Sense of Agency for Mental Actions: Insights from a Belief-Based Action-Effect Paradigm

A substantial body of research in the past few decades has converged on the idea that the so-called “sense of agency”, the feeling of being in control of our own actions, arises from the integration of multiple sources of information at different levels. In this study, we investigated whether a measurable sense of agency can be detected for mental actions, without the contribution of motor components. We used a fake action-effect paradigm, where participants were led to think that a motor action or a particular thought could trigger a sound. Results showed that the high-level sense of agency, measured through explicit reports, was of comparable strength for motor and mental actions. The ‘intentional binding’ effect, a phenomenon typically associated with the experience of agency, was also observed for both motor and mental actions, with the only exception of short action-effect delays. Furthermore, a consistent relationship between explicit reports of agency and intentional binding was found. Taken together, our results provide novel insights into the specific role of intentional cues in instantiating a sense of agency, even in the absence of motor signals. These results may have important implications for future brain-computer interfaces as well as for the study of pathological disruptions of agency.


Introduction
experiment that most participants feel that their actions can produce the 150 alleged effect. In this sense, we expect more participants to have causal beliefs 151 for the motor condition than for the mental condition, given the increased 152 difficulty of creating causal beliefs for mental actions (see "Methods"). 153

154
It is worth mentioning that the experiment was initially thought to be run 155 in a laboratory environment. However, due to the SARS-CoV-2 pandemic, 156 the experiment was adapted to be performed online. More details will be 157 given in the following paragraphs.  For the experiment to be successful, it was crucial to ensure that par-177 ticipants believed that their mental actions could trigger an external event 178 through the apparent BCI. Participants were told that they would partici-179 pate in an experiment to evaluate the users' interaction with a facial motion 180 detection software. They were told that this software could detect micro-   the same (2.55 seconds after dot motion onset), so that the sound could be programmed to occur at varying intervals from the moment in which 220 participants performed the action (if following the instructions properly).

221
The red cross appeared at a different location each time, chosen randomly 222 among the sixty 'minute' positions of the clock, so that participants could 223 not habituate to the clock position where the key press would be made.

224
Participants could nevertheless habituate to the 2.55 seconds period, and 225 solve the task by learning the period and not looking at the dot. However, 226 this possibility was deemed to be fairly remote given the difficulty of the task, 227 and it was assumed that participants would need to track the dot rotation 228 to perform the task. The trial structure is shown in Figure 1.

229
In every trial, the sound always appeared, regardless of when and whether 230 the 'space' key was pressed. The sound appeared at one of three different 231 intervals from the cued time: 200, 400 or 600 ms. Each interval appeared 232 in total of four trials, and the order of appearance of each interval was ran-233 domized. Before the trials, participants were told that their key press would 234 trigger a sound, but that in some of the trials the sound would be produced 235 by the computer instead.

236
In the first part of this block, the JoA was assessed with explicit measures.

237
Following each trial, participants were asked the question: "Who produced 238 the tone?". They could choose between "Me" and "The computer". After 239 this response, they had to respond using a Likert 1-7 scale to the following 240 question: "How confident are you about your answer?". The scale had in its 241 extremes the labels "Not at all" and "Very much". In the second part of the 242 motor action condition, the FoA was evaluated. Participants had to indicate 243 after each trial the clock position corresponding to the instant where they 244 heard the sound, which allowed us to measure the intentional binding over 245 the sound.

246
An important difference with other studies using similar paradigms to 247 measure intentional binding is that in our study participants were instructed 248 at which specific moment the mental or motor action had to be performed.

249
The reason for this was that the experimental design required the subjects' 250 actions not to be causally linked with the sound. By instructing the par-251 ticipants to perform the action at a specific moment, the sound could be 252 programmed to occur after that time (thus, after the action), thereby creat-253 ing an illusion of causality. In brief, we assumed that if participants pressed 254 the key at a given time and the sound appeared right after their action, they 255 would feel that they produced the sound, even if no actual causal relation 256 Figure 1: Trial structure. The dot started rotating from a random position of the clock (one out of the sixty 'minute' positions). After 500 ms, a red cross appeared on screen, and remained for 500 ms. The red cross appears magnified for visualization purposes. The red cross appeared at the position from where the dot had started rotating, so that the dot reached the cross position always at the same time, 2.55 ms (the dot rotation period). Participants were instructed to perform an action when the dot reached the position cued by the cross: a motor action in the first experimental block, a mental action in the second one, and no action in the third. A sound appeared after a given time interval (200, 400 or 600 ms, in randomized order) after the dot reached the cued position. After the trial, participants had to either report their judgement of agency or to judge the time when they heard the sound. It is worth noting that the stimulus presented was identical in all trials of the three blocks of the experiment, with the only difference that in each block participants were instructed to perform different actions at the cued time. existed between both events and even if they were "forced" to perform the action at an indicated time.

258
In our experiment, only tone binding was measured, due to the fact that     • "In the first block of the experiment (press key -sound), do you think 306 that in some trials the sound was produced by your key press?"

307
• "In the second block of the experiment (think 'SMILE' -sound), do you 308 think that in some trials the sound was produced by your thought?"

309
The possible answers where "Yes" and "No". These questions are crucial 310 for the study because they provide information about whether participants 311 had believed or not that they could eventually produce the sound with their 312 motor or mental actions. If a participant replied 'Yes' to the first or second 313 question, this means that they thought that at least in one trial they produced 314 the sound; therefore, the participant believed the "story" about the facial 315 motion detection software, or about their key press producing the sound. It 316 is critical to separate participants who had causal beliefs from those who did 317 not because a strong sense of agency was expected only in those subjects     To verify the fact that the sense of agency would not be produced (or 348 in a very reduced manner) for participants without causal beliefs, we also 349 analyzed the results for the ten participants who reported believing that 350 neither their motor actions nor their mental actions could produce the tone.

351
These results are summarized in Section 3.5. (STD = 122 ms), which means that in general participants pressed the space 358 bar after the cued time but not much later, and that this pattern was con-359 sistent across participants. No key press was detected in 13 trials (out of the 360 763 total trials in the motor action condition). Those trials were removed.

361
Trials with outlier key press times (±3 STD of the group mean) were also 362 removed (11 trials, less than 2% of the total).

363
In the mental action task, behavioral performance could not be measured, 364 as it was not possible to know, by its own nature, the time when participants 365 thought about the word 'smile'. Therefore, we had to expect that participants  (motor action versus mental action) and also by the action-effect delay. 376 We used the data of all trials where an explicit attribution of agency was 377 reported. In 10 trials no report was given, and in two trials participants 378 reported not having heard the sound. In total, data from 543 trials was 379 available (approximately four trials per participant in each condition and 380 interval). 381 We ran a binomial logistic regression where the dependent variable was 382 authorship attribution ("Me"/"The computer"), and the main predictors  self-attributions was significantly larger for short intervals than for medium 389 intervals (χ 2 = 83.7, dof = 1, p < 0.0001), and was also significantly larger 390 for medium intervals than for long intervals (χ 2 = 24.2, dof = 1, p < 0.0001).

391
All pairwise comparisons were Bonferroni-corrected for multiple comparisons.

392
To further assess whether there was a difference in the authorship attribution 393 for the motor and mental conditions, we ran separate Chi-Square tests for Figure 2: Self-attribution decreases for increased action-effect intervals but is independent of whether the action is mental or an overt motor movement. (a) Proportion of selfattributions for each condition, groups divided by interval. The statistical significance in the comparison between groups is shown for illustrative purposes (the statistical tests have not been carried out on the proportion of self-attributions but on the number of attributions). A significant effect of time interval was found, shorter intervals leading to a higher number of self-attributions. No effect of condition and no interaction were found. * p < 0.0001; 'n.s.': non-significant (p > 0.017); Chi-Square tests, Bonferroni corrected. (b) Mean confidence rating for the attribution of agency divided by time interval, reported in a 1-7 Likert-scale. As expected, the confidence for medium intervals is significantly lower than for short and long intervals. Error bars represent 95% confidence intervals.   participants are more likely to think that they caused the sound for short 423 intervals, and that the computer did so for long intervals, but they are less All of the analyses carried out in this section were between participants.

428
The reason for that is that we had a different number of subjects for the 429 action and the motor condition (see Section 3.1).    538 Figure 4: Comparison of the sound time estimation error for the three experimental conditions, data divided by time interval. For short intervals, significant binding (difference with baseline error) was found only for motor actions. For medium intervals, the binding over the sound was significant for both motor and mental actions, in line with the JoA results. For long intervals, the binding was not significant for any of these two conditions, also in consonance with the explicit reports. Bars represent 95% confidence intervals. * p < 0.01, 'n.s.': non-significant, paired samples t-test.
3.5. "Non-believers": participants who reported no causal beliefs displayed a 539 reduced sense of agency. 540 We analyzed the responses of participants who answered "No" to the 541 two questions of the post-experimental questionnaire regarding their belief 542 of causality in the motor and mental action conditions (10 subjects). We 543 assume that those participants did not believe that their actions triggered 544 the alleged effect, and that either this belief was present from the beginning 545 of the experiment or that it appeared during the experimental tasks. 546 We assessed whether the explicit reports of agency in the motor and men-547 tal conditions were different for "non-believers" compared to participants who 548 reported causal beliefs. We ran Chi-square tests to compare the number of 549 self-attributions in each condition and each time interval for both groups. 550 We found that, for short intervals, the number of self-attributions was sig-551 nificantly larger for believers than for non-believers, and that this was the 552 case for the motor action condition (χ 2 = 12.6, dof = 1, p < 0.001) and a "control" group, proving that intentional binding is not the mere result of time contiguity, but that at least the belief of causality has to be present.

591
The experimental design was based on the assumption that a sense of 592 agency would be created if prior causal beliefs were sufficiently established.

593
The results of the experiment confirm this assumption: more than two thirds 594 of the total number of participants claimed to be in control of the sound in 595 the motor action block, and more than one third of the subjects thought that 596 their mental action could trigger the sound in the mental action block. The 597 fact that a sense of agency emerged for those participants purely from the 598 manipulation of prior causal beliefs is a critical success of this experiment.

599
This was further confirmed by analyzing the results for subjects who did not 600 have causal beliefs: the sense of agency was reduced or even absent for these 601 participants. results are also consistent with the hypothesis put forward by Buehner [28] 641 that the IB effect could reflect pure causality and not intentionality. In any 642 case, both accounts agree in the fact that motor cues are not exclusively 643 required for IB to occur. Second, we conclude that mental actions give rise 644 to a sense of agency not only at a higher conceptual level, but also at a lower perceptual level. This, we claim, is a novel finding in the research on agency.
the FoA under non-motor actions. 648 We contemplate that the construction of the FoA from mental actions for 200 ms intervals were higher than for medium intervals. It is widely ac-    and short delays in this experiment, and as also seen in other studies [19].

795
Another interesting implication of our study concerns how legal attribu-796 tions of moral responsibility may have to be rethought in the presence of 797 future BCI technologies. Several authors have already raised issues concern-798 ing moral responsibility over BCI-mediated actions: users might be uncertain 799 about their agency in a BCI context, making difficult the attribution of moral 800 (and therefore, legal) responsibility [47,48]. Moreover, in most legal systems 801 this attribution of responsibility is based on the fact that only "voluntary 802 acts" are punishable, an "act or action" being defined as a "bodily move-