Intending to perform an action and then immediately executing it is a mundane process. The cognitive and neural mechanisms involved in this process of “proximal” intention formation and execution, in the face of multiple options to choose from, are not clear, however. Especially, it is not clear how intentions are formed when the choice makes no difference. Here we used behavioral and electrophysiological measures to investigate the temporal dynamics of proximal intention formation and “change of intention” in a free picking scenario, in which the alternatives are on a par for the participant. Participants pressed a right or left button following either an instructive visible arrow cue or a visible neutral “free-choice” cue, both preceded by a masked arrow prime. The goal of the prime was to induce a bias toward pressing the left or right button. Presumably, when the choice is arbitrary, such bias should determine the decision. EEG lateralized readiness potentials and EMG measurements revealed that the prime indeed induced an intention to move in one direction. However, we discovered a signature of “change of intention” in both the Instructed and Free-choice decisions. These results suggest that, even in arbitrary choices, biases present in the neural system for choosing one or another option may be overruled and point to a curious “picking deliberation” phenomenon. We discuss a possible neural scenario that could explain this phenomenon.
In many daily life situations, we intend to perform an action selected from several alternatives and immediately execute it. The cognitive and neural mechanisms involved in this process of intention formation and execution are not clear, however. For example, intentions can be directed exogenously by an instructive cue (e.g., the green traffic light signal) or formed endogenously according to “free” will (e.g., “I am going to wear my blue jacket today”; for a recent overview, see Janczyk, Nolden, & Jolicoeur, 2014). Experimentally, the latter is typically investigated in Libet-style experiments in which an endogenously driven decision includes selecting between options that are on a par for the participant (Libet, 1985), for instance, pressing a right or left button according to what the participant “freely wants” when a cue appears. This type of selection between options that make no difference to the agent is termed “picking” (following Ullmann-Margalit & Morgenbesser, 1977) and is distinguished from “choosing,” in which the selection is based on reasoning. We focus here on immediate (“proximal”) picking selections (i.e., the intention to do something now, in contrast to long-term commitments and intentions; Furstenberg, 2014; Mele, 2009) and investigate the dynamics that underlies the selection process in these picking scenarios.
The most basic question about true picking situations is why picking situations do not result in paralysis or how, lacking any difference between the choices or any other reason, a decision is made at all. This was nicely described by Spinoza: “if a man were placed in such a state of equilibrium he would perish of hunger and thirst, supposing he perceived nothing but hunger and thirst, and the food and drink were equidistant from him” (Spinoza, 1677/1930).1 A natural solution is that although on one, explicit or reason-based level, there is symmetry between the alternatives (i.e., there is no reason to prefer one option over the other), on a lower, implicit level the symmetry does not maintain. The underlying asymmetry between alternatives is what makes possible the picking of one alternative rather than the other; if the symmetry were maintained throughout all levels, picking would be impossible. In the words of Leibniz: “insensible impressions…can suffice to tilt the balance…we are never indifferent, even when we appear to be most so, as for instance over whether to turn left or right at the end of a lane. For the choice that we make arises from these insensible stimuli.” (Leibniz, 1765/1981; see Furstenberg, Deouell, & Sompolinsky, 2015, for further discussion of this philosophical question). Here, we studied the notion of this implicit asymmetry and its influence on the final decision by measuring brain signals related to motor preparation. These signals, on the verge of overt action, provide evidence that an intention has been formed (Furstenberg, 2014).
For this purpose, we used a simple picking scenario, in which an agent freely selects whether to press a button using the left or right hand with no specific advantage to either decision over the other. We presented an external subliminal prime in attempt to create an implicit asymmetry between the two alternatives. Studies show that, in cases where a subliminal prime points to one alternative, there is a selection bias toward the direction of the prime (Positive Compatibility Effect; Kiesel et al., 2006; Schlaghecken & Eimer, 2004). Moreover, when the agent's immediate executed response is in the direction opposite the prime direction (incongruent response), RT is delayed relative to when the prime and the response match (congruent case; Kiesel et al., 2006; Schlaghecken & Eimer, 2004). There are at least two explanations for delays in the incongruent response: “hesitation” (i.e., increased time to create any directional intention) or “change of intention” (ChoI; i.e., intending first in one direction and then changing the intention).
One potential strategy to address these questions is by using EEG and specifically the lateralized readiness potential (LRP). The LRP reflects the difference between the scalp potentials over the right and left motor cortices. It can be used as a measure of motor preparation to move the left or right arm, as its polarity indicates the hand which is prepared to move (Smulders & Miller, 2011; Eimer & Coles, 2003). Several studies have measured the LRP in paradigms in which a subliminal prime is followed by a supraliminal cue, instructing the participant to use the left or right hand (e.g., Eimer & Schlaghecken, 1998, 2003; Jaskowski, Skalska, & Verleger, 2003; Dehaene et al., 1998). The usual finding is that responses in the incongruent condition are delayed, and in addition, the LRP first reflects preparation to move the hand indicated by the prime and then reverses to reflect preparation of the hand that finally moves based on the instructive supraliminal cue, suggesting a ChoI. What will be the dynamics of intention formation in a free-choice picking situation, that is, with no instructing cue? Under the premise that the decision in this picking situation is based on incidental neuronal asymmetry at the time of the decision, inducing an asymmetry using a subliminal prime should determine the final decision, with no indication of ChoI.2 In contrast, because the LRP reflects a late, output stage of the decision process, observing a pattern of ChoI at this final stage would suggest that the prime induced an actual intention (plan) to move, and nevertheless, the final act was different. If ChoI is found in free-choice picking situations, then the occurrence of neuronal asymmetry is not sufficient to fully determine the outcome, again begging the question of how free-choice selections are made. Note that in this study we use the concept of intention to include immediate nonconscious and nonexecuted neural preparations to perform a motor plan and not necessarily an explicit control mechanism (Furstenberg, 2014; Mele, 2009).
Only one study measured so far the LRP during a Free-choice picking paradigm with subliminal directional priming cues (Kiesel et al., 2006). In that study, the LRP in the Free-choice incongruent condition reflected only the direction of the final choice and not that of the prime, yet it was delayed. This pattern could imply that, unlike the Instructed situations, conflicting intentions in Free-choice situations result in hesitation, rather than in ChoI. However, the interpretation of this study is complicated by the fact that, in contrast to the majority of studies, Kiesel et al. (2006) did not find indication of ChoI even in the Instructed condition. Therefore, the lack of the ChoI pattern could reflect the specific choice of experimental parameters and not a principle difference between Instructed and Free-choice conditions. Here we examined again whether ChoI could be part of a free-choice picking selection process. We induced neural asymmetry and used EEG and EMG as indices by which to follow intentions and changes thereof.
Twenty students aged 19–28 years participated in the experiment (12 women, mean age = 23.7 years, SD = 2.7). All reported normal or corrected-to-normal vision and no history of neurological disorders. Participants received monetary compensation or course credits for their participation in the study. Written consent was obtained after the general experimental procedures were explained. The study was approved by the ethics committee of The Hebrew University of Jerusalem.
Stimuli and Design
Trials consisted of the presentation of a prime for 20 msec, followed by a masking screen for 40 msec, and finally the cue for 70 msec. Prime stimuli included the symbols “>>” (right directional prime, visual angle: 2.8° × 1.24°), “<<” (left directional prime, 2.8° × 1.24°) or “<< >>” (neutral prime, visual angle: 5.68° × 1.24°). The mask was constructed of lines of various lengths and orientation scattered within a virtual rectangle (visual angle: 13.4° × 7.6°). There were eight such different masks that were randomly applied. The cue was a right or left arrow identical to the directional prime arrows or a “+” sign (visual angle: 2.23° × 2.23°), which indicated a free-choice trial (Figure 1). The task was to press a button as fast and accurately as possible upon seeing the cue with the left or right index finger (on the “z” and “/” keyboard buttons, respectively), according to the information given by the cue. In the “Instructed” condition, participants had to press the finger that corresponds to the direction of the cue (i.e., left finger for left pointing arrow and vice versa). In the “Free-choice” condition, participants had to press with either the left or right finger as they choose. Note that until the onset of the cue, participants did not know whether the trial was Instructed or Free-choice.
The experiment took place in a dimly lit, sound-attenuated room (Eckel C-26, Cambridge, MA). Participants were comfortably seated in front of a CRT monitor (ViewSonic G75f, Walnut CA; 100 Hz refresh rate), with a viewing distance of 90 cm. All stimuli were presented at the center of the screen in black color on a gray background. Responses were provided through a standard computer keyboard placed on the participants' lap. The experimental procedure was controlled by a PC running E-Prime software (Psychology Software Tools, Sharpsburg, PA; version 1.2). The intertrial interval was randomly jittered between 1500 and 2000 msec (uniform distribution with 100 msec steps), while presenting a fixation point in the center of the screen (filled black circle, visual angle = 0.38°). RTs are reported relative to prime onset to match the segmentation of the ERPs. Participants were instructed to respond as fast and as accurately as possible to the cue. To motivate fast responses, the participants had a response window of 570 msec from cue onset (630 msec from prime onset) in Instructed trials and 870 msec from cue onset (930 from prime onset) in Free-choice trials. In trials in which responses exceeded these limits, a “too slow” message was presented, and the RT of these trials was discarded from later analyses. An incorrect response to an Instructed arrow cue resulted in an error message “Wrong” and a somewhat unpleasant tone.
The first testing block consisted of 60 Instructed trials, intended to create the association between the cue and the response. This was followed by four mixed blocks, each consisting of 65 Instructed and 65 Free-choice trials. In Instructed trials of these blocks, the cue had equal probability to be left or right, and orthogonally, the prime had equal probability to be left or right. Thus, there were equal chances for the prime to be congruent or incongruent with the cue. In Free-choice trials of these blocks, the prime had also equal probability to point left or right. These blocks were followed by three blocks consisting each of 60 Instructed and 60 Free-choice trials, all with neutral primes. Two practice blocks preceded the experiment, one with 16 Instructed trials and the second with a mixture of 16 Instructed and 16 Free-choice trials. The experimental blocks started after the participants became confident with the task and stimuli. The participants were encouraged to rest their eyes for a few minutes between blocks.
Objective Evaluation of Awareness
Following the main experimental blocks, awareness of the primes was evaluated in an objective two-interval two-alternative forced-choice (2AFC) testing block. Each trial consisted of two different masking screens (each randomly chosen as above) appearing one after the other (each for 40 msec) with an ISI of 1.5 sec. One of these masking screens (randomly chosen in each trial) was preceded by a 20-msec arrow prime (identical to the one used in the main experiment), whereas the other was preceded by a 20-msec neutral prime (“//”; visual angle: 2.76° × 0.6°). Participants were asked successively (1) whether a rapid arrow prime appeared before the first or second mask (detection) and (2) whether the arrow prime pointed left or right (discrimination). The first question (detection) was replied by pressing with the left hand on the “1” or “2” top keyboard keys, and the second question (discrimination) was replied by pressing with the right hand on the keyboard left or right arrow keys. Emphasis was given to accuracy and not to speed. The block consisted of 50 trials. Each trial began after the participant replied to the two successive questions of the previous trial. Twice during the block (after 20 and after 40 trials), participants were presented with a feedback screen with the number of successful detection and discrimination hits. The purpose of this feedback was to keep the participants motivated.
In the Instructed trials, the percentage of errors and RT were calculated separately for congruent, incongruent, and neutral conditions, where congruent trials were defined as those with similar prime and cue, incongruent trials as those with opposing prime, and cue and neutral trials as those with a neutral prime. In the Free-choice trials, no explicit directional cue was given; thus, we define congruency according to the match between the direction of the prime and the chosen response. By definition, there were no errors in this condition. Bias of choices toward the prime direction was measured by the percentage of congruent choices versus incongruent choices. Priming was expected to induce a congruency effect, that is, a higher percentage of congruent versus incongruent responses. As prime direction was equally distributed between left and right, a preexisting response bias for one direction would not be reflected in a congruency effect (e.g., if in the extreme case the participant pressed only with one hand, there would be no congruency bias). RT was also calculated in the Free-choice trials separately for the congruent, incongruent, and neutral trials.
EEG was recorded continuously with preamplified sintered Ag/AgCl electrodes from 64 scalp locations according to the extended 10–20 system (www.biosemi.com/pics/cap_64_layout_medium.jpg), as well as from the tip of the nose, using a BioSemi Active 2 system (BioSemi, Amsterdam, the Netherlands). Blinks and eye movements were monitored using two EOG electrodes located at the outer canthi of the right and left eyes for horizontal movements and an electrode below the center of the right eye together with the FP2 electrode for vertical movements. The EEG was sampled at 1024 Hz with an online low-pass anti-aliasing filter (cutoff: 205 Hz) and stored for offline analysis. Analysis was conducted using Brain Vision Analyzer 2 (Brain Products, Gilching, Germany) and Matlab R2010b (The MathWorks, Inc., Natick, MA). The EEG data were digitally referenced to the nose and filtered with a bandpass of 0.5–15 Hz (zero-phase shift, 24 dB/octave Butterworth filter) for the ERP analysis. Blink artifacts were removed using the independent component analysis method (Jung et al., 2000; as implemented in Analyzer 2). Segments contaminated by other artifacts were discarded (rejection criteria: >100 mV absolute difference between samples within intervals of 100 msec; absolute amplitude beyond the ±100 mV range). Participants were asked to minimize movements and eye blinks during the task.
The continuous EEG signal was segmented into single trial epochs, separately for the experimental conditions of Prime direction (Right/Left/Neutral) × Cue (Right arrow/Left arrow/Free-choice) × Response direction (Right/Left). As a convention, we refer to specific conditions using the abbreviation of the first letters of the prime, cue and response identities, with + as the identifier of the free-choice condition (e.g., RRR for right prime, right cue, right response and L + R for left prime, free-choice cue, right response). We used the LRP ERP to identify preparation of left- or right-hand movement. The LRP was derived by subtracting the preresponse potential recorded from a central, parasagittal scalp electrode (C4), roughly located over the right motor cortex from the potential recorded at a homologous site over the contralateral scalp (C3). Activation of the left motor cortex in preparation of right-hand movement results in a negative shift of the LRP whereas activation of the right motor cortex in preparation to move the left hand results in a positive shift of the LRP. For response-locked analysis, where the LRP is seen most clearly, the EEG was segmented to epochs extending from 1000 msec before until 200 msec after the response. Potentials were calculated relative to a (−1000) to (−800) msec preresponse baseline period. These segments were averaged for each participant per condition. The average signed LRP amplitude in the −200:0 msec preceding the response was used to test the effects of the prime and response direction on the LRP. Only correct Instructed responses were analyzed; thus, response direction was identical to cue direction.
Our main interest was to examine the effect of the prime and the cue on the formation of intentions in the Instructed and in the Free-choice conditions. Because RTs vary from trial to trial, prime and cue-dependent information might not be apparent in response-locked averages because of temporal smearing. To examine the effect of prime–cue relations on the preparation to move, we thus segmented the EEG time-locked to the prime–mask–cue complex. For this analysis, segments extended from 200 msec before until 1000 msec after prime onset. Potentials were calculated relative to a (−200) to (−50) msec prestimulus baseline period. We used a t cluster-based permutation test (following Maris & Oostenveld, 2007) to test for significant differences between conditions without prior assumptions about the timing of the effect. Specifically, we first performed point-by-point t tests comparing two conditions. Then, for each cluster of consecutive time point in which the t test yielded a significant difference (p < .05, uncorrected), we calculated a t sum statistic, defined as the sum of all t values of the cluster. This statistic is affected by both the strength and temporal extent of a cluster. To estimate the probability of finding a cluster with a given t sum value in case that the null hypothesis of no difference between conditions is true, we performed a permutation analysis. This analysis included 10,000 iterations of the t cluster analysis. For each iteration, we randomly permuted the condition labels within participant and obtained the maximal t cluster statistic. These 10,000 values establish the null distribution, which is an estimate of the distribution of the largest t cluster statistics that could result by chance in our data. Finally, a cluster of consecutive significant t tests in the real data was considered to be significant only if its t sum statistic exceeded the 95% cutoff of this null distribution of max t sum values. We used this approach to compare the incongruent condition to the congruent condition, and also the incongruent condition to 0, to verify that the incongruent condition not only affected the degree of left–right asymmetry reflected by the LRP but actually inverted the asymmetry.
EMG Recording and Analysis
The EMG was recorded by placing two Ag/AgCl electrodes, proximal and distal, about 5 cm apart, over the flexor digitorum superficialis muscles, continuously and synchronously with the EEG (same recording parameters). For each hand, the muscle activation signal was calculated as the difference between the proximal and distal electrodes. Offline, the data were digitally filtered with a high-pass filter of 15 Hz (zero-phase shift, 48 dB/octave Butterworth filter) and segmented according to the experimental conditions (see EEG Recording and ERP Analysis sections). Muscle activity was extracted using a moving window (15 points length) root mean square over the EMG signal. Muscle activation was defined as root mean square exceeding 4 standard deviations from the 150 msec baseline interval before prime onset. The onset of the EMG response was defined as the first sample point exceeding this threshold. To evaluate EMG activity without overt action, onset was calculated for left and right hand regardless of the hand executing the final button press.
We examined four manifestations of a ChoI in the EMG activity. First, we tested whether an incongruent prime increased the probability for an EMG response in the nonresponding hand relative to a congruent prime. Because percentages do not distribute normally, we used a permutation-based test, in which the true difference between the average percentages of trials in the two conditions is compared against a permutation null distribution. This null distribution was constructed by performing 10,000 permutations, in each of which the labels of the conditions were randomly shuffled for each participant and the resulting surrogate average difference between the percentages was registered. Second, we tested whether the latency of the EMG onset differed between the responding and nonresponding hands. For this purpose, we calculated for each participant the mean latency of EMG onset across Instructed and Free-choice conditions and compared the responding and nonresponding hands using paired t test. Third, we tested whether, in the Instructed and Free-choice incongruent condition, the latency of EMG onset in the responding hand was delayed when there was an EMG response in the nonresponding hand using a paired t test. Participants were excluded from the second and third analysis if they had less than three trials with an EMG response in the nonresponding hand. Fourth, given that the presence of EMG activity in the nonresponding hand is associated with a delay of response in the responding hand, we asked whether the delay of RT in incongruent relative to congruent trials could be fully explained by the presence of more trials in which there was EMG activation in the nonresponding hand in the incongruent relative to congruent condition. To this end, we tested whether the latency of EMG onset in the responding hand was delayed in incongruent trials relative to congruent trials, exclusively in trials in which there was no EMG response in the nonresponding hand, again using a paired t test.
Participants performed two types of trials (Figure 1). Instructed trials consisted of a masked prime (left or right arrow, or neutral two-way arrow) followed by a left or right arrow cue. The task was to press the right or left button with the respective hand quickly and accurately according to the instructing cue. Free-choice trials consisted of a masked prime, which could be directional or neutral, followed by a Free-choice cue, and the task was to press the left or right button quickly according to what the participant wanted at the moment of the cue. In all trial types, the participants were not informed of the existence of the prime.
Six of 20 participants had higher than chance-level performance in the objective evaluation of awareness task (2AFC detection of the masked prime arrow cue) and were thus excluded from the analysis. The performance of each of the remaining 14 participants (8 women, mean age = 23.8, SD = 2.74) did not differ from chance level in the detection test. The mean of correct responses in the objective detection test in this group of participants was 51.29%, which was not significantly different from chance (t(13) = 0.86, p > .2, lower bound of one-tailed 95% confidence range: 48.83%). In the discrimination task, which was not part of our selection process, these participants scored an average of 49.71%, SD = 7.91.
As expected from earlier studies with a short prime-to-cue interval, participants showed a Positive Compatibility Effect in the Instructed condition (Kiesel et al., 2006; Schlaghecken & Eimer, 2004; Eimer & Schlaghecken, 2003; Dehaene et al., 1998; Neumann & Klotz, 1994). RTs in the Instructed condition were modulated by the congruency of the prime (F(2, 26) = 64.62, p < .001, MSE = 163). RTs were slower for incongruent compared to congruent trials (t(13) = 10.78, p < .0001, Figure 2A). Congruent and neutral trials did not differ in their RT (t(13) = 0.07, p > .05), whereas neutral trials were significantly slower than incongruent trials (t(13) = 9.69, p < .05, Scheffe corrected for post hoc comparisons). Error rates were also modulated by the congruency of the prime (F(2, 26) = 12.07, p < .005, MSE = 26), with higher error rates in the incongruent compared to congruent condition, ruling out a speed–accuracy tradeoff (t(13) = 4.03, p < .005). However, unlike in the case of RTs, the neutral condition revealed higher error rates than in the congruent condition, (t(13) = 3.43, p < .05, Scheffe corrected for post hoc comparisons), but no difference from the incongruent (t(13) = 1.16, p > .05). Overall, masked primes affected participants' responses despite the fact that they were not consciously detectable.
In the Free-choice condition, participants were free to choose the direction of response as they wish. Nevertheless, response direction was biased toward the prime direction, with 57.6% of the responses being congruent with the prime, significantly more than 50% chance level (t(13) = 4.15, p < .05; Figure 2B). This bias was also significant individually in 10 of the 14 participants (using a test similar to the test for objective awareness). In addition, RTs in the Free-choice condition were affected by the congruency between the prime and the chosen response direction (congruent/incongruent/neutral, F(2, 26) = 40.65, p < .001, MSE = 267). Planned contrasts revealed that RTs were slower when choosing a response incongruent with the prime than when choosing a response congruent with the prime (t(13) = 4.93, p < .0005). Thus, although the prime was not perceivable by participants, it affected their responses both when instructed to respond in a given direction and when choosing without instruction based on their own will. Interestingly, the RTs in the Free-choice condition when the masked prime was neutral were even shorter than in the primed congruent choices (t(13) = 5.37, p < .05, Scheffe corrected for post hoc comparisons), suggesting a cost for a meaningful prime regardless of identity.
The behavioral results confirmed that incongruent subliminal primes interfere with the process of initiating and executing a motor plan, whether triggered by an external cue (in the Instructed condition) or by an internal decision (in the Free-choice condition). However, the nature of interference could be manifested as a weakening of the strength of intention, as prolonging the latency of intention formation because of confusion or hesitation, or as the actual initiation of a motor plan to the incongruent direction, followed by a ChoI. We tested this issue by measuring EEG and EMG during performance of the task.
To identify preparation of left- or right-hand movement, we used the LRP, calculated as C3 (left) minus C4 (right). In this derivation, preparation for right- or left-hand movements is typically observed as a negative or positive deflection of the LRP, respectively. Indeed, in the response-locked averages, the LRPs were characterized by a negative or positive deflection peaking around 100 msec before right or left button press, respectively. The button press was followed by a smaller deflection in the same direction as the preresponse LRP, followed by a reversal of the LRP polarity. Statistical analysis of the preresponse interval (Figure 3A) revealed a main effect of Response direction (F(1, 13) = 119.99, p < .001) and no effect of Prime congruency or interaction between Prime congruency and Response directions (both Fs < 1).
We first examine the pattern of brain responses in the Instructed condition. The modulation of the LRPacross hands by prime congruency (congruent, incongruent, or neutral trials) in the Instructed condition is illustrated in Figure 3B. The most prominent pattern in all three conditions is a negative deflection, that is, in the direction of the responding hand, between ∼350 and 550 msec. Notwithstanding, trials with congruent and incongruent primes differed significantly during the interval marked in Figure 3B as “prime effect” (199–339 msec, p < .01, cluster-based permutation analysis; gray line and blue bar in Figure 3B). Importantly, the positive deflection of the incongruent response (i.e., in the direction of the prime and opposite the direction of the responding hand) was significant compared to zero in the “prime effect” window of interest (174–316 msec, p < .01, cluster-based permutation analysis). The LRP elicited in trials with neutral primes was intermediate between the congruent and incongruent conditions and exhibited no such absolute positive shift. These results suggest that the difference between the congruent and incongruent response resulted from an actual preparation to move in the direction of the prime and not from a mere modulation of the amplitude of a response in the direction of the responding hand nor from a change in the latency of the response (note also the opposite polarity effects when the data is not pooled across hands; Figure 3C).
These dynamics of brain activity across time can be further evaluated by examining the current source density (CSD) topographies (scalp surface Laplacian). By filtering out the effect of remote sources, CSDs provide closer approximation of the cortical surface activity than voltage topographies (Perrin, Pernier, Bertrand, & Echallier, 1989; Figure 4). The CSD maps show stronger negativity contralateral to the prime direction within the earlier windows up to ∼330 msec, which becomes stronger contralateral to the cue/response hand thereafter. This is reflected most clearly in a shift in lateralization of the negativity in incongruent trials, from right to left in LRR trials (second row) and from left to right in RLL trials (fourth row).
Next, we turned to the Free-choice condition, asking whether there is a pattern of ChoI in incongruent trials in the LRP, as was found for the Instructed condition. Because in the Free-choice condition, there were no explicit instructions, congruency was defined based on the matching between the prime and the response. Thus, a left arrow prime followed by a right response or vice versa is considered incongruent. Analysis of the LRPacross hands revealed similar results to those found in the Instructed case. That is, there was a deflection in the direction of the responding hand at about 350–650 msec (Figure 5A) yet, as in the Instructed case, the congruent and incongruent conditions differed in the earlier time window (265–330 msec, p < .05, cluster-based permutation test; blue bar in Figure 5A). Specifically, the polarity in this earlier window shifted in the direction of the prime. Again, this pattern is also reflected in the CSD topographies (Figure 5B). These results reflect a ChoI pattern in the Free-choice condition.
To summarize, the pattern of results in both the Instructed and Free-choice trials is compatible with a ChoI scenario, when choices are mapped to right-/left-hand movements: activation over the motor area in the direction of the initial primed intention, which is overcome by activation over the contralateral scalp in the direction of the final intention.
In conscious flanker-type tasks, incorrect hand EMG activity is sometimes found in trials which nevertheless end with a correct hand response (Kopp, Rist, & Mattler, 1996; Smid, Mulder, & Mulder, 1990; Coles, Gratton, Bashore, Eriksen, & Donchin, 1985). We thus tested whether ChoI was observed in peripheral muscle activation measured with EMG. The percentage of trials with EMG activity in the nonresponding hand was significantly larger in the incongruent conditions than in the congruent conditions for both Instructed and Free-choice conditions (permutation-based paired test: p < .005 and p < .05 respectively; Figure 6B). EMG onsets of the nonresponding hand in the incongruent condition (Figure 6C; showing the Instructed and Free-choice combined) preceded the EMG onset of the responding hand (t(11) = 2.61, p < .05; two participants eliminated using criteria of minimum three trials with EMG activity in nonresponding hand). In the congruent condition, there was no such difference (t(8) = 0.16, p = .88; 5 participants eliminated). Nonresponding hand onsets in the incongruent condition occurred most frequently between 200 and 340 msec after the prime onset (Figure 6C, top). This is roughly the same temporal window in which the LRP demonstrated activation commensurate with preparing the nonresponding hand. In contrast, in the congruent condition, rare EMG activity in the nonresponding hand tended to co-occur with the activation of the responding hand (Figure 6C, bottom). Finally, the EMG onset of the responding hand in the Instructed and Free-choice incongruent condition was significantly slower when there was EMG activity in the nonresponding hand compared to the cases where there was no EMG activity in the nonresponding hand (Instructed condition [four participants eliminated]: t(9) = 5.19, p < .001; Free-choice condition [eight participants eliminated]: t(5) = 5.59, p < .005; Figure 6D). However, incongruent trials were still slower than congruent trials even when there was no EMG activity in the nonresponding hand (Instructed condition: t(13) = 8.9, p < .001; Free-choice condition: t(13) = 2.78, p < .05; Figure 6E).
Faced with a decision with a correct choice, changing one's mind may have various reasons, such as change in motivation or the accumulation of evidence during processing (Resulaj, Kiani, Wolpert, & Shadlen, 2009; Song & Nakayama, 2009). In this article, we investigated the nature of picking type selections, where no correct choice exists. What is the temporal dynamics of intention formation when the decision is apparently arbitrary?
Replicating earlier findings (Kiesel et al., 2006; Schlaghecken & Eimer, 2004; Eimer & Schlaghecken, 2003; Dehaene et al., 1998; Neumann & Klotz, 1994), the behavioral results of this study express the Positive Compatibility Effect (Figure 2).3 We found that in the Instructed condition error rates were significantly greater in incongruent trials than in congruent trials and that in correct trials RTs were longer when the instruction cue was preceded by incongruent primes compared to congruent and neutral primes. However, in the Instructed condition, congruent primes did not reduce the RT in comparison to the neutral (this was found also in a Flanker study; Praamstra, Plat, Meyer, & Horstink, 1999), and the error rate in the neutral prime case was closer to the incongruent rate than to the congruent. Thus, if the neutral prime case is taken as baseline, it seems that congruence affects accuracy and incongruence affects speed (but see Klotz & Neumann, 1999; Neumann & Klotz, 1994, who found that a congruent prime reduced RT in comparison to a neutral prime, whereas neutral error rate was statistically at the level of the congruent trials).
In Free-choice trials, responses were biased by the direction of the masked prime as well; in addition, RTs were slower when choosing a response incongruent with the prime than when choosing a response congruent with the prime, as previously found (Kiesel et al., 2006; Schlaghecken & Eimer, 2004). Interestingly, the mean RT in the neutral Free-choice condition was even shorter than in congruent Free-choice condition, suggesting a cost for any biasing prime whether congruent or incongruent; earlier studies on Free-choice conditions have not done or reported the neutral Free-choice condition. Regardless, the results of both the Instructed and Free-choice conditions suggest that the prime biased the response selection in our study.
Prime-induced Motor Cortex Activation
We replicated the typical finding for the Instructed incongruent condition, finding the neural pattern of ChoI—the LRP first reflects preparation to move the hand in the direction indicated by the prime cue and then reverses to the direction of the hand that finally moved, suggesting a ChoI within a temporal window of about 200–340 msec after prime onset (“prime effect” temporal window; Figures 3 and 4; cf. Eimer & Schlaghecken, 2003; Jaskowski et al., 2003; Dehaene et al., 1998). As noted, in one exceptional study, Kiesel et al. (2006) did not find the ChoI pattern of LRP polarity reversal—the LRP in the incongruent Instructed condition was only delayed, reflecting the direction of the final choice. A possible explanation for this atypical result is the physical difference in that study between the primes and the instructive cues, as it was shown that creating an explicit mapping of the stimuli to the response in the current task is essential (Schlaghecken & Eimer, 2004). Using prime stimuli, which are not active in the task set, can result in visual confusion rather than an effect on the motor system.
Notwithstanding, what was missing in most previous studies was an electrophysiological measure of the Free-choice condition expressing the move from an exogenously prompted intention (via priming) to an endogenously prompted intention. The theoretical consequences of this condition could be as follows: (1) That the prime creates only hesitation because “participants choose which response to perform rather early, probably before the prime had induced any motor activation. Then the incongruent prime delays the motor activation of the response that has been chosen” (Kiesel et al., 2006, p. 406). In this case, there should be no formation of a prime-induced motor asymmetry, but only a delay in the LRP in the direction congruent with the response. (2) That the prime induces motor asymmetry, which fully determines the response direction. The incongruent Free-choice trials under this premise are cases in which the prime was not adequately processed and failed to induce motor asymmetry. (3) That the prime induces motor activation and forms an early neural asymmetry in its direction, but this is endogenously substituted by the final neural asymmetry reflecting the chosen response direction. The latter option reflects the ChoI scenario.
In the only previous attempt to test this question, Kiesel et al. (2006) found that the LRP in the incongruent condition was delayed, reflecting the direction of the final choice, but with no change of polarities that would suggest a ChoI (Option 1 above). However, as Kiesel et al. exceptionally did not find a ChoI pattern even in their instructed condition (see discussion above), this result is hard to interpret. In contrast to Kiesel et al.'s results, in our study the ChoI LRP pattern was observed in the Free-choice incongruent case, within a similar temporal window as in the Instructed case (Figure 5). This neural signature rules out both Options 1 and 2 and allows us to explain the behavioral cost of incongruence, in both Instructed and Free-choice trials, as a result of a ChoI scenario (Option 3). Moreover, the prime-induced motor cortex activation can be interpreted as the source for the bias toward the prime in the Free-choice congruent trials. On the basis of the similarity between the ChoI pattern in the Instructed and the Free-choice condition, we carefully suggest that, just as the decision to act in the Instructed condition was formed when instructive cue was presented, in the Free-choice condition the endogenous intention was formed only after the free-choice cue was presented and in some cases reversed an already formed intention (induced by the prime) to move with the other hand.
Indeed, given that the prime induces asymmetrical motor preparation in the Instructed condition, the fact that the same is observed in the Free-choice condition is not surprising, as in our experiment when the prime appeared it was not yet known whether it is an Instructed or Free-choice trial. What is novel and surprising is that despite the fact that the prime clearly created an asymmetry and induced a tendency to move in one direction, which is evidenced by the prime-related LRP, participants “freely” overcame this and moved the other hand. That is, we deliberately created a bias (by presenting the prime), we have evidence that the prime worked and created a final plan to move in a certain direction, yet the choice was in the other direction. This is inconsistent with the idea that the underlying asymmetry (even at this final stage) determines the choice.
Comparing Peripheral and Central Motor Activations
Earlier studies in conscious Flanker and Stroop-type tasks have shown that some responses contained incorrect hand muscle activity (measured via EMG) that nevertheless ended in a correct response (Szűcs, Soltész, & White, 2009; Kopp et al., 1996; Smid et al., 1990; Coles et al., 1985). Similar effects were found in our experiment using masked primes. Although EMG peripheral activity of the nonresponding hand was occasionally recorded in all types of trials, it was significantly more frequent in the incongruent trials, both in the Instructed as well as in Free-choice conditions (Figure 6B). The distribution of the EMG onsets shows that in incongruent trials the nonresponding hand activation occurred usually before the responding hand acted and within the temporal window of the prime effect found by the LRP (Figure 6C). Additionally, the responding hand RTs were slower in trials with preceding EMG activity in the nonresponding hand (Figure 6D). These results suggest that the peripheral activity in the incongruent case is part of the ChoI scenario. The primed intention not only activated the central motor cortex before it was overcome by the final intention, but in a portion of the trials this intention “leaked” to the periphery and activated the muscles. These findings may be interpreted as showing that commands are sent down to the periphery before the final decisions are made and suggest that that the so-called “moment of decision” and the race between the alternatives continue beyond the confines of the brain into the periphery (see further evidence and discussion in McBride, Boy, Husain, & Sumner, 2012). They undermine the common assumption that the executive decision process (central brain activity) and the execution process (peripheral muscle activity), or the deliberation and commitment to act (Thura & Cisek, 2014), are distinct serial stages.
The question of why the EMG activity does not result in actual motion and button press remains open. One option is that the activation is not strong enough to recruit enough muscle units to move the limb. Another option is that there is coactivation of agonists and antagonists preventing movement until the final decision is made. These options can be addressed with more extensive measurement of EMG in future studies.
Implications Regarding Decision-making in Picking Situations
Taken together, our findings reveal a ChoI because of a sequence of conflicting exogenous instructions (the Instructed condition) or because of a change from an exogenously induced intention to an endogenous intention (the Free-choice condition). As asserted at the outset of this article, what drives the decisions in a picking situation—when on an explicit or reason-based level there is complete symmetry between the alternatives—is puzzling. One solution to this question is to invoke a concept of independent “free will,” which arbitrates and determines the choice (as expressed for example by Newton in his reply to Leibniz; see Ariew, 2000, fourth reply). A more biological solution is that, although there is no explicit advantage to one of the choices, on a lower, implicit level the symmetry does not maintain. For example, if one has to choose between two similar items placed side by side, a coincidental spatial attention bias toward either side may determine the decision. In the present example, if the left motor cortex happens to be in a more prepared state at the moment when the choice cue appears, one would use the right hand, whereas if the right motor cortex happens to be more aroused, the left hand will act. That is, low level, implicit neuronal asymmetry between alternatives is what makes possible and determines the picking of one alternative rather than the other. However, this study indicates that in free picking situations instantaneous neuronal asymmetries associated with the agent's choice do not determine the final outcome but reflect complex dynamics that may result in a ChoI. In particular, although neuronal asymmetry was exogenously induced by the masked prime and final stages of an intention was formed in one direction as documented by the LRP and the EMG, the agent could many times act against this established asymmetry and form an opposing intention. Importantly, such reversals are not the result of cognitive deliberations, as they occurred during the course of a single trial of a rapid picking task.
The Role of Internal Noise in Proximal Picking Type Situations
What can then explain the ChoI in the Free-choice scenario? If indeed there was a formation of neural asymmetry in the Free-choice condition, why are there responses in a direction opposing the prime? A possible mechanism for generating this type of dynamics is the presence of continuous neuronal noise that is injected into the system. A commonly postulated model for evidence-based decision-making is drift diffusion, in which the accumulating evidence for two alternatives are integrated over time and compared to a fixed decision boundary (Gold & Shadlen, 2007; Ratcliff, 1978). This model is considered an optimal strategy of overcoming the unavoidable noise that accompanies the evidence signal (for a review, see Miller & Katz, 2013). However, when the information for a certain decision is weak or ambiguous, the threshold crossing is mainly determined by subthreshold neuronal noise (Schurger, Sitt, & Dehaene, 2012; Shadlen, Britten, Newsome, & Movshon, 1996). According to our hypothesis, in the Free-choice case continuous neuronal noise is integrated as a left/right “evidence” by the decision-making circuit, whereas in Instructed tasks, “evidence” consists of both noise and the signal generated by the instructing cue. Thus, rapid changes of intention in Free-choice may reflect the impact of rapid neural noise on the ongoing decision-making system.
As we noted elsewhere (Furstenberg et al., 2015), Ullmann-Margalit and Morgenbesser (1977) point to the fact that young children turn every task into a choosing task. Children see meaning in every selection task, even if they do not know how to articulate this meaning. Therefore, young children trying to select in a picking task (say, select one candy piece from a bag full of identical red candies) deliberate and change their minds several times in the process of selecting, as if it were a reasoned choosing task. This could be explained either by suggesting that the level of internal noise is high, shifting the asymmetry back and forth before it reaches a level which amounts to a decision, or conversely, that children lack sufficient low level noise necessary for decision-making in picking situations, making it necessary to convert the process of picking to one of choosing to break the symmetry, even if it relies on a made-up logic. We speculate that either one of these deficits could be because of prematurity of the frontal lobes.
We thank Venkatakrishnan Ramaswamy for helpful discussions. This research was supported, in part, by the John Templeton Foundation's Big Questions in Free Will program, the James S. McDonnell Foundation, and the Gatsby Charity Foundation. The opinions expressed in this publication are those of the authors and do not necessarily reflect the views of these agencies.
Reprint requests should be sent to Ariel Furstenberg, The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Edmond J. Safra Campus, Givat Ram, Jerusalem, Israel, 91904, or via e-mail: email@example.com.
Similar statements were made by other philosophers including Aristotle, Aquinas, Leibniz, and others.
It could be argued that because Free-choice incongruent decisions still exist (i.e., the participants do not always chose the direction indicated by the prime) then the low-level asymmetry solution for picking is refuted anyway. However, incongruent situations could be explained by a failure of the subliminal prime to have a significant neuronal effect and to cause an asymmetry.