[jrkatz]

> lock their car and take them to the police station and let them contact someone from there. But don't force them to pay a fine or go to court if they haven't caused harm.

Supposing at least some drunk drivers are aware they are drunk but fully convinced they will not harm anyone, this improves their worst case scenario substantially. Before, losing their license; now, getting even closer to home before calling a ride. Does this incentivize drunk driving among people unable to evaluate the danger they present to others (e.g., drunks)?

Obviously, this falls apart somewhat because those same people probably drive drunk today thinking they won't get caught. Nonetheless, the common consensus is that certainty of punishment is the primary deterrent against criminal activity. Certain non-punishment will change some of the calculus.

[pdonis]

> the common consensus is that certainty of punishment is the primary deterrent against criminal activity.

There is no such thing as "certainty of punishment". There isn't even "certainty of punishment if you get caught", since our society is not a sane society in this respect and lets people who do cause harm, up to and including murder, off the hook for all manner of insane ideological reasons.

That said, I didn't say we shouldn't punish people who actually cause harm. The scenario I was describing was a person who is, for example, driving drunk, but hasn't caused harm, and gets stopped by a cop. Should the cop be able to ticket them? Or just prevent them from driving until they are sober? If they haven't caused harm, I would say the latter.

[jrkatz]

I'm sorry, I should have been clear. The 'certainty of punishment' bit should be understood as a probability. The gist of it is that a person considering criminal behavior isn't worried about the severity of the punishment `S` in isolation, but also the likelihood (read: 'certainty') `C` thereof. Simplistically the deterrence model is `C*S`. An extremely severe punishment (execution!) with zero or effectively zero chance of getting caught provides no deterrence. There's more to it - for example doubling severity does not double deterrence in general - but that's the gist of it.

My little thought experiment supposes a person who is drunk evaluates, before deciding to get in the car, the potential outcomes of drunk drive, while entirely convinced they won't harm someone. (I suspect most drunk drives are entirely convinced they won't harm someone, although it is obvious to anybody else there is at least a _chance_ they will.) Today, getting caught with whatever nonzero probability results in a punishment of some severity. By the model we're assuming above, this produces a nonzero amount of deterrence. If we shift away from punishing drunk drivers, the severity drops to zero. If a drunk's options are to call for a ride from the bar or try to drive home, and getting caught at the latter only mean they'll have to call for a ride from some location between the bar and home, driving is the rational choice after all, _so long as they are convinced they won't hurt anybody_.

I was being too circumspect. The point of my scenario is to highlight that decriminalizing drunk driving will likely increase the amount of drunk driving, at least among the population of drunk drivers who are overconfident in their abilities. That in turn will increase the danger drunk drivers present to the rest of us.

I gather you think reaction is morally preferable to prevention in such cases. Is that a fair assertion? In general, I bet we agree there is some amount of risk to myself that I must tolerate to afford others their freedoms, and have just landed at different risk tolerances. I try to take need into account when drawing my line in the sand. "Does anybody _need_ to drive drunk?". Since it's a situation drunk drivers make for themselves I'm happy to punish it. How do you draw your line?