Abstract
Article Figures and data Abstract Introduction Results Discussion Materials and methods Data availability References Decision letter Author response Article and author information Metrics Abstract The readiness potential (RP)—a key ERP correlate of upcoming action—is known to precede subjects’ reports of their decision to move. Some view this as evidence against a causal role for consciousness in human decision-making and thus against free-will. But previous work focused on arbitrary decisions—purposeless, unreasoned, and without consequences. It remains unknown to what degree the RP generalizes to deliberate, more ecological decisions. We directly compared deliberate and arbitrary decision-making during a 1000-donation task to non-profit organizations. While we found the expected RPs for arbitrary decisions, they were strikingly absent for deliberate ones. Our results and drift-diffusion model are congruent with the RP representing accumulation of noisy, random fluctuations that drive arbitrary—but not deliberate—decisions. They further point to different neural mechanisms underlying deliberate and arbitrary decisions, challenging the generalizability of studies that argue for no causal role for consciousness in decision-making to real-life decisions. Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review. The Reviewing Editor's assessment is that all the issues have been addressed (see decision letter). https://doi.org/10.7554/eLife.39787.001 Introduction Humans typically experience freely selecting between alternative courses of action, say, when ordering a particular item off a restaurant menu. Yet a series of human studies using electroencephalography (EEG) (Haggard and Eimer, 1999; Libet et al., 1983; Salvaris and Haggard, 2014), fMRI (Bode and Haynes, 2009; Bode et al., 2011; Soon et al., 2008; Soon et al., 2013), intracranial (Perez et al., 2015), and single-cell recordings (Fried et al., 2011) challenged the veridicality of this common experience. These studies found neural correlates of decision processes hundreds of milliseconds and even seconds prior to the moment that subjects reported having consciously decided. The seminal research that launched this series of studies was conducted by Benjamin Libet and colleagues (Libet et al., 1983). There, the authors showed that the readiness potential (RP)—a ramp-up in EEG negativity before movement onset, thought to originate from the presupplementary motor area (pre-SMA)—began before subjects reported a conscious decision to act. Libet and colleagues took the RP to be a marker for an unconscious decision to act (Libet et al., 1983; Soon et al., 2008) that, once it begins, ballistically leads to action (Shibasaki and Hallett, 2006). Under that interpretation, the fact that RP onset precedes the report of the onset of the conscious decision to act was taken as evidence that decisions about actions are made unconsciously. And thus the subjective human experience of freely and consciously deciding to act is but an illusion (Harris, 2012; Libet et al., 1983; Wegner, 2002). This finding has been at the center of the free-will debate in neuroscience for almost four decades, captivating scholars from many disciplines in and outside of academia (Frith et al., 2000; Frith and Haggard, 2018; Haggard, 2008; Jeannerod, 2006; Lau et al., 2004; Mele, 2006; Wegner, 2002). Critically, in the above studies, subjects were told to arbitrarily move their right hand or flex their right wrist; or they were instructed to arbitrarily move either the right or left hand (Haggard, 2008; Hallett, 2016; Roskies, 2010). Thus, their decisions when and which hand to move were always unreasoned, purposeless, and bereft of any real consequence. This stands in sharp contrast to many real-life decisions that are deliberate—that is, reasoned, purposeful, and bearing consequences (Ullmann-Margalit and Morgenbesser, 1977): which clothes to wear, what route to take to work, as well as more formative decisions about life partners, career choices, and so on. Deliberate decisions have been widely studied in the field of neuroeconomics (Kable and Glimcher, 2009; Sanfey et al., 2006) and in perceptual tasks (Gold and Shadlen, 2007). Yet, interestingly, little has been done in that field to assess the relation between decision-related activity, subjects’ conscious experience of deciding, and the neural activity instantaneously contributing to this experience. Though some studies compared, for example, internally driven and externally cued decisions (Thut et al., 2000; Wisniewski et al., 2016), or stimulus-based and intention-based actions (Waszak et al., 2005), these were typically arbitrary decisions and actions with no real implications. Therefore, the results of these studies provide no direct evidence about potential differences between arbitrary and deliberate decisions. Such direct comparisons are critical for the free will debate, because it is deliberate, rather than arbitrary, decisions that are at the center of philosophical arguments about free will and moral responsibility (Breitmeyer, 1985; Maoz and Yaffe, 2016; Roskies, 2010). Deliberate decisions typically involve more conscious and lengthy deliberation and might thus be more tightly bound to conscious processes than arbitrary ones. Consequently, if the RP is a marker for unconscious decisions, while deliberate decisions are driven more by conscious than by unconscious processes, then the RP might be substantially diminished, or even absent, for deliberate decisions. Another reason that the RP might be completely absent during deliberate decisions has to do with a recent computational model (Schurger et al., 2012). This model claims that the RP—which has been deemed a preparatory signal with a causal link to the upcoming movement—actually reflects an artifact that results from a combination of (i) biased sampling stemming from the methodology of calculating this component and (ii) autocorrelation (or smoothness) in the EEG signal. The RP is calculated by aligning EEG activity (typically in electrode Cz) to movement onset, then segmenting a certain time duration around each movement onset (i.e., epoching), and finally averaging across all movements. Hence, we only look for an RP before movement onset, which results in biased sampling (as ‘movement absent’ is not probed). Put differently, we search for and generally find a ramp up in EEG negativity in Cz before movement onset. But we do not search for movement onset every time there is a ramp up in EEG negativity on Cz. What is more, as EEG is autocorrelated, ramps up or down are to be expected (unlike, say, for white-noise activity). Schurger and colleagues demonstrated that RPs can be modeled using a mechanistic, stochastic, autocorrelated, drift-diffusion process that integrates noise to a bound (or threshold; see Model section in Materials and methods for details). In the model, it is only the threshold crossing that reflects decision completion and directly leads to action. And thus the beginning of (what is in hindsight and on average) the ramp up toward the threshold is certainly not the completion of the decision that ballistically leads to the threshold crossing and hence to movement onset (Schurger et al., 2012). This interpretation of the RP thus takes the sting out of the Libet argument against free will, as the latter was based on interpreting the RP as reflecting an unconscious decision to act. Importantly for our purposes, within the framework of the model, this artificial accumulation of stochastic fluctuations toward a threshold is expected to occur for arbitrary decisions, but not for deliberate ones. Unlike arbitrary decisions, deliberate decisions are generally not driven by random fluctuations. Rather, it is the values of the decision alternatives that mainly drive the decision and ultimately lead to action. Therefore, if the RP indeed reflects the artificial accumulation of stochastic fluctuations, as the model suggests, a key prediction of the model is that no RP will be found for deliberate decisions (see more below). Thus, demonstrating the absence of an RP in deliberate decisions challenges the interpretation of the RP as a general index of internal, unconscious decision-making; if this interpretation were correct, such a marker should have been found for all decision types. What is more, and importantly, it questions the generalizability of any studies focused on arbitrary decisions to everyday, ecological, deliberate decisions. In particular, it challenges RP-based claims relating to moral responsibility (Haggard, 2008; Libet, 1985; Roskies, 2010), as moral responsibility can be ascribed only to deliberate decisions. Here, we tested this hypothesis and directly compared the neural precursors of deliberate and arbitrary decisions—and in particular the RP—on the same subjects, in an EEG experiment. Our experiment utilized a donation-preference paradigm, in which a pair of non-profit organizations (NPOs) were presented in each trial. In deliberate-decision trials, subjects chose to which NPO they would like to donate 1000. In arbitrary-decision trials, both NPOs received an equal donation of $500, irrespective of subjects’ key presses (Figure 1). In both conditions, subjects were instructed to report their decisions as soon as they made them, and their hands were placed on the response keys, to make sure they could do so as quickly as possible. Notably, while the visual inputs and motor outputs were identical between deliberate and arbitrary decisions, the decisions’ meaning for the subjects was radically different: in deliberate blocks, the decisions were meaningful and consequential—reminiscent of important, real-life decisions—while in arbitrary blocks, the decisions were meaningless and bereft of consequences—mimicking previous studies of volition. Figure 1 Download asset Open asset Experimental paradigm. The experiment included deliberate (red, left panel) and arbitrary (blue, right panel) blocks, each containing nine trials. In each trial, two causes—reflecting NPO names—were presented, and subjects were asked to either choose to which NPO they would like to donate (deliberate), or to simply press either right or left, as both NPOs would receive an equal donation (arbitrary). They were specifically instructed to respond as soon as they reached a decision, in both conditions. Within each block, some of the trials were easy (lighter colors) decisions, where the subject’s preferences for the two NPOs substantially differed (based on a previous rating session), and some were hard decisions (darker colors), where the preferences were more similar; easy and hard trials were randomly intermixed within each block. To make sure subjects were paying attention to the NPO names, even in arbitrary trials, and to better equate the cognitive load between deliberate and arbitrary trials, memory tests (in light gray) were randomly introduced. There, subjects were asked to determine which of four NPO names appeared in the immediately previous trial. For a full list of NPOs and causes see Supplementary file 1. https://doi.org/10.7554/eLife.39787.002 Results Behavioral results Subjects’ reaction times (RTs) were analyzed using a 2-way ANOVA along decision type (arbitrary/deliberate) and difficulty (easy/hard). This was carried out on log-transformed data (raw RTs violated the normality assumption; W = 0.94, p=0.001). As expected, subjects were substantially slower for deliberate (M = 2.33, SD = 0.51) than for arbitrary (M = 0.99, SD = 0.32) decisions (Figure 2, left; F(1,17)=114.87, p<0.0001 for the main effect of decision type). A main effect of decision difficulty was also found (F(1,17)=21.54, p<0.0005), with difficult decisions (M = 1.77, SD = 0.40) being slower than easy ones (M = 1.56, SD = 0.28). Importantly, subjects were significantly slower for hard (M = 2.52, SD = 0.62) vs. easy (M = 2.13, SD = 0.44) decisions in the deliberate case (t(17)=4.78, p=0.0002), yet not for the arbitrary case (M = 1.00, SD = 0.34; M = 0.98, SD = 0.32, for hard and easy arbitrary decisions, respectively; t(17)=1.01, p=0.33; F(1,17)=20.85, p<0.0005 for the interaction between decision type and decision difficulty). This validates our experimental manipulation and further demonstrates that, in deliberate decisions, subjects were making meaningful decisions, affected by the difference in the values of the two NPOs, while for arbitrary decisions they were not. What is more, the roughly equal RTs between easy and hard arbitrary decisions provide evidence inconsistent with concerns that subjects were deliberating during arbitrary decisions. Figure 2 Download asset Open asset Behavioral results. Reaction Times (RTs; left) and Consistency Grades (CG; right) in arbitrary (blue) and deliberate (red) decisions. Each dot represents the average RT/CG for easy and hard decisions for an individual subject (hard decisions: x-coordinate; easy decisions: y-coordinate). Group means and SEs are represented by dark red and dark blue crosses. The red and blue histograms at the bottom-left corner of each plot sum the number of red and blue dots with respect to the solid diagonal line. The dashed diagonal line represents equal RT/CG for easy and hard decisions; data points below that diagonal indicate longer RTs or higher CGs for hard decisions. In both measures, arbitrary decisions are more centered around the diagonal than deliberate decisions, showing no or substantially reduced differences between easy and hard decisions. https://doi.org/10.7554/eLife.39787.003 The consistency between subjects’ choices throughout the main experiment and the NPO ratings they gave prior to the main experimental session was also analyzed using a 2-way ANOVA (see Materials and methods). As expected, subjects were highly consistent with their own, previous ratings when making deliberate decisions (M = 0.91, SD = 0.04), but not when making arbitrary ones (M = 0.52, SD = 0.04; Figure 2, right; F(1,17)=946.55, p<0.0001, BF = 2.321029) for the main effect of decision type. A main effect of decision difficulty was also found (F(1,17)=57.39, p<0.0001, though BF = 1.57), with hard decisions evoking less consistent scores (M = 0.66, SD = 0.05) than easy ones (M = 0.76, SD = 0.03). Again, decision type and decision difficulty interacted (F(1,17)=25.96, p<0.0001, BF = 477.47): subjects were much more consistent with their choices in easy (M = 0.99, SD = 0.02) vs. hard (M = 0.83, SD = 0,64) deliberate decisions (t(17)=11.15, p<0.0001, BF = 3.68106), than they were in easy (M = 0.54, SD = 0.07) vs. hard (M = 0.49, SD = 0.05) arbitrary decisions (t(17)=2.50, p=0.023, BF = 2.69). Nevertheless, though subjects were around chance (i.e., 0.5) in their consistency in arbitrary decisions (ranging between 0.39 and 0.64), it seems that some subjects were slightly influenced by their preferences in easy-arbitrary decisions trials, resulting in the significant difference between hard-arbitrary and easy-arbitrary decisions above, though the Bayes factor was inconclusive. Finally, no differences were found between subjects’ tendency to press the right vs. left key in the different conditions (both main effects and interaction: F < 1). EEG results: Readiness Potential (RP) The RP is generally held to index unconscious readiness for upcoming movement (Haggard, 2008; Kornhuber and Deecke, 1990; Libet et al., 1983; Shibasaki and Hallett, 2006); although more recently, alternative interpretations of the RP have been suggested (Miller et al., 2011; Schmidt et al., 2016; Schurger et al., 2012; Trevena and Miller, 2010; Verleger et al., 2016). It has nevertheless been the standard component studied in EEG versions of the Libet paradigm (Haggard, 2008; Haggard and Eimer, 1999; Hallett, 2007; Libet, 1985; Libet et al., 1983; Libet et al., 1982; Miller et al., 2011; Schurger et al., 2012; Shibasaki and Hallett, 2006; Trevena and Miller, 2010). As is common, we measured the RP over electrode Cz in the different conditions by averaging the activity across trials in the 2 s prior to subjects’ movement. Focusing on the last 500 ms before movement onset for our statistical tests, we found a clear RP in arbitrary decisions, yet RP amplitude was not significantly different from 0 in deliberate decisions (Figure 3A; F(1,17)=11.86, p=0.003, BF = 309.21 for the main effect of decision type; in t-tests against 0 for this averaged activity in the different conditions, corrected for multiple comparisons, an effect was only found for arbitrary decisions (hard: t(17)=5.09, p=0.0001, BF = 307.38; easy: t(17)=5.75, p<0.0001, BF = 1015.84) and not for deliberate ones). The Bayes factor—while trending in the right direction—indicated inconclusive evidence (hard: t(17)=1.24, p>0.5, BF = 0.47; easy: t(17)=1.84, p=0.34, BF = 0.97). Our original baseline was stimulus locked (see Materials and methods). And we hypothesized that the inconclusive Bayes factor for deliberate trials had to do with a constant, slow, negative drift that our model predicted for deliberate trials (see below) rather than reflecting a typical RP. As the RTs for deliberate trials were longer than for arbitrary ones, this trend might have become more pronounced for those trials. To test this, we switched the baseline period to −1000 ms to −500 ms relative to movement onset (i.e., a baseline that immediately preceded our time of interest window). Under this analysis, we found moderate evidence that deliberate decisions (pooled across decision difficulty) are not different from 0 (BF = 0.332), supporting the claim that the RP during the last 500 ms before response onset was completely absent (BF for similarly pooled arbitrary decisions was 5.07·104). Figure 3 Download asset Open asset The readiness potentials (RPs) for deliberate and arbitrary decisions. (A) Mean and SE of the readiness potential (RP; across subjects) in deliberate (red shades) and arbitrary (blue shades) easy and hard decisions in electrode Cz, as well as scalp distributions. Zero refers to time of right/left movement, or response, made by the subject. Notably, the RP significantly differs from zero and displays a typical scalp distribution for arbitrary decisions only. Similarly, temporal clusters where activity was significantly different from 0 were found for arbitrary decisions only (horizontal blue lines above the x axis). Scalp distributions depict the average activity between −0.5 and 0 s, across subjects. The inset bar plots show the mean amplitude of the RP, with 95% confidence intervals, over the same time window. Response-locked potentials with an expanded timecourse, and stimulus-locked potentials are given in Figure 6B and A, respectively. The same (response-locked) potentials as here, but with a movement-locked baseline of −1 to −0.5 s (same as in our Bayesian analysis), are given in Figure 6C. (B) subjects’ Cz activity in the four conditions = The line for against time over the last ms before response onset is by a line. The lines have significantly different from 0 for arbitrary decisions only. that the to an RP only in arbitrary decisions. In an to further test for time where the RP is different from 0 for deliberate and arbitrary trials, we a and for all four conditions against the (see Materials and we found a of that differed from 0 in both arbitrary conditions by lines above the x in Figure The same no clusters of activity from zero in either of the deliberate conditions. In a against time for the last ms before response onset, the trend was significant for arbitrary decisions (Figure p<0.0001, BF for both easy and hard but not for deliberate decisions, with the Bayes factor evidence for no effect (hard: p>0.5, BF = easy: BF = all corrected for multiple Notably, this of results was also for (Figure of the subjects had significant for arbitrary is, corrected for multiple against time for every over the last ms before response but only of the subjects had significant for the same for deliberate decisions; see Materials and In the average for deliberate and arbitrary decisions were and a significant BF = The the averaged amplitude above, and further demonstrates that the of baseline our results. This is because the of by of Figure Download asset Open asset of for individual subjects’ RPs for deliberate decisions (in and arbitrary ones (in pooled across decision We further tested differences in reaction time between the conditions, and subjects’ consistency scores might our We also tested the RPs might some stimulus-locked potentials or be to baseline in reaction times between conditions, stimulus-locked potentials and do not drive the effect RTs in deliberate decisions were typically more than as as RTs in arbitrary decisions. We to out the that the absence of RP in deliberate decisions from the difference in RTs between the conditions. We carried out for this we a the subjects two based on their and higher than the for deliberate and arbitrary trials, respectively. We then the same using only the subjects in the deliberate (M = s, SD = and the slower subjects in the arbitrary (M = s, SD = RP we would the RP to be more between these two though there were only the data a of results to those over the was (Figure to Figure Deliberate and arbitrary decisions were different with significant RPs found in arbitrary but not deliberate decisions. In the RPs for arbitrary decisions were not significantly different between the subjects with RTs and the for the easy or hard conditions Similarly, the RPs for deliberate decisions were not significantly different between the subjects with RTs and the for the easy or hard conditions This that RTs do not Cz for deliberate or arbitrary decisions in our results. Figure Download asset Open asset between RTs and RPs between subjects and and within subjects and (A) The subjects with RTs for arbitrary decisions (in and RTs for deliberate decisions (in show the same activity that was found in the main Figure (B) A of the difference between the RPs the difference between the RTs for deliberate and arbitrary decisions for each subject. The of the line is = x dashed red The is had an difference between deliberate and arbitrary decisions that was more than from the difference across all subjects. same subject’s difference was also more than higher than the across all subjects. subject was an and only from this For each subject we the RP using only the deliberate trials and slower arbitrary trials. The is the same as the found for the main We the same between the RP differences and the differences as in but this time the was within subjects. The of the line is = x The is we the difference between RPs in deliberate and arbitrary decisions over the last 500 ms before response against the difference between the RTs in these two conditions for each subject (Figure Again, if RP we would differences between RTs in deliberate and arbitrary conditions to correlate with differences between RPs in the two conditions. But no was found between the two = We further the RP differences on The not any relation between and RP differences = x the was at (as expected from the as the confidence the was not significantly different from While the results of the above suggested that our effects do not from differences between the RTs in deliberate and arbitrary decisions, the average RTs for deliberate subjects were ms slower than for arbitrary subjects. In we had only of the subjects in each to the the that some of our results might have been We also to look at the effect of within subjects and not ones. We a We the two decision and for each decision type and for statistical And then we took the deliberate trials and slower arbitrary trials for each subject this time we had subjects was and better results. Here, deliberate arbitrary trials (M = s, SD = were ms slower than arbitrary decisions (M = s, SD = on This the difference between deliberate and arbitrary by about from the We then the RPs for these deliberate and arbitrary trials within each subject (Figure the there is the same as the main (Figure What is more, deliberate and arbitrary decisions different trials were different from 0 while deliberate trials were not We further the differences between RPs in deliberate and arbitrary decisions as against the differences between the RTs for each subject to that such a would not for trials that are We found no relation between the two differences (Figure = x = Yet that could to the differences the conditions is that the RP in arbitrary might be some potential by the (i.e., the of the two specifically in arbitrary blocks, where the RTs are thus effects could the In particular, a potential might to some to the RP when locked to response onset. To test this we a analysis, the potentials in all conditions, locked to the onset of the stimulus (Figure We also the potentials across an expanded for (Figure the we see in Figures and 6B is to a stimulus-locked we would to see the before the four mean response onset times by lines at and 1.00, 2.13, and s for arbitrary arbitrary deliberate and deliberate in the stimulus-locked plot (Figure which precede the mean response that would further be of