It's shocking how, despite all the response, no one has presented a clear general case.

It's called risk management. We learn about risk when a bad thing happens. When a bad thing happens, that means it was possible for it to happen. The next question is "under what conditions will this bad thing happen again". And as it turns out, there is no deterministic answer to this question. It's all probability.

So condition X increases the probability of bad thing Y happening by amount Z. When a bad thing (Y) happens, we commission studies to look into all the possible conditions (X1, X2, etc) that contribute to Y and what the probability increase is for each X.

Armed with this information, we then set about the political process of determining the negative value of the bad thing and the negative value of eliminating the contributing conditions.

So for example, people get skin cancer (bad thing). The sun is a contributing condition to people getting skin cancer. If there was no sun, less people would get skin cancer. But if we destroy the sun, we destroy all of humanity. So addressing the condition of "the sun exists" is more negative than the bad thing "some people get skin cancer".

Following the same example, another condition for skin cancer is that the skin lacks the physical ability to protect against the sun's harmful rays. We could solve this by forcibly genetically engineering the entire human population, or by forcing everyone to wear a poncho or burka, or by forcing everyone to wear sunscreen. But politically, we have chosen that since skin cancer's largest impact is in the individual who develops it, and that the social cost of skin cancer is small by comparison, that the negative effects of removing bodily autonomy from everyone are worse than the bad thing "some people get skin cancer some times".

So now we arrive at the final component of why we punish people for contributing to the conditions that increase the probability of bad thing happening. Social impact.

If the bad thing we are analyzing affects a lot of people AND the bad thing happens to people who are not exclusively the people who brought about the conditions that increase the probability of the bad thing, then we, as a society, sometimes decide that the best way to reduce the bad thing occuring is to reduce the probability that it will occur and that it is socially acceptable to punish people who raise the probability because that punishment will reduce the frequency and volume of that behavior and therefore reduce the frequency and volume of the bad thing.

In the example of speeding, it's easy. Speeding is totally legal for drag racers on a closed track and no one punishes them for going too fast. They might kill themselves, but that's on them. But speeding on public roads leads to the death of people other than the driver. The driver could also die, but especially with pedestrians and cyclists, the speeder usually survives but other people die because of the speeder's behavior. These things also happen when people drive the legal speed limit. In fact, it's just a reality of living in a society with tons of motor vehicles.

But the research shows speed is a major contributor to deaths of people other than the driver, and that limiting speeds absolutely results in fewer deaths caused by motor vehicles. In fact, literally every single speed limit set in the world is backed by a speed study that you can challenge with empirical evidence and trigger an empirical review to determine whether the speed limit is set too high or too low.

We could go to the next step and automate speed governance. We could require all cars be installed with a speed governor, we could put up digital beacons to announce speed limits and require all cars be designed to respect the speed limit at all times. But, we have decided that the negatives of doing that are worse than the number of deaths caused by motor vehicle accidents where speed was a contributing factor. Costs would be very high, effectiveness would be very low at the outset, false negatives and false positives would be rampant, and people could disable their governor system and then we'd still need traffic monitoring to catch the speeders (although disabling the governor might be punished by jail time and vehicle destruction).

It's all math. We do it because the math shows that it works. Reducing the risk factors of a specific negative outcome reduces the negative outcome. And we do it because we do not want to wait for every single human being to have to experience the bad thing to learn to govern their own behavior, especially when that bad thing could be the death of innocent people.