From the subjective dimension of agency, let us now turn to how philosophical action theory deals with BCI-mediated events. To appreciate the peculiarities of BCI-mediated events, let us introduce two common distinctions of the philosophical literature:
the distinction between basic and non-basic acts and
the distinction between mental action and bodily actions.
Mental Actions, Basic Actions and Non-basic Actions
First, we shall elaborate on the distinction between mental action and bodily action. Thinking about actions usually conjures up images of people intentionally moving their bodies (or parts of them) in space. However, many of the things that humans do intentionally do not involve peripheral bodily movements. For example, we intentionally deliberate about something, try to remember something, or calculate “in our head.” Hence, it seems sensible to distinguish between overt bodily actions and mental actions. Mental actions are actions that do not involve any external movement of the body or limbs (Mele 1997). In other words, in mental actions “[…] there is no motor output to be controlled and no sensory input vector that could be manipulated by bodily movement” (Metzinger 2017, 1) and mental actions bring about a mental event, a mental disposition, or a mental state (Proust 2001).Footnote 6 Despite the differences, there seem to be many similarities between mental and bodily actions. For example, both types of action can be purposeful, goal-directed, inhibited intentionally, or controlled by the agent to some degree. Further, both can include a sense of agency, and both can require motivation, learning and prior experience (Metzinger 2017; Wu 2013; Mele 1997). However, some authors draw a distinction of different forms of mental agency and argue that not all forms of mental actions are similar to bodily actions. For Tillmann Vierkant (2018) and Pamela Hieronymi (2009), the so-called managerial control is the only kind of mental action that is similar to bodily actions. Managerial control is an “intentional mental action with the purpose of creating an environment that will facilitate cognitive activity and the management of attitudes” (Vierkant 2018, 6). Another form of mental agency, cognitive agency, differs from bodily agency because cognitive acts, like judgments, are not intentional.
To shed some light on what mental agency means in the BCI-context, consider the distinction between two kinds of mental agency provided by Thomas Metzinger. There is attentional agency, which is the “ability to control one’s focus of attention”, and there is cognitive agency, which is “the ability to control goal/task-related deliberate thought” (Metzinger 2013, 2). These two kinds of mental agency mirror nicely what is going on in most active and reactive BCIs (see Section 2 above). Remember, the subject has to either direct her attention (reactive BCI) or she has to perform some motor imagery (active BCI), both of which are mental actions in the sense just outlined. Contrast this with passive BCIs where brain activity is monitored but no mental action is required. It is worth pointing out here that the two kinds of mental agency that Metzinger talks about are both intentional and therefore similar to bodily agency. But as we have just seen with reference to Vierkant and Hieronymi, one form of mental agency, namely cognitive agency that includes cognitive acts like judgments, is not intentional. Now, provided that a future BCI would be able to respond to non-intentional mental acts, we would be confronted with a case where a non-intentional mental act directly causes some movement, which has implications for issues of moral agency.Footnote 7 So far as we are aware of, contemporary BCIs do not yet involve this non-intentional mental agency.
Let us now turn to the second distinction: What does it mean to say that an action is basic, while another action is non-basic? Sometimes we do things by means of doing other things. For example, we drive a car by shifting gears and controlling the steering wheel. Based on observations like these, some philosophers draw a distinction between basic and non-basic acts (e.g., Goldman 1970, ch.3.4; Danto 1973). A basic act is an act that you do not do by doing something else.Footnote 8 The paradigmatic example is raising an arm: We simply raise it, without any further steps in between. Consequently, provided the distinction between basic and non-basic acts is sensible, if an event is an action, then it is either a basic action or it is brought about by the performance of a basic action.
Combining this distinction with the one between mental and overt actions yields a particularity of BCI actions. The BCI-mediated event in active and reactive BCIs is a non-basic action (assuming that it is an action at all) because the subject always performs it by doing something else. That is, in order to bring about an external event or movement, the subject has to perform a mental action first (which is regularly a basic act in the previously mentioned sense), which in turn means that the BCI-mediated event in question is a non-basic action. Thus, an action with identical effects—say, an arm reaching for a glass—is a basic action under natural circumstances (that is in non-BCI contexts), but a non-basic action when it is facilitated with a BCI-prosthesis. This is the third peculiarity of BCI actions we wish to highlight.
Standard Theory of Action
But are such BCI-mediated events really actions proper? Many authors simply assume without further questioning that they are actions (e.g., Requarth 2015; Kotchetkov et al. 2010). But this is not as evident as it may appear and leads into intricate debates in action theory. In this section, we will show that according to the popular causal theory of action (e.g., Davidson 1963), some BCI-mediated events qualify as actions,Footnote 9 but not all of them.
What does the so-called standard theory of action say? The standard theory of action is a causal theory,Footnote 10 according to which an event is an action if it is caused, in the right way, by certain mental states and events.Footnote 11 Further, the mental states that cause the action must have a content in virtue of which they rationalize the action.Footnote 12 It is commonly assumed that intentions, beliefs, and desires can play such a role. Further, most authors hold that at least intentions play a crucial causal role in action and that intentions are not reducible to beliefs and desires or a combination thereof (e.g., Bratman 1987).
What are intentions? As one would expect, there is some disagreement on what intentions are exactly. Whatever else is true about intentions, it is plausible to assume that intentions are commitments to action plans (Clarke 2010; Mele 1992). The content of an intention, i.e., the action plan, may be simple or detailed. That is, a person may intend to simply buy milk in the store, or she may intend buying a particular brand by taking a particular route to a specific store.Footnote 13 Further, we can be aware of our intentions when we act but many actions lack conscious intentions (see Lumer 2017). It is not a requirement of the standard theory of action that we are constantly conscious of our intentions.
Further distinctions regarding intentions have been proposed. Pacherie (2007), in developing John Searle’s ideas on intentions, differentiates between present-directed intentions (P-Intentions) and future directed intentions (F-Intentions). F-Intentions are formed before the onset of the action and are closely linked to deliberation. In contrast to F-Intentions, a P-Intention implements the action plan (inherited from the F-Intention) in a specific situation and entails a more precise representation of the goal. In addition, Pacherie introduces another form of intention that she calls motor-intention (M-Intention). At the level of M-Intentions, perceptual-action contents are transformed into sensory-motor representations. The idea is that M-intentions monitor the bodily movement in a fine-grained way.
Now, consider BCI use in the case of the BCI application Brain Painting that we have described in the introduction. It is sensible to assume here that F-Intention and P-Intention play the same role as in paradigmatic actions. The subject may deliberate to produce a painting and she knows how to achieve this goal. That is, she has an action plan that is then implemented via P-Intentions that involve the mental action of directing attention in order to pick the specific painting tools. The M-Intention on the other hand, is missing in this case. A peculiarity of BCI technology is that motor-intentions do not feature prominently, because in most cases, there simply is no motor-intention because there is no motor representation that guides an intentional bodily movement.
Let us now come back to BCI-mediated events and how the standard theory of action handles them. Consider a scenario of an active or reactive BCI. The subject has to perform a mental action (e.g., motor imagery or selective attention), which is a basic action (as established above), in order to perform the non-basic action of selecting an option from a menu. The causal theory of action holds that this is an action if it is caused in the right way by the right sort of mental states. It is uncontroversial to assume that the subject in our scenario has an intention, in that she is committed to an action plan that includes performing a mental action to achieve a certain goal. The subject also knows that performing the mental action will lead to the desired goal, i.e., selection of a certain option (if the BCI works properly, that is). Further, applying Pacherie’s types of intentions here, it is sensible to assume that F-Intention and P-Intention play the same role in BCI-mediated events as in paradigmatic bodily actions. The F-Intention may be described as something like “Selecting a desired option with the help of the BCI”, while the P-Intention might be described as something like “I want to select the red colored circle so I have to focus my attention on it” (in reactive BCIs) or “I want to select the red colored circle so I imagine moving my thumb” (in active BCIs). So, the occurrence in question (e.g., selection of an option), is caused and rationalized by the right kind of mental state, i.e., an intention, and is therefore an action proper. In these cases, BCI-mediated events are similar to paradigmatic actions in their causal ancestry, and therefore qualify as actions.
By contrast, according to the standard theory, events mediated by passive BCIs do not constitute actions, because the events in question are not caused by the right mental states. This is the fourth peculiarity we wish to highlight: events mediated by passive BCIs are no actions.
Action and Control
However, some authors consider causal accounts of agency of the kind just presented as too coarse-grained. A key objection to the standard view is that it places too much emphasis on the causal history of the event in question and overlooks the capacities of the person in the moment of executing a movement (e.g., Frankfurt 1978). A long debate about deviant or wayward causal chains revolves around this issue. To cut a controversial story short, it seems that sometimes, persons do not act—even though their movements are caused by the right antecedent states such as beliefs, desires, and intentions—because they lack control over the movement. This is relevant for the present purposes because BCI-mediated events seem to differ from ordinary actions with respect to control over the execution of the movement. Control is surely a multifaceted concept, especially when applied to the mental realm, and too broad to fully unpack here. However, Pacherie (2007) provides a useful distinction between guidance and executive control and Shepherd (2015) also distinguishes between the executive aspects and the implementational aspects of an action. The first concerns such things as planning and plan revision, whereas the second concerns the establishment of a state of affairs that satisfies a goal state. In line with these distinctions, we suggest distinguishing executory, guidance, and veto control and shall apply them to BCI use.
Executory Control and Libet Experiments
People have many desires, beliefs, and intentions on which they do not act. Something additional has to come in to realize such intentions: an executory command. Often, it is called a volition. But because the term is controversial and ambiguous, we rather speak of an executory command.Footnote 14 From a commonsensical perspective, the idea seems plausible: Sometimes, it occurs to us as if we give a conscious command, a go-signal, to initiate an action, and that it causally executes the action. Think about a camera-woman who carefully observes a landscape and adjusts her equipment until she thinks “This is the perfect moment” and starts the recording. There seems to be a clear executory command that follows a conscious appreciation of the moment (even though the decision was influenced by external stimuli). However, we often move our bodies without consciously giving any such command. Most of the time, we just act. In fact, there seems to be an entire subset of actions—automatisms—that we perform in the absence of such executory commands (we will turn to automatism below). A surprising peculiarity of reactive and active BCIs is that users have to give a go-command via the respective mental action. The device picks up the underlying signal and translates it into movements. In such applications, users give a go-command in a much more pronounced manner—namely via a conscious and effortful mental action—than we usually give go-commands in many ordinary actions. This is the fifth peculiarity of BCI actions worth highlighting.
Interestingly, this difference might bear relevance for the discussion about the implications of experiments by Libet (Libet 1985, 1999) or Wegner (2002). Without rehearsing them here, one of the worries they stirred is that the subjective experience of movement initiation, the executory command, might not “cause the action,” as it is neither causally efficacious, nor necessary for movement execution. Instead of the experience of “consciously willing to initiate a movement,” antecedent unconscious processes detectable in the motor cortex (“readiness potential”) have initiated the movement. The experience of initiating the movement then turns out to be misleading (The role of the conscious experience is controversial. Perhaps it is epiphenomenal; perhaps it comes with a veto-power). Interestingly, and irrespective of one’s position on Libet’s experiments, the initiation of movements via BCIs would not fall prey to such worries, because respective unconscious preparatory processes in the motor cortex (and related brain areas) cannot initiate the movement and remain causally irrelevant, because the output via the neuromuscular system is not used. The BCI commands applications only if it detects the signals correlative of the respective mental act. This requires a conscious and effortful performance of a mental act. BCI-movements thus preserve the causal efficacy of a conscious command (such as mental imagery). In this regard, BCI actions may come closer than ordinary actions to the ideal of consciously initiated action.Footnote 15
Executory control over movement initiation is just one form of control. A famous case example in action theory can be explained by its absence:
Bob desires and intends to shoot the sheriff, but this makes him nervous and causes his finger to cramp, which in turn causes the trigger to be pulled, resulting in the gun being fired and the sheriff being shot.
Did Bob shoot the sheriff? Although the shooting was caused by the correct mental states, it seems that it was not caused in the right way. Instead of the executory command, there was an unintentional control-undermining element at play here—nervousness—that initiated the movement.Footnote 16 Therefore, the shooting does not qualify as an action according to the standard theory of action. Such cases illustrate that executory control is an important element of actions (whether it is necessary is a question we will address in a moment).
Whereas agents with active BCIs have executory control in the way just mentioned, other BCI designs can initiate events without executory commands. In the standard protocols, one option is that the BCI initiates events in an application without a command by the user (Wolpaw and Wolpaw 2012). One line of current research focuses on BCIs that read out brain states to predict users’ movement intentions (Wang and Jung 2011). Through machine learning, BCIs may adapt to plans or intentions of the users, bypassing the cumbersome process of them giving commands via mental imagery for example. Engineers hope that this will lead to a much smoother user experience. Actually, it might be a requirement for the practical success of restorative BCIs. However, if BCI systems initiate movements without any user command, or by reading out users’ preferences or more distal intentions, users do not have executory control. This leads to the sixth peculiarity of BCIs, that it may initiate movements by predictive interpretation of brain signals. This has the somehow paradoxical effect that engineering BCI systems that work best for users, especially patients, may work well precisely by undermining features such as initiation that are central to concepts of action.
There are further forms of control. Sometimes, after initiation, people have control over the ensuing execution of movements, which is the ability to alter and influence the execution of movements, such as the trajectory of a limb movement. Let us call this kind of control guidance control. Guidance control is often limited. For instance, we are not aware of the many muscles and their contractions that are necessary to raise an arm, let alone to perform more sophisticated actions such as skiing. Overall, humans lack fine-grained muscular control. Nonetheless, through various feedback channels, we monitor the progression of our movements and are able to adjust them, i.e., we could get them under conscious control, although only on general levels.
In most BCI applications, however, users do not possess guidance control to a relevantly similar degree. The signals detected by the BCI are processed and transformed by the machine in operations that are inaccessible to the user. Because the BCI is like a black-box, users do not have guidance control over those parts of the causal chain that are processed and executed by the machine. Users may only influence the unfolding of events controlled by the machine by sending another signal, e.g., redirecting the process, if the technical design affords so. Much depends on the precise properties of the technical design. Here, the distinction between goal-selection and process-control becomes relevant. In goal-selection (“autopilot”), users lack guidance control. In process-control, they possess guidance control insofar as they can direct the events controlled by the application. This might be more or less fine-grained, depending on the technical setup. In the future, more sophisticated BCIs might be able to provide fine-grained process-control along with multisensory feedback, so that they are more integrated with the minds of users. The amount of guidance control might then resemble ordinary guidance control of healthy adults. So, we wish to note the seventh peculiarity: in BCI designs with goal selection, users lack any guidance control. In process-control, users may possess some.
The last form of control we wish to distinguish is veto control, the power to stop the further execution of a movement. Evidently an important form of control, users of reactive BCIs possess it insofar as they can simply discontinue sustaining attention. This will stop the movement. The same might be possible through mental actions in active BCIs. In addition, further technical options to stop the operation of the BCI should be implemented. In the best case, users have the power to halt BCI-mediated operations at any time through some other output channel. Thus, we call upon BCI engineers to design sufficient ways and options to provide stop-signals.Footnote 17
We have introduced three forms of control: executory, guidance, and veto control. The question is which of them is necessary or sufficient for a movement to qualify as an action. We presuppose, in accordance with arguments in the literature outlined above, that some form of control is required. But which? We suggest either executory or guidance control. If a person exerts at least one of them, the movement may constitute an action (other conditions notwithstanding). What about veto control? This is important because that is the only form of control many BCI designs will leave users. Future BCI designs will operate by reading out P-intentions or other mental states to predict what users may want to do and initiate the action. Users then have neither executory nor guidance control. Are such BCI-mediated events actions? Whether it suffices in the absence of executory or guidance control is an interesting question that brings us to the gray areas of the concept of action. Is a movement that a person neither initiates, nor controls during execution, an action, if she could stop it?
This question has parallels to another gray area in the investigation of agency—automatism. Automatisms are movements that are not initiated by a conscious executory command, nor do persons exert conscious guidance over them. Classic examples include movements in sports or driving a car. Driving a car requires constant movements such as shifting gears, breaking, accelerating. Most of these movements are executed without any conscious control; they are often even outside of awareness of the driver. They are more or less initiated and guided by unconscious mechanisms. Given the prevalence of automaticity in daily life (Bargh and Chartrand 1999) the question arises whether such automatic behavior are actions or not.Footnote 18 For causal theorists, the answer depends on whether the behavior is non-deviantly caused by a corresponding intention (see Lumer 2017). The behavior does not need the formation of P-intentions according to Pacherie’s taxonomy mentioned above. Suppose a person wishes to drive his car from Paris to Rome, gets in the car, and indeed drives there performing a myriad of automatized movements. It would be strange to deny that the drive is an action. Similarly, many of our daily automatic movements may not be traceable to intentions. Think about the many things we do with our hands all day, e.g., during a conversation. These movements are neither consciously initiated, nor executed. But they could, at any time, be brought under conscious control by attending to it. They are potentially controllable. Is this sufficient? Are such subconsciously initiated and guided movements actions, in virtue of them being potentially consciously control- and stoppable? We wish to suggest that potential control is not enough. These events are thus no actions (they might be omissions). In any case, we can note that such movements are located at the fuzzy borders of the concept of agency. BCI-mediated events will often be set in this gray area between actions, omissions and mere movements. In most current BCI applications, there is no or very little guidance and veto control. The question whether veto control is sufficient to qualify the event as an action might become more relevant when passive BCIs become more common and are equipped with an active veto control option.