On the Role of Free Will in Predictive Models of Human Behavior

Yaneer Bar-Yam (May 10, 2011)

Trying to understand social systems forces us to wrestle with the concept of free will. Does free will make it impossible to build models that reliably predict what human beings will do?

We have tried to do so nonetheless.

We recently built models that predict the location of ethnic violence [1] and the time of market crashes.[2] In these models we are talking about collective actions, behaviors that involve many people acting together and in the same way. Therein lies an important difference: Collective behaviors are more predictable than those of individuals.

In a sense these models do not violate or conflict with the concept of free will. The models focus on occasions when many people do the same thing, and so the model does not need to describe what each person does, only what most people do.

Clarifying that distinction can help us consider the role of free will in building models of individual or social behavior. Doing so is important both for how people view themselves and others, and how everyone views science.

The important role that free will plays in how people assess their own behavior and that of others was brought to light in a recent study in Science. A central point of the article is that people do believe that they and others have free will, and this is important to the responsibility people have for their actions. [3] Another study in Science considers the predictability of individual behavior. It turns out that using the past daily movements of an individual allows one to predict with greater than 90% accuracy where they will go next.[4] Does this mean that people think they have free will but don't?

I will argue that we should consider the importance of individual differences in a scientific discussion of free will.

Free will has been defined as 'the ability to exercise control over actions, decisions, or choices.'[5] Free will is about the process of making decisions. Free will is also about there being more than one choice that is possible. The crux of our discussion revolves around the way in which multiple possible decisions might arise. Would a decision-making process that always results in the same decision constitute free will? If there is only one decision that is possible then it would be predictable. Since free will is commonly associated to a lack of predictability, there should be more than one possible outcome. However, there are really two parts to the concept of free will that are not typically being distinguished:

  1. Free will could be that two people, under the same circumstances, would choose to do different things.
  2. Free will could be that one person, under the same conditions, might do any of several things.

The question we are answering ultimately revolves around what kinds of scientific models are or are not predictive of what people will do.

  • If the first statement of the two is not valid, we think of a model of human action as a model that predicts what each and every person will do.
  • If the first statement is valid, but the second is not valid, we think of a model of each specific person. We might start with a simple idea of such a model, but we also might consider a more in-depth type of model that incorporates their detailed history, background, parents, relationships, genetics, and any of the many unique qualities of that person. Then, we ask, can we predict what that person will do?
  • If the second statement is also valid we have to ask, what does it mean for a person to make different choices under the same circumstances? One idea about how this might happen is if the choices are made randomly. As random events are not predictable, a model wouldn't be able to predict that choice. Does that make a random choice the same as free will? This is saying that free will is the same as a person flipping a coin to make decisions.

Could there be something more? Perhaps. My objective is not to preclude other possibilities but to clarify what we already know. The key to clarifying these points is the subject of free will --- a particular person.

Assaults on the concept of free will are traditionally attributed to the idea that deterministic laws, such as Newton's laws, can predict the future from equations of motion.

There are conditions in which Newton’s Laws can be predictive of what people do, when physical impacts are fast or large enough to preclude volitional action. A crash dummy gives a reasonable representation of a person in an automobile crash.

Another deterministic idea is in the assumptions often used in neoclassical economics or economic game theory, which suggest that people respond to the world by acting in their narrowly defined self-interest, and thus are predictable. Everyone has the same goals and acts in the optimal way to reach those goals.

There are also conditions under which these assumptions are likely to hold. Especially when options are clear and outcomes and objectives unambiguous. Though even here cultural and other differences in values may cause people to make different choices.

Neither of these ideas, however, has been reliably able to predict what individuals actually do more generally. And the issue of cultural differences in values becomes more severe when we consider conditions under which individual differences matter.

When we say that a person has free will, we are talking about a quality of that person. When a person makes a choice, they might review multiple options, but the choice they eventually make is due to who they are. The qualities of that person, his or her personality, values and thoughts, affect the decisions that are made. Free will, in this sense, is a freedom of the individual to act as he or she would choose, differently from what another person would choose.

Science is advancing. It may be possible to consider improved modeling strategies. It will be important to distinguish models that try to predict what a particular individual will do from models that try to predict particular actions that all people will do.

The study in Science that says people believe they have free will can be understood from the experience of decision making. A person can review multiple options and consider them. The choice that is eventually made depends on the qualities of the person and is therefore determined by them. Because they reflect the qualities of the person, the person can be considered responsible for the decisions that are made. Other people may make different choices so the conditions do not determine the choice, the person determines them in the context of the conditions.

According to the study in Science about where people go, individual behavior is highly predictable at least as far as where they go every day: places they work, places they play, and even when and in what order they do so. Based upon knowing the behavior of that person in the past, their behavior in the future is highly predictable. This is not surprising when we think about people and what they do.

This is still not the same as predicting what one person will do based upon the environment that he or she is in, without considering that individual's history and behavior. It is also not the same as predicting what one person does from what other people are doing. In this study, neither of these is shown to be possible.

The study reinforces the idea that the essence of non-predictability in free will is, indeed, about individual differences that distinguish what different people choose to do.

This is also the reason that unpredictability in the form of randomness doesn't seem to be the crux of free will any more than having actions be determined by a deterministic process. A random choice is not predictable, but it is not a decision by the individual, i.e., an expression of that person's identity. It also cannot serve as a basis for holding a person responsible or giving them credit for that decision.

An important question that remains for the future is whether or not we can build models that capture individual differences enough in order to predict them. Whether we can do so, or more correctly for what aspects we can do so, remains to be shown.

If, as we have done thus far, we can predict collective behaviors, this is a case in which enough people do the same thing that individual differences don't play an important role. The ability to choose does not play a role because the nature of the conditions are such that free will, in the sense of different choices, does not manifest itself.

The value of respecting free will in society is tied to the concept that each person can and should be able to make decisions, be responsible for, and gain credit or blame for decisions. This is consistent with the idea that free will is about individual differences. Science can and should also recognize the importance of individual differences in analysis and modeling.

Thanks to Karla Bertrand, Amac Herdagdelen, Yavni Bar-Yam and Shlomiya Bar-Yam for helpful comments, and AFOSR, ONR and DARPA for support of our research.

[1] M. Lim, R. Metlzer, and Y. Bar-Yam. Global Pattern Formation and Ethnic/Cultural Violence, Science 317, 5844 (September 14, 2007).

[2] D. Harmon, M. de Aguiar, D. Chinellato, D. Braha, I. Epstein, Y. Bar-Yam, Predicting economic market crises using measures of collective panic, arXiv:1102.2620v1 [q-fin.ST] February 13, 2010.

[3] S. Nichols, Experimental Philosophy and the Problem of Free Will Science 331, 1401-1403 (18 March 2011)

[4] C. Song,, Z. Qu, N. Blumm, and A.-L. Barabási, Limits of Predictability in Human Mobility Science 327, 1018-1021 (19 February 2010)

[5] “whether, and in what sense, rational agents exercise control over their actions or decisions” en.wikipedia.org/wiki/Neuroscience_of_free_will

Prediction Difficulty vs. Behavior graph



Phone: 617-547-4100 | Fax: 617-661-7711 | Email: office at necsi.edu

277 Broadway Cambridge, MA USA