Voluntary Behavior, Conscious Thoughts

Skinner proposes a surprisingly easy way to dissolve the problem of what it means for an action to be “voluntary”, or “under voluntary control”.

We commonly perceive certain actions as under voluntary control: for example, I can control what words I’m typing right now, or whether I go out for dinner tonight. Other actions are not under voluntary control: for example, absent some exciting technique like biofeedback I can’t control my heartbeat or my core body temperature or the amount of bile produced by my liver.

Other, larger-scale actions also get classified as involuntary. Many people consider sleepwalking involuntary, including the bizarre “sleep-eating” behaviors some people display on Ambien and related drugs. The tics of Tourette’s are involuntary. Our emotions and preferences are at least a little involuntary: office workers might like to be able to will away their boredom, or mourners their sorrow, but most can’t.

Here “involuntary” needs to be distinguished from “hard-to-resist”. Most people do not define smoking as an involuntary behavior, because, although people may smoke even when they wish they wouldn’t, they have the feeling that they could have chosen not to smoke, they just didn’t.

The philosophy of voluntary versus involuntary behavior seems to run up against a wall when it hits the question of “what is truly me?”. If we make the reductionist identification of “me” with “my brain”, well, clearly it’s my brain controlling sleepwalking and boredom, but it still doesn’t feel like I am controlling these things. Trying to go deeper ends up hopelessly vague, usually with talk of “higher level brain processes” versus “lower level brain processes” and an identification of “myself” with the higher ones. There may be a role for this kind of talk, but it couldn’t hurt to look for something more explanatory.

Skinner, true to his quest, explains the distinction without any discussion of “brain processes” or “self”. He says that voluntary behavior is behavior subject to operant conditioning, and involuntary behavior is everything else.

It might be clearer to define voluntary behavior as fully transparent to reinforcement. Imagine a man with a gun, threatening to shoot me if I go out for dinner tonight. The fear of punishment will be effective: I’ll avoid going out. Lust for reward, too, would be effective. If Bill Gates offered me $1 billion to stay in, that’s what I’d do.

But when our masked gunman tells me to increase my body temperature by two degrees or he’ll shoot, he is out of luck. And no matter how much money Bill Gates offers me for same, he can’t make me give myself a fever either.

There is a place, too, for the hard-to-resist behaviors in all this: these are behaviors which can be affected by reward, but as yet have not been. If a masked man held his gun to the head of smokers and told them to stop or he’d shoot, they would stop. But thus far, none of the potential rewards of not smoking have been sufficient to change smokers’ behavior.


The idea of voluntary behavior is tied so intimately to the idea of the self, or of consciousness (the easy problem, not the hard one), that one would hope that a new approach to one might be able to shed some light on the other. If voluntary action depends on transparency to reinforcement, where does that leave consciousness?

I haven’t been able to find Skinner’s beliefs on this subject (when he talks about consciousness, it’s usually to deny it as an ontologically fundamental entity) and I’ve never seen anywhere near as elegant a reduction. But an explanation in the spirit of reinforcement learning would have to start by insisting on treating thoughts and emotions as effects rather than causes. Instead of explaining my choice of restaurant by saying I thought about it and decided McDonalds was best, it would be more accurate to say that previous experiences with McDonalds caused both the thought “I should go to McDonalds” and the behavior of going to McDonalds.

There is an intuitive connection between thought and language, and Soviet psychologist Lev Vygotsky made the connection more explicit; he found that children begin by speaking their stream of consciousness aloud to inform other people, and eventually learn to suppress that stream into nonvocal (subvocal?) thought.

The last post in this sequence discussed different reinforcement of thought and action. Speech and thought make a natural category as opposed to action; both are fast and easy, and so less likely to be affected by time and effort discounting. Both are point actions as opposed to a long project like learning Swahili or quitting smoking. And both bring reinforcement not through normal sensory channels (saying a word doesn’t give pleasure in the same way smoking a cigarette might, nor pain in the same way having to study a boring grammar textbook might) but in what they say about you as a person and how they affect other people’s real (and perceived) opinion of you.

So even if there is no governor anywhere unifying all thoughts and words, they may come out in harmony because they were selected by the same processes for the same reasons. And actions may not end up so harmonious, because they suffer from differential reinforcement.

Such harmony resembles the idea of a core “me”, of whom all my thoughts are a part, and who has complete power over my organs of speech—but who is sometimes at odds with my actions or emotions.

The reinforcement governing thought and speech is most likely to be internal reinforcement based on your own self-perception and on others’ perception of you. If there’s a good reason reputation management processes need to be different from decision-making processes, understanding that difference could help understand the evolutionary history of a perceived difference between the conscious and unconscious mind. One such reason is provided by Robert Trivers’ theory of social consciousness, the subject of tomorrow’s post.