Obeying orders reduces vicarious brain activation towards victims’ pain

Past historical events and experimental research have shown complying with the orders from an authority has a strong impact on people’s behaviour. However, the mechanisms underlying how obeying orders influences moral behaviours remain largely unknown. Here, we test the hypothesis that when male and female humans inflict a painful stimulation to another individual, their empathic response is reduced when this action complied with the order of an experimenter (coerced condition) in comparison with being free to decide to inflict that pain (free condition). We observed that even if participants knew that the shock intensity delivered to the ‘victim’ was exactly the same during coerced and free conditions, they rated the shocks as less painful in the coerced condition. MRI results further indicated that obeying orders reduced activity associated with witnessing the shocks to the victim in the ACC, insula/IFG, TPJ, the MTG and dorsal striatum (including the caudate and the putamen) as well as neural signatures of vicarious pain in comparison with being free to decide. We also observed that participants felt less responsible and showed reduced activity in a multivariate neural guilt signature in the coerced than in the free condition, suggesting that this reduction of neural response associated with empathy could be linked to a reduction of felt responsibility and guilt. These results highlight that obeying orders has a measurable influence on how people perceive and process others’ pain. This may help explain how people’s willingness to perform moral transgressions is altered in coerced situations.

Many examples in the history of Mankind show that when people obey the orders from an 3 authority, they are able to perform highly immoral acts towards others (e.g. Arendt, 1951, 4 1963, Herman & Chomsky, 1988. Even past experimental research, mainly by the work of 5 Stanley Milgram (1963Milgram ( , 1974, showed that many people comply with coerced orders to 6 inflict unbearable electric shock on a person for the sake of the experiment in which they were 1 8 participants rated their explicit sense of responsibility over the outcomes of their actions on an 1 9 analogue scale presented on the screen, ranging from 'not responsible at all' to 'fully 2 0 responsible'. Each delivered shock was rewarded with €0.05 to ensure that reward prediction 2 1 was similar in the two experimental conditions. In coerced blocks, participants were 2 2 instructed to deliver a shock on 50% of trials, see To increase the psychological effect of receiving orders, the experimenter was present in 2 5 the scanner room during the coerced condition. In order to justify that the agent was still able to hear clearly the experimenter giving orders even when she was in the noisy scanner room, a 1 fake microphone was used. The experimenter made it clear that she was present at the 2 beginning of the coerced condition by speaking with the agent, but then moved in the corner 3 of the room to avoid visual interference due to her presence. 4 At the end of the experimental session, participants were asked to fill in the Interpersonal 4 0 parameters computed during the segmentation were used to normalize the gray matter 4 1 segment (1mmx1mmx1mm) and the EPIs (2mmx2mmx2mm) to the MNI templates.

2
Afterwards, images were smoothed with a 6mm kernel.

3
At the first level, we defined separate regressors for Shock and NoShock trials, with window was taken for Shock and NoShock trials. Additional regressors included: (1) The 5 0 8 auditory orders from the experimenter (starting between 8-12s after the start of the trial) 1 together with the button presses (participants could press the key whenever they wanted right 2 after the auditory orders), (2) the pain rating scale (appearing on 6/30 trials randomly 1s after 3 the arrow pointing towards the video feedback disappeared) together with the responsibility 4 rating scale (appearing at the end of each MRI run, 1s after the arrow pointing towards the 5 video feedback disappeared or again 1s after the pain scale). Trials where participants 6 disobeyed were modelled in additional regressors of no interest separately for 'prosocial' 7 disobedience (i.e., they refused to administer a shock while having been ordered to send a 8 shock) and 'antisocial' disobedience (i.e., they administered a shock while having been 9 ordered not to send a shock). Finally, 6 additional regressors of no interest were included to 1 0 model head translations and rotations. Shock-NoShock rather than examining the shock condition alone in each condition to isolate 1 6 the effect of witnessing a shock from carry-over activity associated with pressing the response 1 7 button and seeing the arrow presented during the feedback period. To be noted, some 1 8 participants administered few shocks. Because the reliability with which brain activity in the 1 9 Shock condition can be estimated in the fMRI analysis depends on the number of trials 2 0 included, only including participants delivering a large number of shocks would be ideal.

1
However, for our results to be representative of the population, excluding too many 2 2 participants delivering small numbers of shocks would bias our results towards less 2 3 considerate participants. Given the distribution of shocks (i.e., {1,2,2,3,5,8,9,9,...}, see 2), we thus chose to only exclude participants that had delivered 1, 2, 3 or 5 shocks, and 2 5 retained all participants that had delivered 8 shocks or more. This decision was not based on a 2 6 power-analysis but by a subject cost-benefit analysis that pitched the benefits of inclusiveness 2 7 against the cost of less stable parameter estimates. The five excluded participants were thus 2 8 not included in any neuroimaging analyses. increasing order. The red line represents the cut-off to include (above the red line) or exclude (below or at the 2 level of the red line) those participants from our MRI analyses.

4
In order to test for the collinearity of the regressors, that is, ensuring that the brain 5 activations observed when participants witnessed a shock or no shock on the victim's hand 6 are not the consequence of previous brain activations coming from anterior events, we 7 calculated from the design matrix a Pearson correlation between our contrast of interest the shared. We then transformed the r value of each participant into a z score and conducted a one The same analysis on the rating scale as a regressor also showed no influence on the activations associated with the other regressors. corrected at the cluster level. To decode vicarious pain intensity, we additionally used the vicarious pain signature shock trials in the free condition to perform the analysis, and for each condition. The 3 2 vicarious pain values were then compared using a one-tailed t-test to test our one-tailed 3 3 hypothesis that vicarious pain should be higher in the free compared to the coerced condition.

4
We used the pattern without occipital lobe, as used in Krishnan et al., to reduce visual 3 5 confounds. We performed the same analysis with the neurological pain signature, NPS,

6
Wager et al., 2013) and with the guilt-related brain signature (Yu et al., 2020). To shed further light on the properties of the clusters resulting from the contrast only experimental condition in which participants could freely decide to deliver a shock or not 5 0 to the victim, thus allowing us to assess the association between brain activity and decision- 30/60 shocks, randomly. Two participants never followed the orders of the experimenter, one 2 3 by inflicting 60/60 shocks, one by inflicting 0/60 shocks. They were fully removed from any 2 4 further analyses. Three participants voluntarily disobeyed the orders to send a shock to the 2 5 'victim' on a few trials (respectively, 6 out of 30 shock trials, 5 out of 30 shock trials and 2 2 6 out of 30 shock trials). Those trials were removed from the analysis. In the free condition, 2 7 participants were told that they were entirely free to decide to deliver a shock or not to the between the number of shocks that participants received when they were victim first and the 3 8 number of shocks that they gave when they became agents further showed a tendency to adapt  shocks sent by the participants did not differ according to their role order (victim first, agent shocks or not to the 'victim' is only slightly influenced by the role order during the task. of the same ANOVA resulted in a BF incl =5.580, confirming that the data is over 5 times more CI95=382-532‰) (Fig. 4A), even if they knew that the shock intensity delivered to the 'victim' 1 5 was the same throughout the entire experiment. We also observed an interaction between the difference of pain ratings between the free and the coerced conditions was higher for those 1 9 who were victim first (79‰, CI95=23-135‰) than for those who were agent first (9‰, CI95=- as a covariate. We observed that the pattern of results remained unchanged (see 3 0 Supplementary Material S3), thus confirming that differences in the pain ratings cannot be Responsibility ratings. At the end of each of the four experimental runs, agents had to report vs Victim first) as between-subject factor on the responsibility ratings. The main effect of 4 0 Condition was significant, and Bayesian analyses strongly support the difference between the 4 1 1 2 Free and Coerced condition (F(1, 31)=59.696, p < .001, η 2 partial = .658; BFi ncl=2.048x10 8 ). Participants 1 reported a higher responsibility rating in the Free condition (86%, CI 95 =81-91%) than in the 2 coerced condition (51%, CI95=42-60%) (Fig. 4B). The main effect of Role-Order was not 3 significant (F (1,31) =0.104, p > .7) and the Bayesian analysis provides evidence for the absence of 4 a difference in the order the task was performed (BF incl =0.277). The interaction was not 5 significant (F(1, 31)=0.003, p > .9) and a Bayesian analysis provides evidence for the absence of 6 an interaction (BF incl =0.301). when they freely administered that shock than when they received the order to administer the same shock. B) 1 2 Participants reported that they felt more responsible in the free than in the coerced condition. C) Participants 1 3 reported that they felt worse administering a shock in exchange for money in the free condition than in the 1 4 coerced condition. D) Participants reported that they felt sorrier to administer a shock in the free condition than as between-subject factor on badness ratings was performed using both frequentist and 0.906) than in the coerced condition (-.448, CI 95 =-1.078 -.183) (Fig. 4C) the coerced condition (.025, CI 95 =(-).624 -.675) (Fig. 4D). The main effect of Role-Order The key hypothesis of the present study was to observe reduced activity in regions associated 5 with pain, including the insula and the ACC, while witnessing the pain of the victim in the 6 coerced condition in comparison with the free condition. To identify the brain regions that 7 respond more to the victim's pain when people are free to decide to inflict pain compared to were thus not taken into account in any neuroimaging results (N=5, 3 were agents first). After 1 3 excluding those participants, all remaining participants provided at least 8 shock trials, which 1 4 appears to be enough to reveal regions with reduced shock-triggered brain activation when 1 5 people obeyed orders compared to when they could freely decide which action to execute 1 6 (t=3.37, p < .001, cFWE, i.e. cluster size threshold determined by 5% family wise error=165).

7
These regions include the anterior cingulate cortex, putamen and caudate, the MTG, inferior (See Table 1 and Fig. 5). Decoding the unthresholded map using Neurosynth revealed that 2 1 the two functional terms most associated with this pattern of activation are pain and painful activity in brain regions associated with the processing of pain in the coerced condition in all the ROIs (all ps FDR > .1, all BFs 10 > .41 & < 1.200). We also found no effect of gender in Only clusters surviving a 5% FWE correction at the cluster size are reported (t=3.37, p < .001, cluster size 165).

1
Brain regions are referred to as by the nomenclature of the Anatomy Toolbox (Eickhoff et al., 2005). Columns 2 refer to the size in voxels of each cluster (voxels refer to the resampled 2x2x2mm voxels), the number of voxels 3 of that cluster within a specific cytoarchitectonic region, the percentage of voxels in that region, hemisphere, 4 cytoarchitectonic region (if available) or microanatomical region, percentage of cytoarchitectonic region falling 5 within the cluster, peak t-value within a particular region and MNI coordinates of the peak, respectively. ROI 1 6 was further referred to as 'ACC'. ROI 2 was further referred to as dorsal striatum because over 50% of it fell 7 1 6 within the caudate/putamen. ROI 3 was further referred to as 'TPJ'. ROI 4 was further referred to as 'insula'.

1
ROI 5 was further referred to as 'MTG'.

3
To further test our hypothesis that coercion leads to a reduced pain representation of the 4 victim compared to the free condition, and to overcome the limitations of voxel-wise reverse 5 inference, we used the vicarious pain signature identified by Krishnan et al., (2016). Krishnan 6 and co-workers presented participants with images that suggested three levels of pain in other 7 individuals, and identified a pattern of brain activity that predicts the level of pain (low, vicarious pain intensity to be lower in the coerced compared to the free condition, and 1 6 therefore used a one-tailed t-test to test this prediction. We found that indeed, the VPS for the with the neurosynth decoding in associating the pattern of activity with painfulness. Since we observed that participants reported that they felt less sorry and less bad for the 2 6 shocks sent to the 'victim' in the coerced condition compared to the free condition, we 2 7 further assessed whether coercion reduced neurocognitive processes associated with guilt. To 2 8 do so, we applied the same procedure to a multivariate brain pattern for interpersonal guilt Participants sent on average less shocks in the free condition than in the coerced condition, 3 4 which could thus have led to a higher repetition suppression to viewing shocks in the coerced 3 5 condition. We thus performed additional analyses examining whether we had observed 3 6 significant repetition suppression in our conditions. For that aim we compared brain 3 7 activations when participants witnessed a shock that was preceded by another shock or by no 3 8 shock, separately for the free and the coerced condition. This analysis did not show significant 3 9 results in either condition at p<0.001 (cluster FWE corrected), thus implying that repetition 4 0 suppression was not a significant factor in our analysis and that results are unlikely to be 4 1 explained by differential repetition suppression. should be related to the decision-making to shock or not to shock. Given our low 5 0 disobedience rate, we cannot test this notion in the coerced condition. Instead, we therefore 5 1 1 7 tested whether individual differences in decision-making in the free condition (quantified as 1 the number of shocks delivered to the 'victim') was related to individual differences in 2 activity in empathy related brain regions showing reduced activity following orders (i.e. Table   3 1 CoercedNoShock)] (which we will now call ROIs) and used this parameter estimates in a 6 correlation analysis with the number of shocks given.

8
We had prior hypotheses that based on findings relating brain activity in these regions to 9 empathy and empathy to prosociality, individuals that show higher activity when witnessing a participants freely delivered and activity in the brain regions that we identified in the 1 3 [FreeShock-FreeNoShock] contrast, again using both frequentist and Bayesian approaches.

4
To correct for multiple comparisons for frequentist statistics, we applied a False Discovery 1995) to each p-value with the frequentist approach. We performed separate analyses on those 1 7 who were agent first and those who were agent second, since the choice to send or not a shock 1 8 to the 'victim' for the later can be influenced by the number of shocks that they received as 1 9 victim (see Fig. 3). For those who were agent first, we observed that the number of shocks that they freely 2 2 decided to administer was correlated with the activity in the dorsal striatum (i.e., ROI 2) (r=- [FreeShock-FreeNoShock] correlates with the number of free shocks delivered, including 3 0 only agents first. Figure S1 displays the uncorrected results. For those who were agent second, we know that the number of shocks administered depends 3 3 on the number of shocks received. To test whether there was additional information in brain 3 4 activity, we subtracted the number of shocks received (as a victim) from the number of shocks 3 5 delivered in the free condition (i.e., shocks given -shocks received), and correlated this 3 6 excess with the brain activity in the regions of interest. There was no conclusive evidence that  (Fig. 6B). For ACC the BF -0 =0.285 suggests that data support the null hypothesis of a including only agents that were victim first. Figure S2 displays the uncorrected results. score on the IRI for the empathy related subscales PT, EC and PD. (see Table 2). Analyses 1 5 were performed using Bayesian Linear Regressions in JASP for the BF incl and a frequentist 1 6  .05) and BFincl > 3, two-tailed. Using the predictor starting with the order and ending with the button press, in an 1 6 exploratory analysis we also analysed whether there were differences in activity during the 1 7 decision-making phase of the experiment between Free and Coerced conditions. This chose to give or not to give shocks. At p<0.001 5%FWE cluster corrected, we observed a 2 0 cluster including the left superior medial and mid orbital gyri in the vmPFC that was more 2 1 active while taking a decision in the Free compared to following orders in the Coerced 2 2 condition (See Table S3 and Fig. S3). This finding matches the fact that participants had to We tested the hypothesis that obeying the orders received from an authority would reduce the 1 2 vicarious brain activation when witnessing the pain that one had delivered to a 'victim' witnessed a shock that they delivered after having received the order to do so, activity in brain analyses confirmed that reduced activations in the coerced condition were not related to a 2 5 repetition suppression phenomenon associated with viewing shocks more frequently in the 2 6 coerced condition than in the free condition. Participants also reported a reduced feeling of responsibility in the coerced condition, which the case of a pain that is fully caused by the participants own action, brain activity is altered observed were between two conditions ('observe' vs 'decide and execute') that not only 3 5 differed in terms of responsibility, but also in the sensorimotor information associated with 3 6 performing an action, in the causality between one's own actions and the outcome and in the errors. In our paradigm, participants were always the authors of the actions and these actions 4 0 were always fully predictable, thus isolating the impact of one's own perception of 4 1 responsibility on the empathic response. It may be argued that our results do not only reflect a reduced empathic response when people 4 4 obey orders but a general effect of lack of freedom to make a decision. Such conditions were 4 5 already introduced in a previous study (Lepron et al. 2015) in which participants also had to 4 6 perform a condition in which they had to execute an instruction sent by the computer.

7
Interestingly, the authors did not observe any differences in the neurophysiological empathic 4 8 response between that condition and a condition in which participants could decide and 4 9 execute the action. In our study, we did observe such differences at the neural level between 5 0 2 1 the free and the coerced condition, suggesting that obeying the orders of an authority has a 1 stronger influence on the empathic response towards others' pain than simply following the 2 instructions of a computer.

4
A study (Yu et al., 2014) showed that the activations in anterior middle cingulate cortex 5 (aMCC) and anterior insula (AI) were higher when participants felt guiltier and were the sole 6 responsible for the pain of others compared to when they bear less responsibility. In the 7 present study, participants explicitly reported that they felt less sorry and less bad in the 8 coerced condition compared to the free condition. The reduced activations observed in the 9 ACC and insula could thus also reflect a reduced perception of guilt under coercion. This 1 0 notion was confirmed by the fact that a multivoxel pattern that has been shown to decode 1 1 interpersonal guilt from neural activity (Yu et al., 2020) was also reduced in our Coerced vs. time perception (Humphrey & Buehner, 2009). In order to separate brain activity related to 2 5 motor response from those related to processing the pain of the victim, comparatively long 2 6 jittered action-outcome delay had to be used in our fMRI paradigm, precluding the use of conditions in which brain activity is measured from individuals giving orders, which gives a