The claim that conscious intentions play a causal role in action production should be compatible with our best empirical knowledge on how action is produced. The main empirical worries this claim confronts come from neuroscientific findings that have been interpreted as showing that the time of onset of conscious intentions is not compatible with their being the initiators of actions.
The most famous of these experiments are Libet’s studies on “readiness potential” (Libet et al. 1983; Libet 1985). In these studies, subjects were asked to flex their wrist at will and to note when they felt the urge to move by observing the position of a dot on a special clock. While subjects were both acting and monitoring their urges (intentions, decisions) to act, Libet used an EEG to record the activity of prefrontal motor areas. On average, participants reported the conscious intention to act, which Libet called the W-judgement, about 200ms before the onset of muscle activity. By contrast, the EEG revealed that preparatory brain activity, termed by Libet type II readiness potential (RP), preceded action onset by about 550ms. In other words, their brains started preparing the action at least 350ms before the participants became aware of their intention to act. This led Libet to the conclusion that the wrist-flexing actions in his experiments were not initiated by conscious intentions but were initiated instead by the (unconscious) RPs.
These experiments and Libet’s interpretation of his findings have been widely discussed (see e.g., Banks & Pockett 2007; Bayne & Pacherie 2014; Mele 2009; Nahmias 2002; Pacherie 2014; Roskies 2011) and commentators have pointed out a number of methodological problems with Libet’s paradigm as well as conceptual problems with his interpretation of his results. Let me focus first on one methodological problem and one attempt to address it. I will then consider one conceptual problem
Libet argues that it is the RP rather than the conscious intention that initiate the agent’s action. If RPs are the initiators of the action, there should be a robust correlation between them and the actions they cause: we should expect RP events to be “immediately” followed by the appropriate action, or, to put it the other way round, we should expect that when there is no movement, there is also no RP event. As several commentators have observed (e.g., Mele 2009; Roskies 2011), the back-averaging techniques used in the experiment do not allow us to ascertain whether this is indeed the case. Because the RP on any one trial is obscured by neural noise, what is presented as “the RP data” is determined by averaging the data collected on a large number of trials. In order to compute this average, the EEG recordings on different trials need to be aligned, and this requires some fixed point that can be identified across trials. Since in Libet’s experiments action onset serves as the needed fixed point for the alignment of EEG recordings, any RPs that are not followed by an action simply won’t be measured, and so we don’t know how robust the correlation between the RP and Libet-actions is.
In a recent experiment, Schurger and colleagues (Schurger et al. 2012) used a modified Libet task to circumvent the limitations of back-averaging techniques. Their aim was to test the proposal that RPs correlate with predecision activity rather than, as Libet proposed, with activity that coincides with, or is subsequent to, the agent’s decision. Schurger and colleagues proceeded on the assumption that the decisions of the participants in Libet’s experiment can be modelled—as neural decision tasks typically are—in terms of an accumulator-plus-threshold mechanism: decisions are made when relevant evidence accumulated over time reaches a certain threshold. Given that in Libet’s task subjects are explicitly instructed not to base their decision on any specific evidence, Schurger and colleagues proposed in this instance that the decision process amounts to simply shifting premotor activation closer to the threshold for initiation of the movement and waiting for a random threshold-crossing fluctuation in RP. Thus, Schurger and colleagues predicted the same premotor activation build-up as Libet when a movement is produced. However, whereas on Libet’s post-decision interpretation of this build-up there should be no premotor activity (and hence no RPs) when no movement is produced, on Schurger and colleagues’ stochastic decision model there should be continuous random fluctuations in RPs even when no movement is produced. Schurger and colleagues reasoned that it should be possible to capture these fluctuations by interrupting subjects in a Libet task with a compulsory response cue and sorting trials by their reaction times. On the assumption that the interrupted responses arise from the same decision accumulator as the self-initiated ones, and on the assumption that close-to-threshold activity reflects spontaneous fluctuations of RPs rather than mounting preparation to move building over the course of the entire trial, slow and fast reaction times should be distributed equally within trials. In their Libetus Interruptus task, they found, as they had predicted, that slow and fast responses to interruptions were distributed equally throughout the time span of the trial.
These results cast serious doubt on Libet’s claim that the neural decision to move coincides with the onset of the RP, since spontaneous fluctuations of RPs happen all the time. Therefore, they also cast doubt on his further claim that since RP onset precedes the urge to move by 350ms or more, conscious intentions can play no role in the initiation of the movement. If instead the neural decision to move coincides with a much later threshold-crossing event, it remains at least an open possibility that this event coincides with and constitutes the neural basis of a conscious urge to move. Schurger and colleagues take no stand on the exact relation between the conscious urge to move and their threshold-crossing event. They insist, however, that this threshold-crossing event should not be interpreted as the cause of the movement but rather as just one of the many factors involved in the causation of self-initiated movements. This leads me to my final point.
One conceptual problem with Libet’s interpretation of his findings and also, as Dreßing points out, with most interpretations of neuroscientific experiments and a large part of the philosophical debates on mental causation and causal exclusion lies in the conception of causality that is assumed, “namely a temporal, linear, one-way causality” (Dreßing this collection, p. 10). I agree with Dreßing’s suggestion that a different concept of causation should be considered, one that allows for multiple causal processes to operate in parallel and to exert influence on one another. This is indeed the spirit of the dynamical model of intentions I have proposed elsewhere (Pacherie 2008). In particular, I insisted that a distal intention does not cease to exist and play a role once a corresponding proximal intention has been formed (and the same goes for proximal and motor intentions). What I suggested is that all three levels of intentions operate simultaneously, each exerting its own form of control, as well as operating together with unconscious processes. Following Dretske’s lead, we can think of intentions as structuring rather than as triggering causes of action. On the dynamic hierarchical model of intentions I have proposed, we can further think of the structures set up by intentions as nested. This means that we don’t need intentions to initiate actions for them to play a causal role in the production of action. This also means that the intentional online control that I argued was an important pragmatic function of intention may be best conceived as a form of re-structuring, necessary only when the initial structuring is inadequate.