Blog Coordinator

« Things to Come | Main | Tamler Sommers on the Strawsons »



Feed You can follow this conversation by subscribing to the comment feed for this post.

I was going to post this in the previous thread on luck, following Randolph's first question. But I came back after writing this, and found Randolph's new post. Since it pertains here as well, I will post it here.

This is the objection from luck as I understand it: “If a decisions or actions occur indeterministically, such that two or more alternative decisions might be made at t, each with a non-zero probability, and everything is exactly the same in a world history until t, then there is nothing about the world or the decider prior to t that accounts for one decision being made over the other. Which gets chosen is just a matter of (perhaps weighted) chance, not a matter of agentic influence on specific outcomes. If one decision should turn out better or worse than another, well, that is just a matter of luck and not a matter of agentic choice. But if decisions and consequences are just a matter of luck, then the decider cannot be responsible for the decision made or for its consequences.”

I agree that if the argument from luck works at all, it works not only against libertarians, but against everyone, including compatibilists.

But I believe that there is an escape from the luck objection. Indeed, it is the escape that evolution stumbled upon in forming our types of nervous systems.

Even if the universe is identical in all regards until the moment t of a decision, choice or outcome, decisions, choices and outcomes need not be utterly random at t. They are in part shaped by our prior acts of agency. How? Our nervous systems are capable of setting up the parameters of future responses to input before those inputs occur. As such, inputs will meet preset criteria which were imposed agentically. Since any decision will meet criteria imposed by the agent, the agent is in part responsible for any outcome that meets those criteria, even if which outcome is decided upon is also in part a matter of chance.

For example, preset parameters might be identical until t, and in one possible world a man comes across a book on Japanese first at t, just by chance, and ends up learning Japanese, and in a nearby possible world, he happens to come across a book on Russian first at t, and ends up learning Russian. This is a significant outcome for him, chancy and the fact that he came across one book first by chance was certainly beyond his control. How is he then in any sense agentically responsible for learning Japanese versus Russian? He may have made a prior decision to learn the first language he comes across when he goes to the bookstore, which preset his responses to subsequent inputs. That it was the Japanese book that he came upon first was not something he could control or is responsible for. But he is in part responsible for having the kind of nervous system at t that would respond to the Japanese book in a certain way. In a sense we are all reprogramming ourselves continually, regarding how we will respond to future input.

Agentic control and responsibility do not come from choosing one alternative over another by chance, but from having a given set of agentically chosen informational criteria in place before the decision is chancily made (because that decision happened to meet those criteria first and just by chance). Thus agency and responsibility do not arise from the particular decisions made, but from the criteria the agent sets up for making those decisions before they were ever made.

Let me add, this criterial way of making decisions based on future inputs gives rise to a two-way power in both senses. I might learn language A or B. And if an above-threshold book never comes along, I might never decide to learn a foreign language. So I might learn a language or I might not. And whether I learn language A or B, or don't learn any, is in part a matter of chance.

Granted, some criteria we do not choose agentically. They come prespecified by genetic potential and how that potential is realized by multiple factors. For example, if I am a heterosexual man, I probably could not help but want to marry a woman. I had multiple criteria and some threshold in place that any potential mate would have had to supercede. But there are nonetheless, at least in principle, infinitely many ways to meet any finite set of criteria. That I married woman A versus B was in part a matter of good or bad luck. But it was not solely a matter of luck, because had a mouse or a kangaroo come along, I would not have married her. But since I did not agentically preset the criteria for marrying a woman (this was, say, preset in me from birth), I cannot be responsible for marrying a woman (e.g. rather than a tree or kangaroo). Responsibility only arises when the preset criteria were set in part agentically, which is to say, the chooser had some control over the criteria that were set.

Thanks for this post, Randy.

I can imagine someone assuming that powers are generally multi-track. They may then think that their being multi-track will help us understand two-way powers. But I don't think it'll help.

For those not familiar with the literature on powers in ontology, many regard individual causal powers as best regarded as multi-track and directed at “endless manifestations with an infinity of present or absent, actual or nonactual alternative disposition partners” (C.B. Martin *The Mind in Nature*, 29). Martin suggests thinking of the projectivity of an individual power as constituting a complex web, which he calls a “Power Net." Consider the sphericity of a ball. This property of the ball is capable of diverse manifestations depending upon the property with which it comes into contact that serves as a disposition partner. The ball will roll if it comes into contact with a solid surface. It will leave a concave, reversible impression if it comes into contact with a surface with the appropriate elasticity. And so on. The same disposition is manifested in different ways with different partners. But this is not what some folks have in mind when they speak of "two-way powers."

Turn to the power to decide. Assuming that this power is like (most) other powers, this power will then be manifested in different ways depending upon the manifestation partner(s) it pairs up with (in this case, they may be the powers constitutive of our reasons for action). It seems that, if this power is like most of the powers of an object, once the powers are paired up, we've got a guaranteed outcome. So given the exact same manifestation partners in decision-making, we'd always have the same outcome.

So, what about the idea of the power to decide as a "two-way power"? Such a power would be a special power indeed. It would be a power that can be manifested in different ways with the exact same partner(s).

But you've noted that this may not be a very special power. What about the powers of the photon and the slit you mention, which can be manifested differently even with the same manifestation partners?

Let's assume that this is a "two-way power" in the relevant sense. I am not sure what to think of such cases. But it would seem that our power to exercise rational agency would not be like the powers of these objects, especially if we exercise rational agency in virtue of possessing this power. If it were like the power of the photon, then the exercise of free agency would seem unintelligible (the problem of proximal luck would thus rear its ugly head).

So, like yourself, I cannot for the life of me figure out how a rational two-way power in virtue of which we exercise free agency would work.

Hi Randy,

Interesting post! In terms of the powers that might be relevant to intentional action and agency, it seems the most appropriate conception of a power would be akin to the second one that you described, viz., a power that is manifested or not, depending on the circumstances. This notion seems most appropriate because the kind of power in question should be one whose manifestation or lack thereof is controlled by the agent, not by the circumstances in which the agent finds him- or herself.

Regarding the question with which you ended your post, perhaps you could clarify two points:

Are you asking how there could be powers whose manifestations are independent of the *particular* circumstances at hand (e.g., the manifestation would have occurred differently in different circumstances), or how there could be powers whose manifestations are independent of *every* circumstance (e.g., the manifestation would have occurred regardless of the circumstances)?

And could you say a little bit more about the conception of "chance" (or "chancy"-ness) that you have in mind?


Michael Brent

Thanks for the comments.

Peter: Your proposal concerning the problem of luck is very interesting. It strikes me as similar in some respects to Al Mele's. But I meant to raise a different (even if related) issue. Are agential powers different, in some basic metaphysical way, from other powers throughout nature? The kind of adaptation you describe seems similar to what is modeled by so-called neural network computer programs, which can run on machines that aren't really thinking things at all. I'd suppose that when such a machine runs such a program (NETtalk, for example), it has a power like the ones you describe. This suggests that the kinds of powers we have here aren't unique to agency. But perhaps they are required for it.

Andrei: My 6th paragraph was meant to take in a view like Martin's, which seems to me very sensible. (Though Martin had a puzzling way of accommodating indeterminism, with a flutter of deterministic powers.) In my case, it's not so much the two-way-ness of agential powers that I find puzzling, but rather the respect in which they're supposed to be metaphysically distinctive. Of course, they're powers of AGENCY, and thus they involve various fascinating psychological capacities. But everyone can agree on that.

Michael: Nature might be replete with powers that might be manifested or not, given all circumstances in which they exist on some occasions. The things that manifest them--the objects that have the powers in question--might be said to control whether they're manifested or not when they are; after all, it is these things that manifest the powers then. What I don't see is how powers of agents are supposed to be special in this respect.

Toward the end of my post, I considered powers that are indeterministic in the manner I just described (or in that they might manifest this way or that given the same circumstances). I had in mind a broad notion of chance that doesn't require determinate objective probabilities. If it is compossible with the laws and all of history prior to t that an E event occur at t, and also compossible with the laws and all of history prior to t that no E event occur at t, then it's a chancy matter whether an E event occurs at t. Seems to me a fair understanding of 'chancy'.


Thanks for clarifying.

I guess I don't think any such powers would be metaphysically distinctive. They're like other powers. But they are just directed at different outcomes than other powers. In particular, they'd be directed at producing certain outcomes when they partner with reasons for action. That's the only thing that I can see about any such powers that makes them special.

I'll go further and say that I think that agency of various sorts (not all of it is intentional, some we can call "quasi-intentional") is evinced throughout nature and not just among humans and other animals. So I take it that what we find in rational agents is only metaphysically distinctive, again, because their manifestation partners would include reasons for action.

I guess what I'm curious about (notwithstanding my puzzlement over rational two-way powers) is why we should expect such powers to be metaphysically distinctive.


I’m thinking that the powers of all living things are a little bit different from one another, and there isn’t anything absolutely special about the powers of human agents. Our powers are simply part of the spectrum. It’s interesting to think about different powers associated with different entities in nature, but it’s probably going to be difficult to directly relate to any powers other than our own. I say that because we only have five senses, and when using those senses we only perceive of the *results* of other powers after those powers have been exerted. That’s probably why it’s taking science some time to realize that other entities in nature have real powers.


You mentioned 2 conceptions of a two-way power: (i) a power that can be manifested this or that way, or (ii) or a power that can be manifested or not. Given what you say after introducing the distinction, I believe there is a third conception available. According to it, a two-way power is the power to do something and to refrain from doing that thing. This, I believe, is what dual-power theorists have in mind. Or, at least, it is what I would say if I were one of them.

Notice that this kind of power, if it exists, is different from a type (i) power. It is not, as you put it, “a generic power,” as being elastic or being fragile is. And it is not a type (ii) power either. Here, refraining is the manifestation of the power; it isn’t a case of the power being inert.

In fact, if this kind of power exists, I think it would be distinctive in the sense that you have in mind in your response to Andrei and Michael. These powers would seem to be different in kind from the ones that Martin and other commonly invoke to explain run-of-the-mill causation.

Randy, what is agency? What is the relationship of a power to agency? What is an agentic vs. non-agentic power?

Not being sure about the precise meaning of 'power' as philosophers use it, I will limit my comments here to the issue of agency. I would be curious to hear your definition of “an agent” or at least a specification of those characteristics that make something an agentic choice or decision rather than a non-agentic one.

Coming from a neuroscientific rather than philosophical background, I tend to think in terms of neural circuits that process information, make decisions, and then send outputs such as commands to motor circuits and ultimately muscles that do things. From that perspective, a weak definition of agency might be the capacity to control the outputs of other information-processing circuitry or the muscles to fulfill goals. But is such a weak definition of ‘agency’ as goal-directed control enough? I will argue below that the answer is ‘no’ and that we need to bring consciousness into our definition of ‘agency.’

Why? There are numerous instances in the psychological literature of patients with goal-directed behaviors over which they can exert no conscious control. One is “utilization behavior” ( where a person automatically grabs objects within reach and uses them appropriately, regardless of the context. For example, if they see a comb, they will pick it up and comb their hair with it, regardless of inappropriateness of the context. This is certainly appropriate usage of an object, but if it happens in, say, a courtroom while in the jury box, it is an action inappropriate for the context. Interestingly, some such patients do not feel that they are acting abnormally. Nor do they typically report that the actions are happening as the result of some other agent. That is to say, they feel that they commanded it (whatever ‘they’ means), even though we know that it is an automatism which they could not have helped but enact. On your view Randy, would picking up the comb and combing one’s hair automatically be an agentic act or a non-agentic act?

A more extreme example is Alien Hand Syndrome (, where some body part, typically one of the arms/hands, seems to take on a life of its own, appearing to have its own agenda that ‘goes against one’s will.’ Would the actions of an alien hand count as agentic on your account? Interestingly, such patients typically report that the hand is not subject to their will, being driven by some agent who is alien to themselves (again, whatever ‘self’ refers to here).

Because of such cases, it would seem that central to any definition of ‘agency’ in a stronger sense must be something more than just ‘goal-directed,’ ‘purposeful’ or ‘capable of controlling subsequent mental acts or motoric behavior.’ It would seem that ‘subject to *conscious* control’ should also be central to any definition of strong agency. Otherwise, goal-directed automaticities would count as agentic.

Certainly, the kind of neuronal criterial resetting and randomness I advocate in the first nine chapters of my book are at best necessary but cannot alone be sufficient for the physical realization of mental causation and a strong free will, because without a conscious agent with intentional states that could exploit such criterial resetting to fulfill its ends, we might be no more than mindless, unpredictable “zombies.” So if agency means conscious agency and conscious control (and 'power' refers to conscious powers to volitional do this or that), it would be good to state that explicitly.

In the final chapter of my book I argued that consciousness plays a key role in agentic mental causation by providing a common format for endogenous attentional and other executive operations that permit the assessment of possible behaviors and thoughts against the highest-level criteria held in working memory for successful attainment of goals and fulfillment of desires. Qualia, I argued, are those representations that can be operated on by endogenous attention, giving rise to the possibility of volitional attentional tracking (e.g. tracking whichever bird in a flock that one wills to), which, I argue, cannot happen in the absence of consciousness. Because certain operations can only take place over conscious operands (i.e. qualia), and motor acts can follow and enact the conclusions of such operations, such mental operations and representations play a necessary causal role in their mental and motoric consequences, and are not mere illusions of volition. Thus, consciousness or qualia play a necessary causal role in agency. On this account, current operations in working memory over endogenously attendable operands do not cause present thoughts, experiences or actions, but instead play a causal role in future thoughts, experiences, and actions, given the arrival of inputs that satisfy the criteria that have been preset.

Thus I believe that volitional or ‘endogenous’ attentional operations over the domain of representations accessible to them (which is the conscious set of information maintained in the brain in various short-term and working memory buffers), are central to what we mean by agency. Without consciousness, there would be nothing for volitional attention to operate over, and therefore no agency. There might still be goal-directed zombie processes. These could even operate at a high semantic level of action programs like ‘if see comb, comb hair,’ but these would not count as agentic in a strong sense, even if we might want to count them as agentic in a weak sense.

Finally, I think it is not a coincidence that such automaticities often arise from damage to the core information-processing circuitry involved in both volitional attention, executive self-control and the mental workspace, namely, lateral prefrontal cortex, the anterior cingulate (ACC) and posterior parietal cortex. The ACC is particularly interesting in this regard. I would be happy to discuss its possible roles in agency and executive control if this is not getting too neurosciency for the philosophical readers here.

I suspect that this is getting too long and neurosciency though, so I will stop here and ask the philosophical community on Flickers first, what counts as agency in the philosophical traditions, and second, is consciousness or volitional attention in any way deemed to play a central role in that conception of agency? Thank you in advance for any replies.

Santiago: yes, this is what some proponents of the idea say. I wonder how it isn't covered by conception (i).

Suppose I raise my arm. Suppose I then manifest a power, it is a two way power, and I could have manifested it in refraining from raising my arm.

This seems to be an instance of (i) a power that can be manifested this way or that. It was actually manifested in my raising my arm. It could have been manifested in my refraining from raising my arm.

One might say: refraining isn't acting. It is omitting to act.

I agree that in many cases refraining isn't acting. Still, if the refraining would have manifested a power, a power that was in fact manifested in raising my arm, it seems we have an instance of (i).

Perhaps it is a special kind of instance of (i), a kind that can exist only with agency.

Peter: you raise a difficult question--what is agency--and I don't have an answer to it. Apparently the proponent of the view I'm examining in this post needs an answer, since the view is that there are metaphysically distinctive powers manifested only in agency.

It seems to me that philosophical theorizing about action commonly focuses on paradigm cases and attends little to the borderline cases. I'd like to see more attention to the latter. I wouldn't myself venture a verdict on them without knowing more than I currently do about real cases of this sort.

I certainly agree that consciousness plays a role in many instances of agency. Is it your view that there is no unconscious agency?


Thanks for your reply.

I agree. One way of thinking about the conception of a two-way power sketched is to treat it as a special case of i). That was my original inclination. I didn’t pursue this line, because your comments seemed to suggest a narrower understanding of i) in terms of generic powers.

Anyhow, even granting a broader understanding, this kind of power to do and to refrain would be special; only agents could have type (i) powers of that sub-type.

But, perhaps, the contrast can be strengthened if one thinks that refrainings are, ontologically speaking, different from doings. (I think you think this). Then, it would seem that agentic two-way powers would not be cases of regular two-way powers of doing this or that.

Again, that’s what I would say if I were a two-way power theorist—which I am not!


>Is it your view that there is no unconscious agency?

Christof Koch speaks about 'zombie agents' in his excellent book 'The Quest for Consciousness.' These are neural circuits that can execute goal-directed actions without involving consciousness. They are typically automatic, not subject to voluntary control, and are enacted rapidly and ballistically. There are whole families of zombie agents governing our behavior, thoughts, dreams, and emotions. For example, people will respond to images of snakes with physiological markers of fear even if the snake images are masked so rapidly that they cannot be consciously experienced. Even whole hierarchies of nested goal-seeking action sequences can occur without consciousness, as occurs in sleepwalkers who have even been known to drive to certain destinations in order to carry out seemingly complex intentional acts, such as engaging in sex.

It is important to distinguish non-voluntary zombie agents or z-agency from the c-agency that depends on voluntary control of operations over conscious operands. Volitional control and c-agency take place over and require conscious representations. C-agentic circuitry that operates over conscious operands has the opposite properties of unconscious zombie agents. C-agency is not automatic, is subject to voluntary control, is relatively slow and enacted non-ballistically, which is to say that it is subject to online modification via feedback. Despite its slowness, the great benefit offered by such c-agentic processing is that it affords flexibility and the consideration of multiple courses of action. C-agency is evident in the playing out of multiple possible courses of action in the mental workspace or imagination. For example, if I ask you to plan a dinner party, you can imagine various guest combinations, switch people out, and play out various meals and conversations in the 'virtual reality' afforded by the conscious experiences fashioned in your imagination. This is slow to be sure, but permits non-ballistic, iterative and open-ended consideration of options until a solution to the problem of who to invite has been solved.

Evolution has provided us with two classes of solutions in executing actions. There is a rapid but inflexible set of z-agents, and a set of c-agents that are slow but flexible. Besides imagining outcomes, another c-agentic operation would be the voluntary control of attention. Attention can be voluntarily allocated to any portion of any space that we represent based on almost any imaginable criterion held in working memory. For example, I can will to attend to and keep track of any person I choose in a given room, even if I don't necessarily look at them. I can maintain almost any arbitrary criterion, such as 'keep track of people wearing red tops, and I can then execute such attentional tracking. Similarly, I can choose to attend to the oboe in a symphony, then attend at will to the violin. This is to say that c-agency is not primarily stimulus-driven, and it is also not automatic or reflex-like. Instead such c-agency is free in the sense that it is subject to volitional control and is not slavishly driven by the input.

Having two classes of agents realized in our nervous system allowed us to have the best that both types have to offer. For decisions and actions that do not require flexibility of response, there are the rapid ballistic and automatic responses selected by zombie agents. For those that require flexibility, such as planning what to do on the weekend, there are conscious operations such as volitional attentional tracking.

So, to answer your question Randy, I think slow but flexible c-agency requires consciousness, whereas rapid but inflexible z-agency does not require consciousness.

Peter, some intermediate cases are interesting. I drive a manual shift car. Driving to work today, I depressed the clutch pedal many times. How many? I have no clue. I don't recall doing it once. I was never attending to it. But it doesn't seem exactly like the things you call zombie actions. I was conscious of various things while I was doing it, and I could have shifted attention to it and done it consciously.

It seems to me that a great deal of our agency is like this.

Hi Randy,

Thanks for the reply. I agree: it is not obvious that, in this specific respect, the powers of agents are unique among the powers found in nature.

As I understand the distinction, a one-way power manifests in one specific manner: when the circumstances for the manifestation of the power obtain, the manifestation of that power occurs. And a two-way power manifests in a different manner: when the circumstances for the manifestation of the power obtain, it is nevertheless up to the agent to determine that the manifestation occurs. Maria Alvarez, gesturing towards Aristotle, speaks this way in her 2013 paper on agency and two-way powers.

The label “two-way” is slightly misleading, though, since it suggests that the manifestation of that power can take place in two ways, which is not quite correct. What seems to be of importance is the fact that the agent determines that the manifestation of the power takes place, not the circumstances. In this respect, depending on how one describes the way in which the agent determines that the manifestation of the power takes place, the powers of agents could be quite unique.


Michael Brent

Randy, the intermediate cases you speak of, like driving a car on 'auto-pilot,' are often cases where once attentionally demanding processes have been automatized through practice, repetition or mere repeated exposure. It appears c-agency is not only slow and inefficient though flexible, it is also expensive in that it burdens limited attentional resources. It appears the nervous system automatically automatizes whatever it can, including executive decisions that no longer require an exhaustive and exhausting volitional search among many options. Once the best option can be selected without consideration of alternative options, why not automatize that selection and execution? It appears that z-agency replaces c-agency wherever possible because z-agency is faster and cheaper in terms of conserving attentional resources. If need be, c-agency can take such processes over again, as when we attend to the clutch in your example, or transition from automatic, unconscious to conscious breathing subject to volitional control. But the default mode appears to be automatization and zombification wherever attentional resources can be spared. A danger of this is that we operate in zombie mode inappropriately and make mistakes because of inattention. My favorite example of that is:

The case of breathing is an interesting one. I can intentionally breathe in and out. But it seems to me that ordinarily my breathing isn't action at all--not zombie action or any other kind. Isn't breathing ordinarily controlled in the brain stem? I seem to recall reading that someone can keep breathing after loss of their entire cerebral cortex.

Randy, please define what counts as an action in your view. On a general account, isn't anything involving a neural command that triggers the contraction of muscles an action? Why should motor commands that originate in cortex count as actions, but those that originate subcortically not count as actions?

I'd think it would take more than muscle contraction caused by neural command. I'd want to rule out constriction of blood vessels in thermoregulation. And I wouldn't think the brain location of the motor commands was essential. Exactly how to draw the line seems to me a hard question.

One place to draw the line might be at intention. Some brain commands that result in muscle contraction follow the having of an intention to act in a way that directly or indirectly realizes some goal, while others do not. But if this is the dividing line between agentic and non-agentic action, then we are left trying to define what exactly we mean by the having of an intention. a useful summary of philosophers' attempts to do just that is here ( If intentions require agentic circuits in the brain that entertain goals and purposes that acts should then realize, then it would seem that we would want to limit such neural circuitry to those that represent goals. This would seem to at the very least to require possessing information about likely causal links between self-actions and outcomes in some representational space, whether of objects and the body in the world, or of the consequences of speaking in a representation of social agents (other minds) in the social world. Placing the dividing line at intention in this way would rule out the rather thermostatic regulation of blood vessels or the contraction of muscles in the gut. Yes, those actions do realize goals like warming the body or digesting food more efficiently. But there was no entertainment of goals in a representational space that permitted the consideration of alternatives or consequences of various alternatives. Placing the dividing line at intention might actually end up limiting agency to that realizd in neural circuitry found in the cortex.

The comments to this entry are closed.

Books about Agency

3QD Prize 2014: Marcus Arvan