# Agency and Embodiment: Groups, Human--Machine Interactions, and Virtual Realities **Authors:** Johannes Himmelreich **Published in:** Ratio 31(2), 2018 **DOI:** [10.1111/rati.12158](https://doi.org/10.1111/rati.12158) *This is the author's manuscript. Please cite the published version.* --- # Introduction Up until autumn 1989, when Germany was divided into East and West, soldiers at the Inner German border had orders to shoot fugitives. During 30 years, 67 were shot trying to escape in Berlin alone (Hertle and Nooke 2009, 18–25). Following Germany’s reunification, the border soldiers have been prosecuted and convicted for manslaughter and murder. In addition to the soldiers, the courts also convicted military and party leaders. They too were convicted for manslaughter and murder. This is common legal practice.[^1] This practice is puzzling. It is a widely held view in philosophy of action that intentional bodily movements are a good guide to agency. When a person moves her body, the resulting action is something she did. Call this view the *embodiment view*. According to this view, when a soldier pulls the trigger to shoot a fugitive, then this soldier is the agent of the pulling and the shooting. Legal practice stands in opposition to this view because, despite resting in his bed, the courts held the commander responsible for the shooting. The commander did not move his body in the shooting of the fugitive and could not have been the agent of killing. It is puzzling that courts convicted military and party leaders for manslaughter and murder. Following the embodiment view there is no such action at a distance.[^2] Following the embodiment view, there is also no action by groups. It stipulates that doing something requires a body. A group does not have a body and therefore a group cannot do anything, it cannot be an agent. The embodiment view rules out group agency. In this paper I investigate the view that agency has to be embodied. I argue that this view is mistaken. The paper is structured as follows. After giving a minimal definition of agency, I give an account of the embodiment view. I then argue that in some cases the embodiment view contradicts the definition of agency. This leads to the embodiment trilemma. Any one of the three parts of the embodiment view can be given up to escape the contradiction. In the remainder of the paper I discuss objections and defend one particular escape route. # Agency This paper focuses on the notion of someone *being the agent of* something. This is a relation between agents on the one side and actions on the other. The relation has two relevant properties. First, it requires that for each action there is an agent. There cannot be an action without an agent. Call this property *surjectivity*. Second, the relation obtains between an agent and an action only if the agent has *intentional states* that are adequate for seeing to it that the action is performed.[^3] The argument that agency does not need to be embodied relies on these two assumptions. is a relation between agents and actions, such that for each action $a$ there exists an agent $i$, and $i$ has intentional states adequate to seeing to it that $a$. Surjectivity is a plausible assumption. It seems to be a conceptual truth about actions that they require an agent.[^4] Intentional States is also a plausible assumption, when attention is restricted to intentional actions. Given a focus on intentional agency, it seems to be a conceptual truth that the agent needs to be capable of having intentions. # The embodiment view Paradigmatic examples of agency suggest that actions involve bodily movements. Pouring milk into your coffee would be one such example, shutting the door another. Some might even go so far as to argue that not just paradigmatic examples but that *all* agency involves bodily movements. But that can’t be right. There are *mental actions*, such as making a decision, adding numbers, listening to music, or paying attention not to spill the milk that do not involve bodily movements. Furthermore there are *omissions*, such as staying in bed, not helping your friend, or not going to class. By definition, omissions – while perhaps involving the body – do not involve bodily movements. Leaving mental actions and omissions aside, it seems plausible that all other actions necessarily involve bodily movements by the agent.[^5] This is the view that agency is *embodied*.[^6] Accordingly, it is perhaps not surprising to find passages in the literature suggesting that agency is embodied. I will consider some of those passages and then formulate embodied agency as three conditions. A prominent statement of the view that agency is necessarily embodied can be found in Davidson (1971). > \[A\]ll primitive actions are bodily movements. (Davidson 1971) There is at least a whiff of body chauvinism in this quote. Taking Davidson (1971) at face value, he claims that agency necessarily has to be embodied. While Davidson wants to be very permissive about what counts as a “movement” (including omissions and mental actions), he still adds the modification “bodily”. Even if we are very permissive about what counts as a movement, this modification suggests that it always has to be the body that moves. Given his theory on the individuation of actions, it’s plausible to understand Davidson as claiming that actions are bodily movements. This view that actions are identical with certain bodily movements is also defended in Melden (1956). > It appears as though an action were a bodily movement of a special sort and that we only need to specify the distinctive features of bodily movements that count as actions in order to elucidate the concept of action. (Melden 1956, 523). Some statements on the connection between the agent and her body are ambiguous between two notions of necessity. Under a broad notion of *conceptual necessity* what is necessary must be true at all metaphysically possible worlds. Identity statements are metaphysically necessary. This also the sense of necessity relevant here because the project is to clarify the concept of agency. More narrowly, there is the notion of *technical necessity*. For technical necessity, what is necessary must be true at all possible worlds that could be brought about from the actual world focusing effort and technology available at present. For example, flying to the moon was technically possible by the mid 1960s. It was conceptually possible long before. Consider now Goldman (1976). Some quotations could be read as referring to either technical or conceptual necessity. Consider Goldman (1976). > \[The concept of basic act-type\] may not be restricted exclusively to “bodily-movement” properties, such as raising one’s hand, turning one’s head, lifting one’s foot, etc. For the sake of simplicity, however, I shall henceforth assume that basic act-types include only bodily movement acts of this sort. (Goldman 1976, 68) The locution “\[f\]or the sake of simplicity” could be taken to indicate that he is not really talking about conceptual necessity here. Or consider Dretske (1991). > \[T\]he things one does are usually achieved by some change (or internally caused absence of change) of one’s body. One initiates a lawsuit by calling one’s lawyer (and this, in turn, by lifting the telephone, and so on), cancels an order by nodding when asked, and refuses an invitation by writing a note. Or one does it in some other way. But *the way* always involves some bodily movement or change. (Dretske 1991, 14) Qualifying his statement with “usually” suggests that Dretske (1991) has merely technical necessity in mind. Finally, perhaps even Smith (2004) could remotely be understood as talking about technical necessity and not conceptual necessity. > \[I\]t is a truth of action theory that, though agents do all sorts of wonderful things, and all sorts of awful things, they are able to do these things only in so far as there are things that they can just do, movements they can make with their bodies. (Smith 2004, 228) All these quotes can plausibly be read as using necessity in the sense of conceptual necessity. Goldman (1976) gestures at alternatives to “‘bodily-movement’ properties” but leaves it open what these alternatives could be. He can be understood as claiming that bodily movements are conceptually necessary but “\[f\]or the sake of simplicity” omitting the argument for this claim. Or consider Dretske (1991). While “usually” suggests he mean necessity in the technical sense, just a sentence later he writes that bodily movements are “always” involved in actions. “Always” can be read as indicating conceptual necessity. Finally, consider Smith (2004). Nothing indicates that he understands basic actions, “things that they can just do”, other than bodily movements in the sense of conceptual necessity. This reading is plausible also because it squares well with Smith’s overall Davidsonian sympathies. The quotes so far have not been explicit about the notion of the *body*. It is sensible to assume that unless an author explicitly theorizes what he or she means by the body, they mean the biological structure. So in talking about bodily movements, they talk about movements of the biological structure. I refer to this notion of the body by saying that the body is *individuated narrowly*. Tuomela (1989) is an example of using the embodiment view for an argument against collective agency.[^7] How can this view be summarized? The claims in the quotes above suggest three necessary conditions.[^8]   Being the agent of $a$ requires performing an action. Performing an action requires a bodily movement. The body is individuated narrowly. Paradigmatic cases of agency meet all three conditions. If I pour milk in my coffee then my biological body (E3) moves, such that there is an action of pouring the milk in the coffee (E2), of which I am the agent (E1). I can be the agent *of* pouring the milk in the coffee since I meet the necessary conditions for Embodied Agency. # Agency without embodiment There are cases where the minimal definition of agency, of $i$ being the agent of $a$, is inconsistent with Embodied Agency. In this section I give a reductio argument by counterexample. It shows that there can be instances of agency that cannot meet all three conditions of Embodied Agency at once. Consider the following unproblematic case of agency. It is unproblematic because it does not lead to an inconsistency between the minimal definition of agency and Embodied Agency. Twin Jim Jim wants to break a window of an old garden shed. He picks up a stone, throws it and the window breaks. Just as any paradigmatic case of agency, Twin Jim satisfies Surjectivity, because there is an agent (Jim) for the action (breaking the window). The case meets the Intentional States assumption because Jim wants to break the window. Finally the three conditions of Embodied Agency are met because Jim has a biological body (E3), which moves to perform an action (E2), so that Jim is the agent of the window breaking (E1). Imagine a variation of the Twin Jim case. Before long, technology will be available to build transmitter implants and tele-controlled robots. Imagine Jim commanded such technology. Jim Jim wants to break a window of an old garden shed. He has a ball machine, which he can control with his mind. He gets used to the machine and learns to control it. Jim fills the machine with stones and walks off. The machine fires and the window breaks. The only difference between the cases of Jim and Twin Jim is that the former uses a ball machine as a tool to break the window. The difference between the machine and a conventional tool is that, unlike hammers, guns or screwdrivers, Jim’s body does not need to move in order to operate it. Apart from this use of a tool, the two cases are symmetric. The use of a tool should not make a difference to the judgment of whether or not there is an intentional action. Therefore, Jim is the agent of the window breaking. However, following the embodiment view, Jim is not the agent of the window breaking. Jim does not meet all the conditions of Embodied Agency at once. This is a contradiction. I will go through it in more detail. Jim *is* the agent of the window breaking. This is established by Surjectivity and Intentional States. Let $a$ be the event of the window breaking. If $a$ is an action then there has to be an agent. Either Jim or the machine are plausible candidates for agency. The machine cannot be the agent of $a$ because it does not have intentional states. Therefore, Jim must be the agent of $a$. He meets the Intentional States requirement because he “wants to break the window”. However, Jim is *not* the agent of the window breaking. Jim does not meet the three conditions of Embodied Agency simultaneously. By the hypothesis of the case, his body does not move (E3). However, a bodily movement is necessary for an action (E2), and an action is required for being the agent of $a$ (E1). Hence, it follows from Embodied Agency that Jim is not the agent of the window breaking. This is a contradiction. On the one hand, the minimal definition of agency leads to the conclusion that Jim *is* the agent of $a$, on the other hand, Embodied Agency leads to the conclusion that Jim is *not* the agent of $a$. Since the minimal definition of agency together with Embodied Agency lead to a contradiction, one of them needs to be given up. The minimal definition of agency should not be given up. It consists only of two assumptions, each of which seems to be a conceptual truth. Therefore, Embodied Agency must be false. In effect, since Embodied Agency consists of a conjunction of three claims E1 to E3, one of these three claims has to go. [^9] # Embodiment Trilemma Giving up one of the three conditions of Embodied Agency is sufficient for avoiding the contradiction. There is one escape route for each condition in Embodied Agency. Hence, the argument constitutes a trilemma. The following are its three horns. Deny E1, the condition that there is a necessary connection between action and agency. Absent such a necessary connection between action and agency, Jim can be the agent of $a$ without $a$ being his action. Jim would be the agent of the window breaking, but the action of breaking the window would be the machine’s. It is also the machine that performs the bodily movement. The remaining conditions E3 and E2 would be met. E3 would be met because the machine’s body can be narrowly individuated, leaving aside that it is not a biological body. And E2 would be be met because the machine’s body moves in the machine performing its action. Jim can be the agent of the machine’s action because E1 is denied. Deny E2, the requirement that a bodily movement is necessary for performing an action. Breaking the window would then be Jim’s action despite his body not moving. Instead, the movement involved in the action would be the machine’s. The machine’s movement and the window breaking would be Jim’s action. He would also be the agent of the machine’s movement and of the window breaking. However, his body does not move. The remaining conditions E3 and E1 would be met. E3 would be met because Jim’s body is individuated narrowly. E1, that an action is necessary for agency, would be met because Jim performs an action of which he is the agent. Deny E3, the view that the body is individuated narrowly. In this case, Jim’s body would go beyond his biological structure. The machine would be part of Jim’s body. Jim would indeed perform a bodily movement even though his biological structure would not move. The machine’s movements would be movements of Jim’s body. The remaining conditions E2 and E3 would be met. E2 would be met because for the action of breaking the window, Jim’s body moves in virtue of the machine moving. Also E1 would be met because Jim performs an action of which he is the agent. When one after another condition of Embodied Agency is denied in turn, the boundary between agency and movement continuously shifts (see Table 1). Denying E1 leads to the result that Jim is the agent of $a$, but the action and the movement is the machine’s. Denying E2 leads to the result that Jim is the agent of $a$, $a$ is Jim’s action, and the machine is performing the movement of this action. Denying E3 leads to the result that Jim is the agent of $a$, it is his action, and his movement. These three escape routes lead to substantively different judgments about who is the agent of $a$, whose action it is, and whose movement is involved. |   | **Movement** | **Action** | **Agency** | |:--------|:-------------|:-----------|:-----------| | Deny E1 | $M$ | $M$ | $J$ | | Deny E2 | $M$ | $J$ | $J$ | | Deny E3 | $J$ | $J$ | $J$ | Mapping of whose movement, action and agency is involved in the window breaking, depending on which condition of Embodied Agency is denied. $M$ is for the machine, $J$ is for Jim. How should the trilemma be solved? Different escape route lead to substantively different results. I will defend that escape route two should be taken, i.e. that E2 should be denied. Before making this argument, I offer an interpretation of the three escape routes and discuss some objections. # Interpreting the Escape Routes The trilemma becomes more tangible when the escape routes are given an interpretation. In short, the interpretations of the three escape routes are the following. By denying that there is a necessary connection between action and agency, the first escape route says that agency and action are two different concepts. This conflicts with the very natural view that agency and action are related in the same concept. The second escape allows that there can be agents of movements that are not theirs. Hence the second escape route leads to a liberal position on how agents are individuated. Metaphorically speaking, one’s agency can extend to the movement of other’s. For example, Jim’s agency might extend over the movements of the machine. The third escape route, by denying that bodies are individuated narrowly, leads to a liberal position on how bodies are individuated. For example, the machine might be a part of Jim’s body. ## Conceptual Rift The first escape route is to deny the necessary connection between agency and action. On this escape route, when this connection is severed, agency and action can come apart. Jim can be the agent of actions that are not his. Some may argue that intuitively agency and action cannot come apart because they form one concept. To reflect this intuition, call this escape route *Conceptual Rift*. Consider how agency and action form one concept. Agency and action are one concept just in case the two predicates concerning agency and action have the same extension. Agency and action form one concept if and only if for any individual $i$ and any action $a$ “$i$ is the agent of $a$” is equivalent with “$a$ is an action of $i$”.[^10] On the first escape route agency and action do not form one concept. On the first escape route the movement of throwing the ball is an action of the machine. Despite the fact that the action is the machine’s, Jim is the agent of this action. Let $i$ denote Jim and $a$ denote the action of breaking the window. It is true that Jim is the agent of breaking the window. However, it is not true that breaking the window is an action of Jim. The two predicates concerning agency and action are not equivalent. Hence, on the first escape route agency and action do not form one concept. Of course, it would be important to know how precisely the two concepts differ. What is it that makes somebody the agent of $a$ while $a$ is not his or her action? Which condition is it that the one concept has but the other lacks? These questions need to be answered elsewhere. ## Extended Agency, Extended Body The second and the third escape route are about individuation. On the second escape route the *agent* is individuated. The contradiction is avoided by individuating the agent more broadly than her body. On this route, Jim is the agent of some movements of the machine. The machine is a part of the agential system “Jim”, as it were. However it is not a part of Jim’s body. On the third escape route the *body* is individuated. The contradiction is avoided by individuating the body more broadly than the biological structure. On this route, movements of the machine are Jim’s movements. The machine is a part of his body, as it were. Call the second escape route *Extended Agency* and the third *Extended Body*. There are familiar examples for Extended Body to illustrate the idea that the body is individuated broadly. In defining Extended Agency and Extended Body I shall therefore start with the latter. Consider Captain Hook or the sprinter Oscar Pistorius. Both have their bodies extended by prostheses. Captain Hook has his hook, which arguably could be individuated as a part of his body, and Oscar Pistorius has prosthetic legs. When the hook is taken to be a part of Captain Hook’s body, then his body is individuated broadly. The hook is not part of the biological structure of Captain Hook. However, E3 requires that the body is individuated narrowly as the biological structure. Hence, the hook is a part of Captain Hook’s body only if E3 is denied. When the hook is a part of Captain Hooks body, then his body extends beyond his biological structure. The metaphor of “extends” can be made more precise on a set theoretic way. Assume that two sets are given. Both are subsets of a total set of all objects. The first set contains all the things that are individuated as Captain Hook’s body. This set contains at least his biological structure. The second set contains all the things that are individuated as the environment around Captain Hook. Let this set contain only things that are not part of his biological structure. Captain Hook’s body is extended just in case these two sets, the body and the environment, are overlapping.[^11] Consider his hook. It counts as Captain Hook’s body but it is not a part of his biological structure. The hook is in the environment of the biological structure. Since the hook is a part of his body and a part of his environment, Captain Hook’s body is extended. Consider another example of Extended Body. Take Oscar Pistorius’ prosthetic legs. They are in the environment of his biological structure. Nevertheless they might be considered a part of his body. In this case, his body and the environment are overlapping, his body is extended, it extends beyond his biological structure. Extended Agency is defined analogously. Again, assume that two sets are given such that both are subsets of a total set of all objects. This time, the first set contains all the things that are individuated as the agent. This set at least contains the agent’s body. The second set contains all the things that are individuated as the environment. Let this set contain only things that are not part of the agent’s body. Agency is extended if and only if these two sets are overlapping.[^12] Agent and environment are overlapping. Body and environment are overlapping The case of Jim may provide an example of Extended Agency. The machine is in the environment of Jim. Nevertheless Jim has in some sense agency over the movements of the machine. On the second escape route, it may be said that Jim’s agency extends because his agency and his environment are overlapping. This would be the result of denying E2, the condition that action requires a bodily movement. Hence, the second escape route, denying E2, can be interpreted as Extended Agency. The case of Jim may also provide an example of Extended Body. In this case the machine would be individuated as a part of Jim’s body. Jim’s body would be extended by the machine and therefore the machine’s movements would be his movements. This would be the result of denying E3, the claim that the body is individuated narrowly. Hence, the third escape route, denying E3, can be interpreted as Extended Body. # The trilemma revisited Before concluding, I will defend taking the second escape route out of the trilemma. This solution is not a general one. It is restricted to the case of Jim. For other cases, other escape routes might be taken. I first set aside the first escape route of Conceptual Rift. To decide between the remaining escape routes of Extended Agency and Extended Body, I put forward a three step heuristic. In the case of Jim this heuristic rules in favor of Extended Agency. The first escape route should not be taken in the case of Jim. The force of the argument so far does not seem strong enough to give up condition E1. Many consider this condition, that being the agent of $a$ requires that the agent performs $a$, a conceptual truth. At any rate, as I have argued in the interpretation of the trilemma, this condition (together with Surjectivity) secures conceptual unity in the sense that the two predicates ‘$i$ is the agent of $a$’ and ‘$a$ is an action of $i$’ are equivalent. In order to deny E1, more would need to be said about what it means for agency and action to come apart. Lacking such an analysis the other escape routes should be preferred.[^13] When there is a defensible analysis of how agency and action are two different concepts, then the escape route of the Conceptual Rift might be as preferable as the second one of Extended Agency, in the case of Jim. The choice is now between the second and the third escape route, between Extended Agency and Extended Body. The question is whether Jim’s agency extends or whether his body extends. I put forward the following heuristic to decide whether or not some $x$ extends. This heuristic can be applied to the case of Jim. The heuristic has three steps. First, the candidate and the object of the extension is specified.[^14] In this case, the candidate for extension is the movement of the machine. And the object of the extension is either Jim’s agency or Jim’s body. This candidate-object pair serves as an input into the second and the third step of the heuristic. The second step is a test on the object of the extension. Is the object such that it can be extended in principle? If this is not the case, then there heuristic stops here and the object does not extend to the candidate of the extension. Otherwise, the heuristic proceeds to the third step. What does it mean that something “can be extended in principle”? Consider the example of the body. If functions of the body can only be performed by the biological structure, and if they were performed external to this structure they would cease to be bodily functions, the the object “body” cannot be extended in principle. However, when the location of where a function is implemented, whether in the biological structure or external to it, does not make a difference to whether or not something is a function of the body, then the body can be extended.[^15] So in short, the second step of the heuristic tests whether or not the object of the extension can be extended in principle. Some object can be extended in principle just in case it is function is indifferent to the location of its implementation. If a body requires that its functions are implemented in a biological structure, then the body cannot be extended in principle. The third step is a test on the relation between the object and the candidate of the extension. This third step can be put as a question: Are the candidate and the object coupled in the right way? Consider again the example of the object “body”. The candidate that implements bodily functions needs to be coupled in the right way to the biological structure in order to extend it. It seems that prostheses, such as the hook of Captain Hook, are coupled to the biological structure in the right way because they are directly attached to it. An extension of the body seems to require fairly stringent criteria on this relation between the object and the candidate of the extension. For other objects of extension, looser requirements on the coupling between the objects and the candidate of extension might suffice.[^16] In short, the three steps of the heuristic are as follows. In the first step, a candidate-object pair of the extension is specified. The second step is a test whether the object can extend in principle. This test consists in whether or not the location where functions of the object are implemented makes a difference to whether or not they are a function of this object. The third step is a test whether the object extends in fact. This test consist in whether or not the candidate is related, or coupled, in the right way to the object. This three step heuristic can be applied to the case of Jim. I argue that Jim is a case of Extended Agency. I will first apply the heuristic to Extended Body. I will argue that while the body might extend in principle (second step), for the case of Jim and the machine, it does not extend in fact (third step). Second, I will apply the heuristic to Extended Agency. I will argue that Jim’s agency extends in principle and in fact. First, apply the heuristic to Extended Body. This means that the candidate-object pair is the movement of the machine (candidate for extension) and Jim’s body (object of extension). This specification is the first step of the heuristic. I take the second step for granted. Some may argue otherwise, but I assume that the body can extend in principle. The heuristic for Extended Body will fail in the third step. I argue that the machine does not extend Jim’s body because it is not coupled to the biological structure in the right way. Extended Body would be implausible for two reasons. First, it would violate a continuity condition for the body. It seems absurd that there are large gaps within one’s body such that a body consists of more than one continuously connected macroscopic object. If the machine and Jim’s biological structure are both his body, then his body would have such a macroscopic gap. This suggests that the machine is not coupled to Jim’s biological structure in the right way in order to extend his body. Second, Extended Body would lead to the situation where Jim would at least be partly present at different locations at the same time. Since the machine is a part of him, Jim would be where the machine is, and since his biological body is a part of him, he would be where his biological body is too. However, it seems absurd to say that a person can be partly present at two very different locations at the same time. This again suggests that the machine is not coupled to Jim’s biological structure in the right way. Hence, Extended Body seems false. The heuristic to test for whether Jim’s body extends to the machine fails in the third step. Now, apply the heuristic to Extended Agency. This means that the candidate-object pair is the movement of the machine (candidate for extension) and Jim’s agency, or the agential system “Jim” (object of extension). This specification is the first step of the heuristic. The second step is to test whether or not an agential system can extend in principle. Consider what the function of agency is. It seems, the function of agency is to intentionally interfere in the environment in the right way. The function takes a mental state or intentions as its input and results in movements or an interference with the environment as its output. Let this be the agential function. This function does not seem to be sensitive to where or how its output is implemented or realized. In other words, as long as the agential function takes Jim’s mental states and his intentions as its input, the interferences can be called Jim’s actions regardless of whether or not it is Jim’s bodily movements that implement them. In short, it seems that an agential system can extend in principle. The location of where the output of an agential function is realized can change as long as the input to the agential function remains unchanged. The third step of the heuristic is to test whether an agential system extends in practice. Is the machine coupled with Jim in the right way? Unlike the coupling conditions for the body, which seemed to require macroscopic continuity, the conditions to extend an agential system seem to be weaker. Assume that the machine is coupled to Jim’s body in the right way just in case it is reliable and easily available (cf. Clark and Chalmers 1998). The description of the case suggests that the machine is coupled to Jim in the right way. Hence, the agential system “Jim” extends from Jim’s body to the machine. The machine implements the output of the agential function of Jim and it is coupled to Jim in the right way because, by assumption of the case, the machine works reliably and is available to Jim as he needs it. # Conclusion I argued that agency does not need to be embodied. I showed that there are cases for which the embodiment view leads to a contradiction with a minimal definition of agency. Giving up the view that agency needs to be embodied solves the puzzle from the introduction. If agency does not need to be embodied then there can be an agent behind the action. Consider again the case of the soldier at the border who shoots a fugitive. The soldier may or may not be the agent of the killing of the fugitive. More importantly, the courts held his commander responsible for the killing. This is puzzling because, following the embodiment view, the commander cannot be the agent of the killing of the fugitive. When the embodiment view turns out to be false, and absent further blockers, then the commander might be the agent of the killing of the fugitive, despite not having performed a bodily movement in the killing. To discuss the case of the soldier at the border I abstracted to an relevantly analogous case of Jim who has a machine, which he can control with his mind. A review of the literature yielded three conditions that characterize the embodiment view. I called these three conditions Embodied Agency. Using the case of Jim, and two minimal assumptions about agency, I showed that Embodied Agency leads to a contradiction. Since there are three conditions in Embodied Agency, and since giving up one condition is sufficient to avoid the contradiction, the argument poses a trilemma. I called this trilemma the Embodiment Trilemma. The Embodiment Trilemma offers the following choice. Either agency and action are two different concepts. On this route, Jim would be the agent of an action that is not his. This is the first escape route, which I called Conceptual Rift. Or, agential systems can extend. This is the second escape route, which I called Extended Agency. On this route, the movement of the machine would be an action of Jim, while Jim performs no bodily movement at all. Or, and this is the third escape route, bodies can extend. On this route, the machine would be a part of Jim’s body, its movements would be movements of Jim’s body. This trilemma gives three different accounts of how agency can distribute over several objects. Each way is weaker than common notions of joint action. For the case of Jim, I argued that the second escape route should be taken. The case illustrates a case of Extended Agency. Since Jim can control the machine with his mind, and since the machine is coupled to him in the right way, it is an extension of his agency. This trilemma can help to characterize other cases of how individual agency can extend. As soon as one progresses from simple cases, such as that of Jim, to cases with more than one intentional agent involved, such as the case of the border soldier and his commander, a further issue comes up. Even if the commander can be the agent of the killing, as I tried to show in this paper, how does the commander’s agency combine with the agency, which the soldier seems to have in the killing? This question has not been discussed in this paper. But, in the light of the argument here, which defends the possibility of agency behind the action, it receives renewed urgency. # References Blackburn, Simon. 1996. *The Oxford Dictionary of Philosophy*. Oxford University Press. Clark, Andy, and David Chalmers. 1998. “The Extended Mind.” *Analysis* 58 (1): 7–19. . Davidson, Donald. 1971. “Agency.” In *Agent, Action, and Reason*, edited by Robert Binkley, Richard Bronaugh, and Ausonio Marras. University of Toronto Press. Dretske, Fred. 1991. *Explaining Behavior: Reasons in a World of Causes*. MIT Press. Goldman, Alvin I. 1976. *A Theory of Human Action*. Princeton University Press. Hertle, Hans-Hermann, and Maria Nooke. 2009. *Die Todesopfer an Der Berliner Mauer 1961-1989: Ein Biographisches Handbuch*. Ch. Links Verlag. Hornsby, Jennifer. 1980. *Actions*. Routledge. Melden, A. I. 1956. “Action.” *The Philosophical Review* 65 (4): 523–41. . Smith, Michael. 2004. *Ethics and the a Priori: Selected Essays on Moral Psychology and Meta-Ethics*. Cambridge University Press. Tuomela, Raimo. 1989. “Actions by Collectives.” *Philosophical Perspectives* 3 (January): 471–96. . [^1]: For a more recent examples, see decisions of the International Criminal Court, in particular ICC-01/04-01/07-717. [^2]: While both are held responsible, legal systems still draw a distinction. German jurisprudence distinguishes between the “Vordermann”, the front-man, and the “Hintermann”, the man behind a delict. The latter is the so-called mediated perpetrator, the “mittelbarer Täter”. [^3]: What exactly counts as adequate should not distract from cases that are sufficiently clear. [^4]: Note that this is different from E1, to be introduced later, which is that each agent requires an action. The entailment is in the other direction. [^5]: There is an ambiguity between two senses of “to move” (Hornsby 1980, 1–5). Notice the syntactic difference between (a) “my body moves” (one-place) and (b) “I move my body” (two-place). In the former case (a), my body may move because I am sitting in the bus, or because somebody else moves it. This is the *intransitive sense*. In the latter case (b), I move my body because I *make* my body move. This is the *transitive sense*. This paper is only on intransitive movements. [^6]: I use “embodied”, “embodiment” and “Embodied Agency” interchangeably. Note that this sense of embodiment is different from that of cognitive science (cognitive functions are not restricted to the brain) and robotics (exploitation of body-world interaction loops by design). [^7]: See also Miller (2001) who assumes that “so-called basic individual action consists of a mental state, such as an intention, a bodily movement, and a causal nexus between the intention and the bodily movement.” [^8]: I have not given textual support for the first condition. However, consider Blackburn (1996) who defines action as “\[w\]hat an agent does”. [^9]: In a deductive form the argument is the following.\ 1. Intuition: $a$ is an action or result of an action.\ 2. Surjectivity: there must be an agent of $a$.\ 3. Hypothesis: Only Jim x-or the machine can be agents of $a$.\ 4. Intentional States: The machine cannot be an agent of $a$.\ Therefore, Jim is the agent of $a$.\ 5. Embodied Agency: Jim is not an agent.\ Therefore, E1 or E2 or E3 is false. [^10]: Consider the literature on the logical form of action sentences \[cite Davidson and Parsons\]. In formalising action sentences this literature usually uses one only one two-place predicate in the formal language to express “being an agent of” and “being an action of” at the same time. [^11]: By “overlapping” I mean that the intersection of the sets is not empty. [^12]: Note that the two sets “environment” are not identical. In Extended Agency the environment contains only things that are not (part of) the agent’s body. In Extended Body the environment contains only things that are not (part of) the biological structure. [^13]: The following is a rough sketch of such an analysis of a) ‘$i$ being the agent of $a$’ and b) ‘$a$ being an action of $i$’. The relation a) obtains between an individual and an event if and only if the individual’s intention that $a$ is necessary and sufficient for $a$ to obtain. The relation b) obtains between an event and an individual if and only if that individual’s behavior is sufficient to bring about the event. [^14]: Here “object” is meant in the syntactic sense as “referring to”, not in the ontological sense as other occurrences of “object” earlier in the paper. [^15]: In the discussion on the Extended Mind, what I call “extension in principle” is known as whether or not a *Parity Principle* is accepted. [^16]: In the discussion on the Extended Mind, these conditions are known as *Coupling Criteria*.