely to ever lead to non-consensual sex and his beliefs that women consent when they satisfy this standard is actually quite reasonable. After all it is VERY unlikely that his judgement will be wrong. But imagine he meets someone who happens to have that middle name and, despite convincing evidence that she meant another David Jones or did it on a lark, really believes she consents to sex and ends up raping her. True, his rule for inferring consent was extremely unlikely to be wrong because it is super improbable to meet someone with that middle name but we don't feel the improbability of meeting such a person is what should govern how much he should be punished. True, this was a really implausible rule for behavior but the same issue comes up with many more plausible ones.

Moreover, this highlights a theoretical problem for smooth laws. There simply is no fact out there about how dangerous or unreasonable a defendant's behavior truly was. Any such conclusion about dangerousness requires you pick a particular rule to describe the defendant's conduct. Did the man who mistakenly shot the police officer have the rule "Fire at anything moving in the dark around the house." or just "Fire at something dark moving around the house during a murder spree after hearing glass break"? Either one correctly describes what the defendant actually did yet they differ starkly in how reasonable and dangerous they are. Yet, no fact about what actually happened can prove one correct or the other.

Thanks very much for your comments, Peter. A few thoughts that might help:

* While I speak of smooth and bumpy laws as a shorthand, my more detailed description describes relationships between a single input and a single output. So in your speeding ticket example, the $500 penalty is bumpy with respect to speed. (There is a smooth relationship between speed and probability of detection, though probability of detection may not be conventionally treated as a legal output.) We can separately evaluate the merits or demerits of making either relationship smooth or bumpy.

* I agree that we cannot always trust legal actors to understand probabilities. In cases where we expect errors to be introduced, I imagine we should weigh the expected magnitude of those errors relative to alternative approaches.

* I like the point you make in the last paragraph. Suppose a defendant was attacked by a man with pink hair and a goatee carrying a big, sharp knife. The defendant shoots and kills the attacker. The defendant will likely have a good claim of self-defense. It's true that he might have been operating under a different rule. Kill anyone with pink hair and a goatee. If so, our defendant is a dangerous person. But the law will often try to sort out, at conviction and/or sentencing, what a defendant's beliefs and motivations were. It's true that such inferences may be mistaken. But it's not as though all possible beliefs and motivations are equally probable.

Moreover, I'm not sure why this concern is directed especially at smooth laws. Errors in assessing beliefs and motivations are very dangerous in cases where the law draws thresholds (e.g., bumpy laws) because we may end up on the wrong side of the threshold. They are also dangerous in the context of smooth laws because we might deviate from ideal penalties. Both sides can suffer from error, and I'm not sure that it's worse in the bumpy context.

The comments to this entry are closed.