Less Wrong has held many discussions of willpower. All of them have focused on the cases where willpower fails, and its failure causes harm, such as procrastination, overeating and addiction. Collectively, we call these behaviors akrasia. Akrasia is any behavior that we believe is harmful, but do anyways due to a lack of willpower. Akrasia, however, represents only a small subset of the cases in which willpower fails, and focusing on it too much creates an availability bias that skews our perception of what willpower is, how it works and how much of it is desireable. To counter this bias, I present here some common special cases where strong willpower is harmful or even fatal.

Consider the human sleep cycle. By default, we settle into an equilibrium in which we sleep for about eight hours per day, starting at about the same time each evening. We can override the normal sleep schedule, and stay awake when we should be asleep, or sleep when we should be awake, by applying willpower. For this example, we'll confine the definition of willpower to just this one ability. Willpower is the ability to control sleep times; we'll leave the ability to work hard and resist cake for later.

We use sleep-willpower to arrange our hours strategically. However, we can only do so within certain constraints, which cannot be overriden by conscious choice. It isn't possible to refrain from sleeping entirely, because the amount of willpower required to stay awake increases with the amount sleep missed, until eventually it exceeds the amount of willpower available. Attempting to minimize sleep for a long enough period will eventually cause a shift to an alternative equilibrium, polyphasic sleep, in which all but the most important stage of sleep (REM) are skipped. It's harder to stay awake in some circumstances, like dark rooms and boring lectures, than in others. And it's impossible to sleep too much more than normal; the better rested we are, the harder it is to fall asleep. These restrictions are protective; when rats are artificially prevented from sleeping for too long, they die, and when humans use stimulants to stay awake too long, they die too. If a human had unlimited sleep-willpower, then they would risk serious injury every time they used it.

Next, consider diet-willpower. Humans have a system that decides what is good to eat, which manifests as cravings and taste preferences. Many people try to override this mechanism, usually for health reasons, replacing them with advice from a perceived authority. Consider a hypothetical person who decides to follow a highly restrictive diet, to lose weight. Unfortunately, most of the diet advice available is bad; diet advice available in the recent past was very bad; and some of the diet advice floating around is disastrously bad. Restrictive diets tend to have harmful deficiencies. If that deficiency is one of the ones that our biochemical diet manager knows how to handle, then following the diet will require willpower. The more severe the deficiency, the more willpower will be required. Try to follow a no-fat diet, and lapses will be the only thing keeping you alive.

In these examples, we have a mechanism which provides defaults, and a mechanism for overriding them. In general, we call the defaults "System 1", and the override mechanism "System 2"; and whatever influence System 2 has over System 1, we call willpower. System 1 represents a general policy, and System 2 adjusts it for the present circumstances. This pattern applies across many different and sometimes unrelated mechanisms, and we use the term "willpower" for all of them. The larger the adjustment, or the longer it is applied, the more willpower is required; and the different implementations of willpower - sleep-willpower, diet-willpower, work-willpower, etc - all require resources, such as glucose and focused attention, which are finite and fungible. When willpower fails, it usually "breaks" - that is, it fails all at once - rather than failing gradually. Deviating from innate or habitual behavior requires willpower. Willpower is reduced by pain, discomfort, and many types of biochemical imbalance. Sustained application of willpower may adjust habits and produce a new equilibrium, but if no such equilibrium exists, it will eventually fail and cause a reversion or overcorrection in the opposite direction.

Willpower lets us adjust our behavior to match our decisions, and willpower failures protect us from mistakes. Generals often successfully convince their soldiers that they should be courageous and always fight to the death, but when stressed enough, they'll break, desert or surrender, and survive. For soldiers on the losing side of a battle, excess willpower is fatal. Teenagers sometimes pledge to stay celibate. Their psychology is designed to make sure they don't keep that pledge. Married couples pledge monogamy, but women with infertile husbands and men with infertile wives tend to cheat. When stubborn people argue, one of them has to give up and end the conversation eventually. The sooner that happens, the less time they waste and the fewer black eyes they get. Workers may be convinced to keep a frantic pace, but sheer laziness will keep them from working hard enough to injure themselves.

The next time you see a willpower hack, remember that willpower isn't always a good thing. Sometimes, what we think of as akrasia is actually protecting us from harm.