[SEQ RERUN] Dissolving the Question
Today’s post, Dissolving the Question was originally published on 08 March 2008. A summary (taken from the LW wiki):
Proving that you are confused may not make you feel any less confused. Proving that a question is meaningless may not help you any more than answering it. Philosophy may lead you to reject the concept, but rejecting a concept is not the same as understanding the cognitive algorithms behind it. Ask yourself, as a question of cognitive science, why do humans make that mistake?
Discuss the post here (rather than in the comments to the original post).
This post is part of the Rerunning the Sequences series, where we’ll be going through Eliezer Yudkowsky’s old posts in order so that people who are interested can (re-)read and discuss them. The previous post was Variable Question Fallacies, and you can use the sequence_reruns tag or rss feed to follow the rest of the series.
Sequence reruns are a community-driven effort. You can participate by re-reading the sequence post, discussing it here, posting the next day’s sequence reruns post, or summarizing forthcoming articles on the wiki. Go here for more details, or to have meta discussions about the Rerunning the Sequences series.
I have a mental model of myself, based on observations and analogies. The model is very imprecise, partly because of not having enough data and not processing them precisely enough (it is not even possible to simulate oneself on a 1:1 scale), partly because as a member of social species I have an incentive to introduce many biases into modelling.
When I ask “what will I do in situation X?”, this model gives multiple answers. This is because uncertain data allow multiple solutions. (Turning individual biases on/off could further increase the number of solutions.) As an analogy, if I know that number N is either 10 or 11 or 12, then the question “how much is N×N?” would give answer 100 or 121 or 144.
But when the specific instance of situation X really happens, I will do only one of the possible actions. But I do not completely understand the details that led to this specific action, so I can’t update my mental model. My model is still saying that there were multiple possible actions—but only one of them really happened.
“Free will” is then a hypothetical human ability that (in last minute) collapses the set of possible actions into one action.
In some sense, it is a confusion between a map and the territory—as long as my model is not able to predict which action will I choose, I perceive all of them as real. This error becomes more obvious in situations when I can’t predict my own behavour, but other people can, because they know me better than I know myself. (In religion, assuming an omniscient deity makes this confusion plainly visible.)
I think this confusion is so persistent, because predicting human behavior is really very difficult. In some situations very small differences can cause a different outcome, so it really feels like the choice of outcome has no visible cause.
These are important observations. But they still leave a lot unexplained. For example, you have models of many physical systems besides yourself. Many of these physical systems are too complicated for your model to predict. Imagine, for example, a mechanical random number generator whose inner workings you don’t know. Here too, your model gives multiple answers about what the random number generator will do, even though the generator will do only one of those possible things. Why do you not attribute free will to the generator? What is different about the way that you think about yourself (and other people) that leads you to attribute free will in one case, but not in the other?
Humans naturally do it, but we can learn otherwise. So the question is, why can we learn (and then really feel) that the generator does not have free will, but the same process would not work for humans, especially for ourselves.
First, we are more complex than a random number generator. The random generator is just… random. Now imagine a machine that a) generally follows some goals, but b) sometimes does random decisions, and c) rationalizes all its choices with “I was following this goal” or, in a case of random action: “that would be too much” or “it seemed suspicious” or “I was bored”. Perhaps it could have a few (potentially contradicting) goals, and always randomly choose one and do an action that increases this goal, even if it harms the other goals. Even more, it should allow some feedback; for example by speaking with it you could increase a probability of some goal, and then even if it randomly chooses not to follow that goal, it would give some rationalization why. This would feel more like a free will.
On the other hand I imagine that some people with human-predicting powers, like the marketing or political experts do not believe in so much human free will (with regard to their profession’s topic; otherwise they compartmentalize), because they are able to predict and manipulate human action.
By the intra-species competition we are complex enough to prevent other people from understaning and predicting us. As a side effect, it makes us incomprehensive and unpredictable to ourselves, too. Generally, we optimize for survival and reproduction, but we are not straightforward in it, because a person with simple algorithm could be simply abused. Sometimes the complexity brings some advantage (for example when we get angry, we act irrationally, but as far as this prevents other people from making us angry, even this irrational emotion is an evolutionary advantage), sometimes the complexity is caused merely by bugs in the program.
We can imagine more than we can really do; for example we can make a plan that feels real, but then we are unable to follow it. But this lie, if it convinces other people (and we better start by convincing ourselves), can bring us some advantage. So humans have an evolutionary incentive to misunderstand themselves—we do not have such incentive towards other species or machines.
I know this is not a perfect answer, but it is as good as I can give now.
So, this is an explanation of why humans think they have free will. But it does not fully explain the debate over free will—why, after hearing this, can someone still insist that this does not resolve the question of whether they have free will or not? What’s going on in their head?
They still cannot predict themselves, so the theory that “in principle they are predictable” still feels wrong.
Predicting humans is really difficult (when they are thinking; not when they act on reflex or habit), because very small changes in input can bring significantly different output. Trying to predict oneself is trying to predict oneself predicting oneself etc. It is like trying to understand what a difficult recursive function does.
We also have psychological fear related to predictability. Imagine what would it be like if another person would be able to precisely predict your reaction. They would be able to choose an action that together with your reaction brings them maximum utility, usually at your expense. It feels safer to insist that your actions are absolutely unpredictable. This is not related only to philosophy; in the same way people claim that advertising never works on them, or they could never become cult members. People love to pretend that there is absolutely no way to influence them. Paradoxically they often blame their failures on the environment; but they mean only economical and similar influences of environment, never psychological; at most they admit the environment has bad psychological influence on their neighbors, which in turn creates “real” problems for them. Some people even deny they could do things unconsciously; they pretend everything is a rational decision.
It is worth noting that people don’t insist on having free will all the time in the same way. It is possible to convince people that they have no free will when they sleep, or when they are unconscious. It would be also possible to argue that when people are acting on habit or on addiction or in hypnosis, then their free will is weakened. And those are situations when it is easier to predict them. Having free will means having a possibility of choice that no one (including the person doing the choice) can predict. It feels like the choice is not based on laws, because a lawful choice would be predictable. It is a confusion between “I don’t see a law” and “there is no law”.
Does your model give you this feeling?
There are at least two paths I did not follow through:
The model of myself is not only imprecise, but also wrong, and it is very difficult to get it right. First, my cognitive biases and social pressures to signal lead me away from the right path; I may have socially or biologically caused blind spots in understanding myself. (For example as a social animal, I may have blocks about thinking some thoughts when I am near the bottom of social hierarchy, and different blocks when I am near the top.) Second, my knowledge influences my behavior, so I am following a moving target. My behavior is also influenced by circumstances of my life, so even my little gained knowledge soon becomes obsolete. And all the time I am influenced by wrong information from other people, which is hard to filter or ignore, because it is positively correlated with my own biases. So not only having a correct model of myself (simulating myself) is impossible, but even rough approximations are very difficult.
Related to the “free will” is a concept of “decision”. It is supposed to be the act which reduces many options to one, but it is often something else… either a socially accepted signal, or a recognition that some invisible mental process has already chosen the preferred action. Anyway, for an external observer there is a correlation between decisions and actions, so they assume causation.
And even this is not complete. I feel most certain about the part that free will is a mysterious explanation of a missing link between feeling “I could have done N different things” and knowing “but I did this one” (or extrapolated to future: “I can do N different things… but I will do only one”); and that the feeling of “I can do N different things” means “my model of myself tells me all these N different things are possible”. It is generally impossible to model oneself perfectly; it is difficult to make even approximate predictions when small changes in inputs can cause big differences in outputs; and our models of ourself are usually horribly wrong.
Or more simply, belief in a free will is a belief that “if I can’t predict myself, no one can”. Speaking about humans, it is actually true, but it is wrong to extrapolate it too far (to hypothetical omniscient beings, to hypothetical observers capable of modelling all particles in my body, etc.).
Note for many-worlds fans: Perhaps in different Everett branches I have done different things, but that is not an answer. It still requires at least an explanation why some things happened in more branches than other things; what caused that specific probability distribution instead of e.g. uniform or Solomonoff probability distribution.
I’ll take it as a “No”. Which means, according to EY, that you should keep dissolving the question.