Abstract
Socially engaging robots have been increasingly applied to alleviate depressive symptoms and to improve the quality of social life among different populations. Seeing that depression negatively influences social reward processing in everyday interaction, we investigate this influence during simulated interactions with humans or robots. In this study, 35 participants with mild depression and 35 controls (all from nonclinical populations) finished the social incentive delay task with event-related potential recording, in which they received performance feedback from other persons or from a robot. Compared to the controls, the mild depressive symptom (MDS) group represented abnormalities of social reward processing in the human feedback condition: first, the MDS group showed a lower hit rate and a smaller contingent-negative variation (correlated with each other) during reward anticipation; second, depression level modulated both the early phase (indexed by the feedback-related negativity (FRN)) and the late phase (indexed by the P3) of reward consumption. In contrast, the effect of depression was evident only on FRN amplitude in the robot feedback condition. We suggest that compared to human–human interaction, the rewarding properties of human–robot interaction are less likely to be affected by depression. These findings have implications for the utilization of robot-assisted intervention in clinical practice.
Similar content being viewed by others
Introduction
Social interaction is essential to people’s livelihood, career development, mental health, and psychological well-being. In human beings, one of the major motivators for social interaction is to pursue social rewards [1, 2]. Here, “social reward” refers to social information that has rewarding properties, including social approval, social belonging, and social support [3]. Depression is closely associated with the abnormal processing of social rewards at both behavioral and neural levels [4]. Severe expression of depression symptoms and syndromes has been linked to blunted response to social rewards (i.e., social anhedonia) [5,6,7]. Depression severity negatively correlates with the activation level of the human reward system, including dopaminergic neural circuits [8,9,10,11]. Consequently, these problems in social reward processing inhibit depressed individuals’ motivation to engage in social interaction [12], which could help understand the impairments of social functioning in depression [13].
Nowadays, human-like robots have been prevailing in daily life, many of which are specifically designed to establish social relations with humans (i.e., artificially intelligent social machines) [14]. These socially engaging artificial agents have been increasingly employed for psychosocial intervention on depression [15]. Using social robots for clinical practice has important implications, seeing that some populations (e.g., older adults and disabled persons) need to maintain a certain degree of social interaction and engagement, but medical resources for providing social support are often limited [16]. Some recent studies have shown that interacting with social robots has promising effects on geriatric depression and counteracting feelings of loneliness [17,18,19,20,21]. Considering these new developments, a question that emerges is whether social reward dysfunctions in depression could influence not only human–human interaction but also human–robot interaction, which is the main interest of this study. Investigating this topic would help determine: (1) the value of a robot-assisted intervention as a means of nonpharmacological treatment, and (2) the nature of social interaction from a theoretical point of view [14].
Interacting with an artificial entity is like interacting with fellow humans in many aspects. First, people are willing to follow their usual habits when communicating with human-like agents, using social skills, such as gestures and facial signals [22,23,24]. Second, people are prone to apply the same social rules and expectations to automations as they do to humans, indicating that social psychological knowledge derived from human–human interaction also guides human–machine interaction [25,26,27,28]. For example, people can “recognize” personality cues from computer-synthesized speech [29]. Likewise, De Kleijn et al. found that participants in the classic Ultimatum Game allocate an equal amount of resources to human and robot opponents [30,31,32,33]. Human faces and humanoid faces are equally likely to trigger an automatic orientation of attention [34], and they both activate face-responsive brain regions including the fusiform gyrus and superior temporal sulcus as well as the mirror neuron system [35]. These phenomena could be regarded as manifestations of anthropomorphism, that is, the tendency to attribute human characteristics to non-human entities—especially those having human-like appearances [36, 37]. Therefore, it is understandable that communications with artificial agents could be perceived to have social rewarding properties [38, 39]. For instance, “friendly” computer feedback is evaluated as more favorable than “unfriendly” feedback by individuals [25]. Also, happy compared to neutral expressions of both human and robot faces shortened individual reaction time (RT) and enhanced the P1 component of event-related potentials (ERPs) [40,41,42]. People may even develop a long-term “parasocial” relationship with their companion robots (e.g., at health service facilities) [43,44,45,46].
To our knowledge, no study to date has directly addressed whether depression would impair the social rewarding nature of human–robot interaction. However, this possibility is indicated by the research focusing on other types of mental disorders: for instance, Raffard et al. discovered that negative symptoms of schizophrenia negatively affect the accuracy of facial expression discrimination in both human and robotic conditions [38, 47]. In this study, we used the classic social incentive delay (SID) task designed by Spreckelmeyer et al. to investigate social reward processing [48, 49]. In each trial of the SID task, participants observe an incentive cue indicating the amount of potential reward, then respond to a target stimulus as quickly as possible, and finally receive social feedback (e.g., a smiling face or a “thumbs-up” gesture) as rewards according to their performance. Here, a major advantage of the SID task is that it enables to discriminate between the anticipation stage (corresponding to cue presentation) and the consumption stage (corresponding to feedback presentation) of social reward processing [50, 51]. Using this paradigm, recent studies have revealed that depressive symptoms are associated with deficits in both stages [52,53,54,55].
In order to observe not only behavioral but also neural manifestations of reward processing abnormalities, this study relies on the ERPs that are suitable to reflect the dynamics of brain activity over the time course of reward processing [56,57,58,59]. Combining the SID task with the ERP technique has been proven to be successful [60,61,62]. According to our research interest, we focus on three ERP components, including the contingent-negative variation (CNV) associated with cue presentation, and the feedback-related negativity (FRN) and P3 associated with feedback presentation. The CNV is a frontocentral distributed negativity that is elicited by a preceding signal, indicating preparation processes for an upcoming event [63, 64]. In the SID context, the CNV elicited by cue presentation indicates the anticipation of the reward-related target [61]. The FRN is another frontocentral distributed negativity in response to feedback stimulus, being larger for unfavorable than favorable feedback [65,66,67]. It is possible that the FRN indicates a quick, coarse, and bottom-up detection of feedback value [68,69,70]. Following the FRN, the P3 is also an important feedback-locked component that is positive-going and centro-parietal distributed [71, 72]. This component is suggested to reflect a more deliberate, comprehensive, top-down evaluation of feedback information [73,74,75]. According to previous studies, the amplitudes of these ERP components are all sensitive to individual depression level [10, 11, 76, 77]. Most relevantly, one of our recent studies using the SID task has found that the mild depressive participants showed a lower hit rate and a smaller CNV compared to the controls [53]. Taking a step further, this study examines whether these indexes would be sensitive to the depression level when individuals are interacting with a robot. Seeing that limited ERP studies have directly compared human–human and human–robot interactions, we did not formulate a priori hypotheses.
Materials and methods
Participants
Before the experiment, we conducted a priori power analysis using G*Power 3.1.7 [78, 79]. According to the effect size (\({\eta}_{p}^{2}\) ≥ 0.051 for interaction effects) reported in one of our recent studies using a similar experimental design [53], the statistical power would be higher than 95% when there are 35 participants per group. A total of 70 adult participants were recruited from a large sample pool (approximately 500) in Shenzhen University. Participants completed the Beck Depression Inventory Second Edition (BDI-II) [80] during the recruitment stage. Based on BDI-II scores, two groups of individuals were invited, namely, a mild depressive symptom (MDS) group and a control group. The MDS group consisted of participants who scored larger than 13 (indicating mild depression), while the control group scored <13. The key exclusion criterion was any lifetime Axis I disorders according to Structured Clinical Interview for DSM-IV-TR Axis I Disorders, Research Version, Non-Patient Edition (SCID-I/NP) [81]. Other exclusion criteria included: (1) seizure disorder; (2) a history of head injury with possible neurological sequelae; (3) self-reported prior use of any psychoactive drugs; and (4) current alcohol or drug dependence. Among the students who met the above criteria, 70 individuals (35 with mild depression and 35 controls) were invited to participate in the experiment. According to their self-reports, all participants were medical-free at the time of the experiment.
On the day of the experiment, each participant was required to complete several questionnaires as soon as they come to the lab: (1) BDI-II; (2) the Trait form of Spielberger’s State-Trait Anxiety Inventory (STAI-T) [82]; (3) the Revised Social Anhedonia Scale (RSAS) [83]; (4) the Liebowitz’s Social Anxiety Scale (LSAS) [84]; (5) Rosenberg Self-esteem Scale (RSS) [85]; and (6) the “hedonic attitudes” subscale of Robot Acceptance Scale (RAS) [86], which measures how much a participant likes robots. As shown in Table 1, no significant difference was found between these two groups with respect to gender, age, handedness, and hedonic attitudes to robots, while the MDS group scored higher than the controls in depression, trait anxiety, social anhedonia, and social anxiety, but lower in self-esteem. Written informed consent was obtained prior to the experiment. The study was approved by the Ethics Committee of Shenzhen University.
Experimental design and stimuli
A life-sized humanoid robot Karl was used in this study (version 1088, Guangzhou, China). Karl could respond to simple oral messages with relevant body gestures and screen-based facial expressions. For example, if somebody tells Karl: “you are so cool,” it would reply: “your praise is making me shy,” move hands up, and show a facial expression of embarrassment. Screen-based robot faces have several advantages compared to traditional mechanical faces, such as a wider variety of facial emotions and less cost of construction [87].
This experiment used 36 human and 36 robot facial expression pictures, with each cue condition (reward/punishment/neutral) having 12 of each kind of picture. Human facial pictures were taken from two undergraduate students (one male and one female) from the first author DZ’s lab. Robot facial pictures were taken from the robot Karl. Thirty volunteers (15 males; aged 22.3 ± 1.9 years; BDI scored 4.0 ± 3.8) assessed the emotional valence and arousal of these pictures on 9-point scales (Table 2); none of them participated in the formal experiment. Repeated-measures analyses of variances (ANOVAs) were performed on the valence (from 1: very negative to 9: very positive) and arousal (from 1: very low to 9: very high) ratings, respectively, with feedback provider (human vs. robot) and cue valence (reward vs. punishment vs. neutral) as two within-subject factors.
The results show that the main effect of cue valence was significant on valence rating (F(2,22) = 561.4, P < 0.001, \({\eta}_{p}^{2}=0.981\); reward [6.5 ± 0.3] > neutral [4.6 ± 0.2] > punishment [3.4 ± 0.4], pairwise Ps ≤ 0.001) and arousal rating (F(2,22) = 284.5, P < 0.001, \({\eta}_{p}^{2}\) = 0.963; reward [5.7 ± 0.3] > neutral [4.1 ± 0.2]/punishment [4.2 ± 0.3], pairwise Ps ≤ 0.001). Neither the main effect of feedback provider (F ≤ 1.6, P ≥ 0.229) nor the interaction effect (F ≤ 2.2, P ≥ 0.135) was significant, indicating that human and robot facial pictures provide comparable emotional experiences.
Experimental procedure
Prior to the task, participants were invited to interact with the robot Karl, as well as the two undergraduate students whose facial pictures were used as experimental materials, for 5 min. Then their individual threshold of visual RT was assessed using a simple RT task (averaged across 30 trials), which was used to determine the target presentation time in the formal task (see below). After that, participants practiced in the human feedback condition and the robot feedback condition for ten trials, respectively. The practice trials and the formal task shared the same temporal structure. The only difference was that we used video clips in practice trials to enhance the experience of social interaction; in contrast, stationary pictures were employed in the formal task. This was because the ERP technique has a time resolution of milliseconds (much shorter than the presentation of video clips); accordingly, stationary pictures rather than video clips are more suitable for ERP research [88].
The formal SID task consisted of two human feedback and two robot feedback blocks. Each block consisted of 36 rewards, 36 neutral, and 36 punishment-cued trials, the order of which was pseudorandomly determined. The formal task (432 trials in total) lasted for ~40 min. The sequence of these two kinds of blocks was counterbalanced across participants. An overview of the trial structure is shown in Fig. 1. In both human feedback and robot feedback blocks, participants first saw a cue (i.e., the anticipation stage) indicating a potential reward (an upward arrow), punishment (a downward arrow), or neither (neutral condition: a horizontal arrow) for 500 ms. This cue was followed by a delay for a varied duration ranging from 2 to 4 s. The target stimulus then appeared, and participants were required to press the space button on a keyboard as quickly as possible to gain reward or avoid punishment. If participants’ RT was shorter than the duration of the target presentation, the ongoing trial would be labeled as a “hit” trial, otherwise, it would be labeled a “miss” trial. The target presentation time was initially set according to each participant’s behavioral performance in the simple RT task, then it was adjusted (±10 ms) in a trial-by-trial manner to keep the hit rate at ~50% [53, 89].
After the target presentation, each participant received outcome feedback for 1 s (i.e., the consumption stage). In the reward condition of human feedback blocks, successful hitting the target would lead to a picture of a smiling human face (indicating positive social evaluation), while missing the target would lead to a neutral human face; in the neutral condition, the feedback was always a neutral human face regardless of hit or miss; finally, in the punishment condition, successful hitting the target would lead to a picture of the neutral human face, while missing the target would lead to a human face with contempt (indicating negative social evaluation). These settings were consistent with the classic monetary incentive delay (MID) task, in which missing the target in the reward condition leads to no punishment, while hitting the target in the punishment condition leads to no reward [90, 91]. The same settings were used in robot feedback blocks, except that pictures of human faces were replaced by pictures of robot faces.
Electroencephalographic (EEG) recording and analysis
Brain electrical activity was recorded by a 32-channel wireless amplifier with a sampling frequency of 250 Hz (NeuSen.W32, Neuracle, Changzhou, China). Data were online recorded referentially against the left mastoid and off-line re-referenced to average activities over the scalp. EEG data were collected with electrode impedances kept below 10 kΩ. Ocular artifacts were removed from EEGs using a regression procedure implemented in NeuroScan software (Scan 4.3: NeuroScan, Inc., Herndon, VA).
This study focused on the anticipation stage and the consumption stage of reward processing, corresponding to the ERPs evoked by cues and those evoked by feedback. The recorded EEG data were filtered (half-amplitude cutoff: 0.1–30 Hz) and segmented beginning 200 ms prior to stimulus (cue/feedback) onset. The cue-evoked ERP epochs lasted from −200 ms to 2.5 s while the feedback-evoked epochs lasted from −200 to 800 ms. All epochs were baseline-corrected with respect to the mean voltage over the 200 ms preceding stimulus onset, followed by averaging for each experimental condition. Epochs containing artifacts with peak-to-peak deflection exceeding ±100 μV were rejected. This procedure deleted 4.8 ± 2.2 trials per condition. The numbers of remaining trials for further analyses are reported in supplementary materials (Supplementary Table S1).
The electrode sites and time window for each ERP component were selected before data analysis based on prior knowledge. The main reference was one of our previous ERP studies which also used the SID task [53]. For cue-evoked ERPs, this study focused on the CNV elicited by reward, neutral, and punishment cues in human feedback and robot feedback blocks. The mean amplitude of the CNV was calculated using the arithmetic average of the electrode sites in the mid-frontocentral area (including Fz, FCz, FC1, and FC2) within a time window of 750–2500 ms post cue onset [92]. For feedback-evoked ERPs, we focused on the FRN and P3 elicited by feedback for hits or misses in human feedback and robot feedback blocks. The mean amplitude of the FRN was calculated using the arithmetic average, also at electrode sites in the mid-frontocentral area (i.e., Fz, FCz, FC1, and FC2), within a time window of 200 to 300 ms post feedback onset [93]. Finally, the mean amplitude of the P3 was calculated using the arithmetic average at electrode sites in the mid-parietal area (i.e., Pz, P3, and P4) within a time window of 300–600 ms post feedback onset [94, 95].
Statistics
Repeated-measures ANOVAs were used to analyze the behavioral and ERP measures. Regarding the anticipation stage (i.e., cue presentation), a 2 × 3 × 2 mixed ANOVA was applied with feedback provider (human vs. robot) and cue valence (reward vs. punishment vs. neutral) as two within-subject factors and group (MDS vs. control) as the between-subject factor. The same model was applied to the consumption stage (i.e., feedback presentation), except that feedback valence (hit vs. miss) was added as the third within-subject factor (that is, a 2 × 3 × 2 × 2 mixed ANOVA). In addition, considering that anxiety and depressive symptoms are highly comorbid [96,97,98], we also performed ANOVAs on all the above dependent variables with trait anxiety as a covariate (see Supplementary Materials).
Descriptive data are presented as mean ± standard deviations, unless otherwise specified. The significance level is set at P = 0.05. Significant interactions were analyzed using a simple effects model. Partial eta-squared (\({\eta}_{p}^{2}\)) values are provided to demonstrate effect size where appropriate. For the sake of brevity, below we report significant interactions only when the group factor is involved. Other significant interactions could be found in supplementary materials.
Results
Behavioral results
Hit rate
The main effect of the group was significant (F(1,68) = 25.3, P < 0.001, \({\eta}_{p}^{2}\) = 0.271; control vs. MDS = 50.2 ± 7.0% vs. 48.7 ± 7.7%). The main effect of cue valence was also significant (F(2,136) = 55.1, P < 0.001, \({\eta}_{p}^{2}\) = 0.448; reward (53.7 ± 4.7%) >punishment (50.6 ± 5.3%) >neutral (44.1 ± 8.0%), pairwise Ps < 0.001). Moreover, the interaction of feedback provider × group was significant (F(1,68) = 38.2, P < 0.001, \({\eta}_{p}^{2}\) = 0.360; Fig. 2A); simple effect analysis reveals that the control group showed a higher hit rate (51.2 ± 6.9%) than the MDS group (47.9 ± 7.6%) in the human condition (F(1,68) = 57.3, P < 0.001, \({\eta}_{p}^{2}\) = 0.457), but not in the robot condition (F < 1; control vs. MDS = 49.2 ± 6.8% vs. 49.5 ± 7.7%).
RT
The RT was averaged across hit trials in each condition [53, 54]. The main effect of cue valence was significant (F(2,136) = 12.4, P < 0.001, \({\eta}_{p}^{2}\) = 0.154; rewards (216.4 ± 19.1 ms) < punishment (218.0 ± 20.5 ms) = neutral (219.3 ± 21.0 ms), pairwise Ps ≤ 0.002).
ERP data
Cue-evoked CNV
The main effect of cue valence was significant (F(2,136) = 8.6, P = 0.001, \({\eta}_{p}^{2}\) = 0.112; reward (−2.21 ± 3.54 μV) = punishment (−2.12 ± 3.50 μV) > neutral (−1.38 ± 3.78 μV), pairwise Ps ≤ 0.007). Moreover, the two-way interaction of feedback provider × group was significant (F(1,68) = 37.6, P < 0.001, \({\eta}_{p}^{2}\) = 0.356; Fig. 2B, C); simple effect analysis reveals that the control group (−3.14 ± 4.77 μV) exhibited a larger (i.e., more negative-going) CNV than the MDS group (−0.74 ± 5.87 μV) in the human condition (F(1,68) = 8.8, P = 0.004, \({\eta}_{p}^{2}\) = 0.114), but not in the robot condition (F(1,68) = 1.4, P = 0.235, \({\eta}_{p}^{2}\) = 0.021; control vs. MDS = −1.46 ± 3.85 vs. −2.29 ± 5.96 μV).
Seeing that the hit rate and the CNV amplitude showed a similar pattern of results (i.e., a significant feedback provider × group interaction), we tested the two-tailed Pearson correlation between the above two indexes. This follow-up analysis showed that the hit rate and CNV amplitude were negatively correlated in both feedback conditions (r = −0.245 ~ −0.474, P = 0.041–0.0004, corrected for multiple comparisons using the false discovery rate method; Table 3). Again, please note that the CNV is a negative-going component; thus, these negative correlations indicate that the CNV amplitude increased as a function of the hit rate.
Feedback-evoked FRN
The main effects of feedback provider (F(1,68) = 7.8, P = 0.007, \({\eta}_{p}^{2}\) = 0.103; human vs. robot = 0.36 ± 3.63 vs. 0.86 ± 3.46 μV), cue valence (F(2,136) = 14.3, P < 0.001, \({\eta}_{p}^{2}\) = 0.174; reward (0.94 ± 3.73 μV) = punishment (0.76 ± 3.35 μV) < neutral (0.14 ± 3.52 μV), pairwise Ps ≤ 0.002), and feedback valence were significant (F(1,68) = 74.7, P < 0.001, \({\eta}_{p}^{2}\) = 0.524; hit vs. miss = 1.00 ± 3.57 vs. 0.22 ± 3.49 μV).
The interaction of feedback valence × group was significant (F(1,68) = 6.8, P = 0.011, \({\eta}_{p}^{2}\) = 0.091). Further, the three-way interaction of cue valence × feedback valence × group was also significant (F(2,136) = 9.5, P < 0.001, \({\eta}_{p}^{2}\) = 0.123; Fig. 3B). To break down this three-way interaction, we examined the feedback valence × group interaction for reward, punishment, and neutral cues, respectively (Fig. 3A). Results reveal a similar pattern of the feedback valence × group interaction for reward (F(1,68) = 10.4, P = 0.002, \({\eta}_{p}^{2}\) = 0.133) and neutral cues (F(1,68) = 11.2, P = 0.001, \({\eta}_{p}^{2}\) = 0.142): the MDS group showed a smaller (i.e., less negative-going) FRN than the controls in response to miss feedback (reward: F(1,68) = 4.1, P = 0.046, \({\eta}_{p}^{2}\) = 0.057, control vs. MDS = −0.59 ± 3.33 vs. 1.08 ± 3.55 μV; neutral: F(1,68) = 4.8, P = 0.032, \({\eta}_{p}^{2}\) = 0.066, control vs. MDS = −0.54 ± 3.57 vs. 1.22 ± 3.11 μV), but not hit feedback (F < 1; reward: 1.45 ± 3.58 vs. 1.90 ± 3.59 μV; neutral: −0.36 ± 3.26 vs. 0.36 ± 3.25 μV). In contrast, the feedback valence × group interaction was not significant for punishment cues; specifically, the two groups showed comparable FRN amplitudes in response to hit (F(1,68) = 1.8, P = 0.184, \({\eta}_{p}^{2}\) = 0.026; control = 0.84 ± 3.38 μV, MDS = 1.85 ± 2.90 μV) and miss feedback (F < 1; control = 0.05 ± 3.23 μV, MDS = 0.29 ± 3.17 μV).
Feedback-evoked P3
The main effects of all the four factors were significant, including group (marginally, F(1,68) = 3.9, P = 0.052, \({\eta}_{p}^{2}\) = 0.055; control vs. MDS = 1.55 ± 3.66 vs. 0.19 ± 3.57 μV), feedback provider (F(1,68) = 6.9, P = 0.011, \({\eta}_{p}^{2}\) = 0.092; human vs. robot = 1.16 ± 3.69 vs. 0.58 ± 3.64 μV), cue valence (F(2,136) = 52.7, P < 0.001, \({\eta}_{p}^{2}\) = 0.437; reward (1.44 ± 3.62 μV) = punishment (1.50 ± 3.44 μV) > neutral (−0.33 ± 3.66 μV), pairwise Ps < 0.001), and feedback valence (F(1,68) = 36.3, P < 0.001, \({\eta}_{p}^{2}\) = 0.348; hit vs. miss = 1.32 ± 3.60 vs. 0.43 ± 3.69 μV).
The three-way interaction of feedback provider × cue valence × group was significant (F(2,136) = 5.8, P = 0.006, \({\eta}_{p}^{2}\) = 0.079; Fig. 4B). To break down this three-way interaction, we examined the feedback provider × group interaction for reward, punishment, and neutral cues, respectively (Fig. 4A). Results reveal a similar pattern of the feedback provider × group interaction for reward (F(1,68) = 4.0, P = 0.050, \({\eta}_{p}^{2}\) = 0.055) and punishment cues (F(1,68) = 8.8, P = 0.004, \({\eta}_{p}^{2}\) = 0.114): the MDS group showed a smaller P3 than the controls in the human feedback condition (reward: F(1,68) = 6.7, P = 0.012, \({\eta}_{p}^{2}\) = 0.090, control vs. MDS = 2.75 ± 2.91 vs. 0.84 ± 3.25 μV; punishment: F(1,68) = 9.1, P = 0.004, \({\eta}_{p}^{2}\) = 0.118, control vs. MDS = 2.91 ± 3.03 vs. 0.60 ± 3.36 μV), but not in the robot feedback condition (F ≤ 1.7, P ≥ 0.195; reward: 1.74 ± 3.34 vs. 0.72 ± 3.16 μV; punishment: 1.60 ± 3.43 vs. 0.75 ± 2.94 μV). In contrast, the feedback provider × group interaction was not significant in the neutral cue condition (F < 1); specifically, the two groups showed comparable P3 amplitudes in response to human (F(1,68) = 1.5, P = 0.230, \({\eta}_{p}^{2}\) = 0.021; control = 0.32 ± 3.42 μV, MDS = −0.65 ± 3.33 μV) and robot feedback (F(1,68) = 1.3, P = 0.263, \({\eta}_{p}^{2}\) = 0.018; control = −0.16 ± 3.39 μV, MDS = −1.06 ± 3.22 μV).
Discussion
Seeing that machines and automations are becoming increasingly important in our society, understanding how people interact with these intelligent agents is necessary [99]. This study investigates whether social reward processing in depression would show different patterns for human–human and human–robot interactions in nonclinical populations. At the anticipation stage of social reward processing, the MDS group had a lower hit rate and a smaller cue-evoked CNV compared to the control group in the human condition but not the robot condition; then at the consumption stage, the MDS group had a smaller feedback-evoked P3 than the control group in the human condition but not the robot condition, after receiving reward and punishment (but not neutral) cues. Overall, these findings indicate that depressive symptoms are more likely to be associated with abnormalities in the anticipation and consumption stages of social reward processing during human–human interaction, compared to human–robot interaction.
Anticipation stage
Since this experiment used a block design for two conditions (human vs. robot), participants were able to predict whether the upcoming feedback would come from a person or a robot in each block. As revealed by our results, this prediction has significantly modulated both task performance and neural signals among mild depressive participants. Specifically, when the evaluative feedback was expected to be delivered by a human companion, the depressive participants were less likely to hit the target stimulus within a given time window compared to the controls. Also, the CNV elicited by cue presentation (reflecting the preparation process for the forthcoming target) [64] was smaller among the depressive participants than the controls. These results are in line with our recently published data [53].
According to the literature, depression symptoms are closely associated with low anticipation of social rewards, which is one of the major characteristics of social anhedonia [7, 100]. In our opinion, the depressive participants had difficulties in conceiving and expecting social reward during their interactions with other people; consequently, they mobilized less effort to finish the follow-up task and thus provided worse performance [101,102,103]. Our interpretation is supported by the correlation between the hit rate and the CNV amplitude, which has also been observed in our previous research [53]. In contrast, when the feedback was expected to be delivered by the robot Karl, no between-group difference was detected in the hit rate or CNV amplitude. Therefore, we suggest that participants’ motivation to achieve social reward in this condition was not significantly affected by depression.
Consumption stage
The effect of depression was significant at both the early, coarse phase (indexed by the FRN) and the late, deliberate phase (indexed by the P3) of social reward consumption [52, 54]. First, the depressive participants showed a smaller FRN elicited by miss feedback than the controls in the reward and neutral cue conditions. In both conditions, miss feedback indicates the omission of social reward. Thus, we suggest that the FRN finding indicates blunted sensitivity to social reward in depression [52, 102, 104]. In line with these results, previous studies have consistently reported that the FRN amplitude decreases as a function of depression severity in various reward-related paradigms [105, 106]. However, the relationship between depression and the FRN remained the same pattern regardless of whether participants were interacting with other persons or with the robot Karl. In our opinion, this was because the early phase of feedback evaluation is susceptible to the similarities in perceptual appearance between humans and human-like machines, that is, an effect of anthropomorphism (see “Introduction”).
More importantly, the effect of depression on the P3 component (indicating a more top-down phase of feedback evaluation) differentiated between human–human and human–robot interactions. That is, the P3 elicited by human feedback, but not robot feedback became smaller among the depressive participants than the controls in the reward and punishment cue conditions. While the P3 has been related to diverse cognitive operations [107, 108], many studies suggest that this component is closely associated with a motivational level in decision-making tasks, such that a larger P3 indicates the stronger motivational significance of the ongoing event [56, 71, 109,110,111]. For instance, a larger P3 was observed when participants were comparing their task performance with that of a disliked opponent’s, indicating stronger motivation to outperform that opponent [112]. Given this background knowledge, we propose a possible interpretation of the P3 finding: compared to the controls, mild depressive individuals’ motivation to provide better performance in a social scenario is less likely to be modulated by their expectation of receiving positive social feedback (or avoiding negative social feedback) from a human companion. In agreement with our explanation, Frey, Frank, & McCabe recently point out that depression negatively affects the ability in using social feedback to appropriately update future actions [101, 113]. In contrast, the motivational level reflected by P3 amplitude was unresponsive to depression when participants received feedback from the robot Karl.
As an alternative explanation, the P3 amplitude may indicate the amount of attentional resources during feedback processing. This explanation stems from the knowledge that the P3 component has been frequently related to attentional function [108, 114]. From this perspective, mild depressive participants might have allocated fewer attentional resources than the controls on human feedback in the reward and punishment cue conditions, possibly because they were pessimistic about receiving favorable feedback from humans. Future studies are awaited to determine between the “attentional” hypothesis and the “motivational” hypothesis, though both of them are based on the relationship between depression and feedback expectation.
Conclusions and concluding remarks
In the science fiction I, Robot by Isaac Asimov, its protagonists argue that robots are actually more reliable and trustworthy than humans—many people in real life share the same belief [115, 116]. For instance, de Visser et al. reported that anthropomorphic robot agents were associated with greater trust resilience (i.e., a higher resistance to breakdowns in trust) than human agents [28, 117]. Relevantly, the current study discovers that social reward processing during human–robot interaction is less prone to be affected by depression compared to that during human–human interaction. That is, depressive participants held a more pessimistic expectation about getting social rewards and were less likely to be motivated by those rewards in the human condition, therefore they showed poorer task performance than the controls; nevertheless, the same was not true in the robot condition.
According to the clinical literature, depressive episodes are strongly associated with negative social experience (e.g., peer rejection in schools) in which people fail to receive social rewards [118,119,120]. Among depressive individuals, this kind of experience generates social evaluative concerns and accordingly inhibits the willingness to engage in social interactions [121,122,123,124]. As a result, depressive individuals perceive themselves to be less socially competent than ordinary people and underestimate the likelihood of obtaining favorable feedback (e.g., social approval) from other persons [7]. On the contrary, we suggest that interacting with robots may not be interfered with by these concerns, because people tend to trust the objectivity of robot feedback. From a theoretical perspective, our findings indicate that human–robot interaction not only has social rewarding properties, but is also safe from some risk factors associated with human–human interaction (e.g., social evaluative threats and peer pressure); depression symptoms interplay with these risk factors during human–human interaction, but not human–robot interaction. These findings may help understand to what extent classical theories of social psychology (which are based on human–human interaction) could be used to explain human–robot interaction [14, 22]. From a practical perspective, our findings address why companion robots significantly improve quality of life, mood, and loneliness for adults with depression, and highlights the value of this kind of robots as a useful tool in clinical practice [19, 46].
Below we list some limitations and future directions for follow-up research to consider. Most importantly, our task did not allow life-like communication between participants and different agents due to technical restrictions; instead, we provided human and robot facial expression pictures to participants as social feedback. Although face processing is a fundamental part of social interaction [125], the real-time exchange of verbal and gestural signals should have enhanced the ecological validity of this study. Thus, we encourage follow-up studies to re-examine the reliability of our findings in more ecological contexts, and to investigate the potential influence of social intelligence skills of robots (not just their appearance). Second, seeing that the robot Karl has an anthropomorphic appearance, future studies could try using robots with alternative designs (e.g., zoomorphic or caricatured) for practical reasons. Moreover, it should be noted that the current findings are derived from the SID task; using alternative paradigms associated with social reward processing (e.g., the Ultimatum Game or the Trust Game [126]) would help examine the generality of these findings. Finally, it would be interesting to explore the processing of robot feedback in other clinical conditions. For instance, Raffard et al. point out that patients with schizophrenia may have difficulties in recognizing and interpreting complex social cues from humans and human-like robots [47, 127]. Accordingly, we hypothesize that schizophrenia would affect social reward sensitivity in not only human–human but also human–robot interactions.
Data availability
All the data and code used in this study could be available by contacting the first author, Prof. Dandan Zhang (e-mail: zhangdd05@gmail.com).
References
Hill CA. Affiliation motivation: people who need people… but in different ways. J Personal Soc Psychol. 1987;52:1008–18.
Kawamichi H, Sugawara SK, Hamano YH, Makita K, Kochiyama T, Sadato N. Increased frequency of social interaction is associated with enjoyment enhancement and reward system activation. Sci Rep. 2016;6:1–11.
Rademacher L, Schulte-Rüther M, Hanewald B, Lammertz S. Reward: from basic reinforcers to anticipation of social cues. In: Wöhr M, Krach S, editors. Social behavior from rodents to humans. Cham: Springer; 2015. p. 207–21.
Bishop SJ, Gagne C. Anxiety, depression, and decision making: a computational perspective. Annu Rev Neurosci. 2018;41:371–88.
Kupferberg A, Bicks L, Hasler G. Social functioning in major depressive disorder. Neurosci Biobehav Rev. 2016;69:313–32.
Forbes EE, Dahl RE. Research review: altered reward function in adolescent depression: what, when and how? J Child Psychol Psychiatry. 2012;53:3–15.
Silk JS, Davis S, McMakin DL, Dahl RE, Forbes EE. Why do anxious children become depressed teenagers? The role of social evaluative threat and reward processing. Psychological Med. 2012;42:2095–107.
Enneking V, Krüssel P, Zaremba D, Dohm K, Grotegerd D, Förster K, et al. Social anhedonia in major depressive disorder: a symptom-specific neuroimaging approach. Neuropsychopharmacology. 2019;44:883–9.
Klawohn J, Burani K, Bruchnak A, Santopetro N, Hajcak G. Reduced neural response to reward and pleasant pictures independently relate to depression. Psychological Med. 2020;51:1–9.
Nelson BD, Perlman G, Klein DN, Kotov R, Hajcak G. Blunted neural response to rewards as a prospective predictor of the development of depression in adolescent girls. Am J Psychiatry. 2016;173:1223–30.
Barch DM, Whalen D, Gilbert K, Kelly D, Kappenman ES, Hajcak G, et al. Neural indicators of anhedonia: predictors and mechanisms of treatment change in a randomized clinical trial in early childhood depression. Biol Psychiatry. 2020;88:879–87.
He Z, Liu Z, Zhao J, Elliott R, Zhang D. Improving emotion regulation of social exclusion in depression-prone individuals: A tDCS study targeting right VLPFC. Psychological Med. 2019;50:1–12.
Rappaport BI, Barch DM. Brain responses to social feedback in internalizing disorders: a comprehensive review. Neurosci Biobehav Rev. 2020;118:784–808.
Cross ES, Ramsey R. Mind meets machine: towards a cognitive science of human–machine interactions. Trends Cogn Sci. 2021;25:200–12.
Chen SC, Jones C, Moyle W. Social robots for depression in older adults: a systematic review. J Nurs Scholarsh. 2018;50:612–22.
Penning MJ. Health, social support, and the utilization of health services among older adults. J Gerontol Ser B: Psychological Sci Soc Sci. 1995;50:S330–S339.
Jøranson N, Pedersen I, Rokstad AM, Ihlebaek C. Effects on symptoms of agitation and depression in persons with dementia participating in robot-assisted activity: a cluster-randomized controlled trial. J Am Med Dir Assoc. 2015;16:867–73.
Petersen S, Houston S, Qin H, Tague C, Studley J. The utilization of robotic pets in dementia care. J Alzheimer’s Dis. 2017;55:569–74.
Chen SC, Moyle W, Jones C, Petsky H. A social robot intervention on depression, loneliness, and quality of life for Taiwanese older adults in long-term care. Int Psychogeriatr. 2020;32:981–91.
Wada K, Shibata T, Saito T, Sakamoto K, Tanie K. Psychological and social effects of one year robot assisted activity on elderly people at a health service facility for the aged. In: Proceedings of the 2005 IEEE International Conference on Robotics and Automation. Barcelona: IEEEE; 2005. p. 2785–90. https://ieeexplore.ieee.org/xpl/conhome/10495/proceeding.
Wada K, Shibata T. Robot therapy in a care house—its sociopsychological and physiological effects on the residents. In: Proceedings 2006 IEEE International Conference on Robotics and Automation. Orlando, FL: IEEE; 2006. p. 3966–71.
Krämer NC, von der Pütten A, Eimler S. Human-agent and human-robot interaction theory: Similarities to and differences from human-human interaction. In: Zacarias M, de Oliveira JV, editors. Human-computer interaction: the agency perspective. Berlin, Heidelberg: Springer; 2012. p. 215–40.
Krämer NC. Social communicative effects of a virtual program guide. In: Panayiotopoulos T, Gratch J, Aylett R, Ballin D, Olivier P, Rist T, editors. Intelligent virtual agents. Berlin, Heidelberg: Springer; 2005. p. 442–453.
Ozeki T, Mouri T, Sugiura H, Yano Y, Miyosawa K. Use of communication robots to converse with people suffering from schizophrenia. ROBOMECH J. 2020;7:1–14.
Nass C, Steuer J, Tauber B., Tauber ER computers are social actors. In: Adelson B, Dumais S, Olson J (eds) In: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems. New York, NY: Association for Computing Machinery; 1994. p. 72–78.
Nass C, Moon Y. Machines and mindlessness: social responses to computers. J Soc Issues. 2000;56:81–103.
Nass C, Moon Y, Fogg BJ, Reeves B, Dryer DC. Can computer personalities be human personalities? Int J Hum-Computer Stud. 1995;43:223–39.
de Visser EJ, Monfort SS, McKendrick R, Smith MA, McKnight PE, Krueger F, et al. Almost human: anthropomorphism increases trust resilience in cognitive agents. J Exp Psychol: Appl. 2016;22:331–49.
Nass C, Lee KM. Does computer-synthesized speech manifest personality? Experimental tests of recognition, similarity-attraction, and consistency-attraction. J Exp Psychol: Appl. 2001;7:171–81.
De Kleijn R, van Es L, Kachergis G, Hommel B. Anthropomorphization of artificial agents leads to fair and strategic, but not altruistic behavior. Int J Hum-Computer Stud. 2019;122:168–73.
Torta E, van Dijk E, Ruijten PA, Cuijpers RH. The ultimatum game as measurement tool for anthropomorphism in human–robot interaction. In: Herrmann G, Pearson MJ, Lenz A, Bremner P, Spiers A, Leonards U, editors. International Conference on Social Robotics (ICSR 2013). Cham: Springer; 2013. p. 209–17.
Sandoval EB, Brandstetter J, Obaid M, Bartneck C. Reciprocity in human-robot interaction: a quantitative approach through the prisoner’s dilemma and the ultimatum game. Int J Soc Robot. 2016;8:303–17.
Wu T, Luo Y, Broster LS, Gu R, Luo YJ. The impact of anxiety on social decision-making: behavioral and electrodermal findings. Soc Neurosci. 2013;8:11–21.
Chaminade T, Okka MM. Comparing the effect of humanoid and human face for the spatial orientation of attention. Front Neurorobotics. 2013;7:12.
Gobbini MI, Gentili C, Ricciardi E, Bellucci C, Salvini P, Laschi C, et al. Distinct neural systems involved in agency and animacy detection. J Cogn Neurosci. 2011;23:1911–20.
Duffy BR. Anthropomorphism and the social robot. Robot Autonomous Syst. 2003;42:177–90.
Złotowski J, Proudfoot D, Yogeeswaran K, Bartneck C. Anthropomorphism: opportunities and challenges in human–robot interaction. Int J Soc Robot. 2015;7:347–60.
Cohen L, Khoramshahi M, Salesse RN, Bortolon C, Słowiński P, Zhai C, et al. Influence of facial feedback during a cooperative human-robot task in schizophrenia. Sci Rep. 2017;7:15023.
Guo F, Li M, Qu Q, Duffy VG. The effect of a humanoid robot’s emotional behaviors on users’ emotional responses: evidence from pupillometry and electroencephalography measures. Int J Hum–Computer Interact. 2019;35:1947–59.
Dubal S, Foucher A, Jouvent R, Nadel J. Human brain spots emotion in non humanoid robots. Soc Cogn Affect Neurosci. 2011;6:90–97.
Chammat M, Foucher A, Nadel J, Dubal S. Reading sadness beyond human faces. Brain Res. 2010;1348:95–104.
Cerda L, Graziani P, Del-Monte J. Emotional valence recognition on virtual, robotic, and human faces: a comparative study. J Technol Behav Sci. 2020;6:151–8.
Banks MR, Willoughby LM, Banks WA. Animal-assisted therapy and loneliness in nursing homes: use of robotic versus living dogs. J Am Med Dir Assoc. 2008;9:173–7.
Bickmore TW, Picard RW. Establishing and maintaining long-term human-computer relationships. ACM Trans Computer-Hum Interact. 2005;12:293–327.
Koay KL, Syrdal DS, Walters ML, Dautenhahn K. Living with robots: investigating the habituation effect in participants’ preferences during a longitudinal human-robot interaction study. In: Proceedings of the 16th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN 2007). New York, NY: IEEE; 2007. p. 564–9. https://ieeexplore.ieee.org/xpl/conhome/4415041/proceeding.
Robinson H, Macdonald B, Kerse N, Broadbent E. The psychosocial effects of a companion robot: a randomized controlled trial. J Am Med Dir Assoc. 2013;14:661–7.
Raffard S, Bortolon C, Khoramshahi M, Salesse RN, Burca M, Marin L, et al. Humanoid robots versus humans: how is emotional valence of facial expressions recognized by individuals with schizophrenia? An exploratory study. Schizophrenia Res. 2016;176:506–13.
Spreckelmeyer KN, Krach S, Kohls G, Rademacher L, Irmak A, Konrad K, et al. Anticipation of monetary and social reward differently activates mesolimbic brain structures in men and women. Soc Cogn Affect Neurosci. 2009;4:158–65.
Rademacher L, Krach S, Kohls G, Irmak A, Gründer G, Spreckelmeyer KN. Dissociation of neural networks for anticipation and consumption of monetary and social rewards. Neuroimage. 2010;49:3276–85.
Oldham S, Murawski C, Fornito A, Youssef G, Yücel M, Lorenzetti V. The anticipation and outcome phases of reward and loss processing: a neuroimaging meta-analysis of the monetary incentive delay task. Hum Brain Mapp. 2018;39:3398–418.
Martins D, Rademacher L, Gabay AS, Taylor R, Richey JA, Smith DV, et al. Mapping social reward and punishment processing in the human brain: a voxel-based meta-analysis of neuroimaging findings using the social incentive delay task. Neurosci Biobehav Rev. 2021;122:1–17.
Pegg S, Arfer KB, Kujawa A. Altered reward responsiveness and depressive symptoms: an examination of social and monetary reward domains and interactions with rejection sensitivity. J Affect Disord. 2021;282:717–25.
Zhang D, Shen J, Bi R, Zhang Y, Zhou F, Feng C, et al. Differentiating the abnormalities of social and monetary reward processing associated with depressive symptoms. Psychol Med. 2020. https://www.cambridge.org/core/journals/psychological-medicine/article/abs/differentiating-the-abnormalities-ofsocial-and-monetary-reward-processing-associated-with-depressivesymptoms/CB36DEEEFD12FF6441AA2F0FA4E0230E.
He Z, Zhang D, Muhlert N, Elliott R. Neural substrates for anticipation and consumption of social and monetary incentives in depression. Soc Cogn Affect Neurosci. 2019;14:815–26.
Ait Oumeziane B, Jones O, Foti D. Neural sensitivity to social and monetary reward in depression: clarifying general and domain-specific deficits. Front Behav Neurosci. 2019;13:199.
San Martín R. Event-related potential studies of outcome processing and feedback-guided learning. Front Hum Neurosci. 2012;6:304.
Walsh MM, Anderson JR. Learning from experience: event-related potential correlates of reward processing, neural adaptation, and behavioral choice. Neurosci Biobehav Rev. 2012;36:1870–84.
Glazer JE, Kelley NJ, Pornpattananangkul N, Mittal VA, Nusslock R. Beyond the FRN: broadening the time-course of EEG and ERP components implicated in reward processing. Int J Psychophysiol. 2018;132:184–202.
Krigolson OE. Event-related brain potentials and the study of reward processing: methodological considerations. Int J Psychophysiol. 2017;132:175–83.
Flores A, Munte TF, Donamayor N, Event-related EEG. responses to anticipation and delivery of monetary and social reward. Biol Psychol. 2015;109:10–19.
Ait Oumeziane B, Schryer-Praga J, Foti D. “Why don’t they ‘like’ me more?”: comparing the time courses of social and monetary reward processing. Neuropsychologia. 2017;107:48–59.
Wei S, Zou Z, Xue Z, Cao S, Yu H, Han J, et al. Social incentives anticipation and consummation: Investigating neural activity in women using methamphetamine. Front Psychol. 2020;11:88.
Walter WG, Cooper R, Aldridge VJ, McCallum WC, Winter AL. Contingent negative variation: an electric sign of sensori-motor association and expectancy in the human brain. Nature. 1964;203:380–4.
Kononowicz TW, Penney TB. The contingent negative variation (CNV): timing isn’t everything. Curr Opin Behav Sci. 2016;8:231–7.
Gehring WJ, Willoughby AR. The medial frontal cortex and the rapid processing of monetary gains and losses. Science. 2002;295:2279–82.
Sambrook TD, Goslin J. A neural reward prediction error revealed by a meta-analysis of ERPs using great grand averages. Psychological Bull. 2015;141:213–35.
Holroyd CB, Coles MGH. The neural basis of human error processing: reinforcement learning, dopamine, and the error-related negativity. Psychological Rev. 2002;109:679–709.
Philiastides MG, Biele G, Vavatzanidis N, Kazzer P, Heekeren HR. Temporal dynamics of prediction error processing during reward-based decision making. Neuroimage. 2010;53:221–32.
Wu Y, Zhou XL. The P300 and reward valence, magnitude, and expectancy in outcome evaluation. Brain Res. 2009;1286:114–22.
Yang Q, Tang P, Gu R, Luo W, Luo YJ. Implicit emotion regulation affects outcome evaluation. Soc Cogn Affect Neurosci. 2015;10:824–31.
Polezzi D, Sartori G, Rumiati R, Vidotto G, Daum I. Brain correlates of risky decision-making. Neuroimage. 2010;49:1886–94.
San Martín R, Appelbaum LG, Pearson JM, Huettel SA, Woldorff MG. Rapid brain responses independently predict gain maximization and loss minimization during economic decision-making. J Neurosci. 2013;33:7011–9.
Lin Y, Gu R, Luan S, Hu L, Qin S, Luo YJ. The hierarchical sensitivity to social misalignment during decision-making under uncertainty. Soc Cogn Affect Neurosci. 2020;16:565–75.
Zhang D, Gu R, Broster LS, Jiang Y, Luo W, Zhang J, et al. Linking brain electrical signals elicited by current outcomes with future risk decision-making. Front Behav Neurosci. 2014;8:84.
Gu R, Lei Z, Broster L, Wu T, Jiang Y, Luo YJ. Beyond valence and magnitude: a flexible evaluative coding system in the brain. Neuropsychologia. 2011;49:3891–7.
Karaaslan F, Gonul AS, Oguz A, Erdinc E, Esel E. P300 changes in major depressive disorders with and without psychotic features. J Affect Disord. 2003;73:283–7.
Hansenne M, Ansseau M. Contingent negative variation and personality in depression. Neuropsychobiology. 2001;44:7–12.
Faul F, Erdfelder E, Lang AG, Buchner AG. *Power 3: a flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behav Res. Methods. 2007;39:175–91.
Faul F, Erdfelder E, Buchner A, Lang AG. Statistical power analyses using G*Power 3.1: tTests for correlation and regression analyses. Behav Res Methods. 2009;41:1149–60.
Beck AT, Steer RA, Brown GK. Beck depression inventory. 3rd ed. San Antonio, TX: The Psychological Corporation; 1996.
First MB, Gibbon M, Spitzer RL, Williams JBW. Structured clinical interview for DSM-IV-TR axis I disorders, research version, non-patient edition (SCID-I/NP). New York: Biometrics Research Department, New York State Psychiatric Institute; 2002.
Spielberger CD, Gorsuch RL, Lushene R, Vagg PR, Jacobs GA. Manual for the state-trait anxiety inventory. Palo Alto, CA: Consulting Psychologist Press; 1983.
Eckblad ML, Chapman LJ, Chapman JP, Mishlove M. The revised social anhedonia scale. Unpublished test copies available from T. R. Kwapil. Champaign, ILU: IUC Department of Psychology; 1982.
Liebowitz MR. Social phobia. Mod Probl Pharmacopsychiatry. 1987;22:141–73.
Rosenberg M. Society and the adolescent self-image. Princeton, NJ: Princeton University Press; 1965.
de Graaf MM, Ben Allouch S, van Dijk JA. Why would I use this in my home? A model of domestic social robot acceptance. Hum–Computer Interact. 2019;34:115–73.
Kalegina A, Schroeder G, Allchin A, Berlin K, Cakmak M. Characterizing the design space of rendered robot faces. In: Hoffman G, Tapus A, (eds) Proceedings of the 2018 ACM/IEEE International Conference on Human-Robot Interaction 2018. IEEE. New York, NY: 2018. p. 96–104.
Picton TW, Bentin S, Berg P, Donchin E, Hillyard SA, Johnson R Jr, et al. Guidelines for using human event-related potentials to study cognition: recording standards and publication criteria. Psychophysiology. 2000;37:127–52.
Landes I, Bakos S, Kohls G, Bartling J, Schulte-Körne G, Greimel E. Altered neural processing of reward and punishment in adolescents with major depressive disorder. J Affect Disord. 2018;232:23–33.
Knutson B, Westdorp A, Kaiser E, Hommer D. FMRI visualization of brain activity during a monetary incentive delay task. Neuroimage. 2000;12:20–27.
Knutson B, Adams CM, Fong GW, Hommer D. Anticipation of increasing monetary reward selectively recruits nucleus accumbens. J Neurosci. 2001;21:RC159.
Chronaki G, Soltesz F, Benikos N, Sonuga-Barke EJS. An electrophysiological investigation of reinforcement effects in attention deficit/hyperactivity disorder: dissociating cue sensitivity from down-stream effects on target engagement and performance. Developmental Cogn Neurosci. 2017;28:12–20.
Gu R, Jiang Y, Kiser S, Luo YJ, Kelly TH. Impulsive personality dimensions are associated with altered behavioral performance and neural responses in the Monetary Incentive Delay Task. Neuropsychologia. 2017;103:59–68.
Bergstrom ZM, de Fockert JW, Richardson-Klavehn A. ERP and behavioural evidence for direct suppression of unwanted memories. Neuroimage. 2009;48:726–37.
Drollette ES, Scudder MR, Raine LB, Moore RD, Saliba BJ, Pontifex MB, et al. Acute exercise facilitates brain function and cognition in children who need it most: an ERP study of individual differences in inhibitory control capacity. Developmental Cogn Neurosci. 2014;7:53–64.
Kessler RC, McGonagle KA, Zhao S, Nelson CB, Hughes M, Eshleman S, et al. Lifetime and 12-month prevalence of DSM-III-R psychiatric disorders in the United States: results from the National Comorbidity Survey. Arch Gen Psychiatry. 1994;51:8–19.
Kessler RC, Chiu WT, Demler O, Merikangas KR, Walters EE. Prevalence, severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry. 2005;62:617–27.
Mineka S, Watson D, Clark LA. Comorbidity of anxiety and unipolar mood disorders. Annu Rev Psychol. 1998;49:377–412.
Goodyear K, Parasuraman R, Chernyak S, de Visser E, Madhavan P, Deshpande G, et al. An fMRI and effective connectivity study investigating miss errors during advice utilization from human and machine agents. Soc Neurosci. 2017;12:570–81.
Forbes EE. Where’s the fun in that? Broadening the focus on reward function in depression. Biol Psychiatry. 2009;66:199–200.
Frey AL, Frank MJ, McCabe C. Social reinforcement learning as a predictor of real-life experiences in individuals with high and low depressive symptomatology. Psychological Med. 2021;51:408–15.
Pegg S, Ethridge P, Shields GS, Slavich GM, Weinberg A, Kujawa A. Blunted social reward responsiveness moderates the effect of lifetime social stress exposure on depressive symptoms. Front Behav Neurosci. 2019;13:178.
Brinkmann K, Franzen J, Rossier C, Gendolla GH. I don’t care about others’ approval: dysphoric individuals show reduced effort mobilization for obtaining a social reward. Motiv Emot. 2014;38:790–801.
Keren H, O'Callaghan G, Vidal-Ribas P, Buzzell GA, Brotman MA, Leibenluft E, et al. Reward processing in depression: a conceptual and meta-analytic review across fMRI and EEG studies. Am J Psychiatry. 2018;175:1111–20.
Proudfit GH, Bress JN, Foti D, Kujawa A, Klein DN. Depression and event-related potentials: emotional disengagement and reward insensitivity. Curr Opin Psychol. 2015;4:110–3.
Proudfit GH. The reward positivity: from basic research on reward to a biomarker for depression. Psychophysiology. 2015;52:449–59.
Verleger R. Effects of relevance and response frequency on P3b amplitudes: review of findings and comparison of hypotheses about the process reflected by P3b. Psychophysiology. 2020;57:e13542.
Polich J, Criado JR. Neuropsychology and neuropharmacology of P3a and P3b. Int J Psychophysiol. 2006;60:172–85.
Nieuwenhuis S, Aston-Jones G, Cohen JD. Decision making, the P3, and the locus coeruleus-norepinephrine system. Psychological Bull. 2005;131:510–32.
Zhang D, Gu R, Wu T, Broster LS, Luo Y, Jiang Y, et al. An electrophysiological index of changes in risk decision-making strategies. Neuropsychologia. 2013;51:1397–407.
Yeung N, Sanfey AG. Independent coding of reward magnitude and valence in the human brain. J Neurosci. 2004;24:6258–64.
Zhang H, Gu R, Yang M, Zhang M, Han F, Li H, et al. Context-based interpersonal relationship modulates social comparison between outcomes: an event-related potential study. Soc Cogn Affect Neurosci. 2021;16:439–52.
Murphy FC, Michael A, Robbins TW, Sahakian BJ. Neuropsychological impairment in patients with major depressive disorder: the effects of feedback on task performance. Psychological Med. 2003;33:455–67.
Gray HM, Ambady N, Lowenthal WT, Deldin P. P300 as an index of attention to self-relevant stimuli. J Exp Soc Psychol. 2004;40:216–24.
Dzindolet MT, Pierce LG, Beck HP, Dawe LA. The perceived utility of human and automated aids in a visual detection task. Hum Factors. 2002;44:79–94.
Dzindolet MT, Peterson SA, Pomranky RA, Pierce LG, Beck HP. The role of trust in automation reliance. Int J Hum-Computer Stud. 2003;58:697–718.
Goodyear K, Parasuraman R, Chernyak S, Madhavan P, Deshpande G, Krueger F. Advice taking from humans and machines: An fMRI and effective connectivity study. Front Hum Neurosci. 2016;10:542.
Kendler KS, Hettema JM, Butera F, Gardner CO, Prescott CA. Life event dimensions of loss, humiliation, entrapment, and danger in the prediction of onsets of major depression and generalized anxiety. Arch Gen Psychiatry. 2003;60:789–96.
Nolan SA, Flynn C, Garber J. Prospective relations between rejection and depression in young adolescents. J Personal Soc Psychol. 2003;85:745–55.
Monroe SM, Rohde P, Seeley JR, Lewinsohn PM. Life events and depression in adolescence: relationship loss as a prospective risk factor for first onset of major depressive disorder. J Abnorm Psychol. 1999;108:606–14.
Davey CG, Yucel M, Allen NB. The emergence of depression in adolescence: development of the prefrontal cortex and the representation of reward. Neurosci Biobehav Rev. 2008;32:1–19.
Slavich GM, Thornton T, Torres LD, Monroe SM, Gotlib IH. Targeted rejection predicts hastened onset of major depression. J Soc Clin Psychol. 2009;28:223–43.
Lee A, Hankin BL, Mermelstein RJ. Perceived social competence, negative social interactions, and negative cognitive style predict depressive symptoms during adolescence. J Clin Child Adolesc Psychol. 2010;39:603–15.
Allen NB, Badcock PB. The social risk hypothesis of depressed mood: evolutionary, psychosocial, and neurobiological perspectives. Psychological Bull. 2003;129:887–913.
Little AC, Jones BC, DeBruine LM. The many faces of research on face perception. Philos Trans R Soc B: Biol Sci. 2011;366:1634–7.
Stallen M, Sanfey AG. The cooperative brain. Neuroscientist. 2013;19:292–303.
Scassellati B, Admoni H, Matarić M. Robots for use in autism research. Annu Rev Biomed Eng. 2012;14:275–94.
Acknowledgements
This study was funded by the National Natural Science Foundation of China (31970980, 32071083, 32020103008), Shenzhen-Hong Kong Institute of Brain Science (2021SHIBS0003), Science and Technology Planning Project of Guangdong Province of China (2019A050510048), Major Program of the Chinese National Social Science Foundation (17ZDA324), and Youth Innovation Promotion Association CAS (2019088).
Author information
Authors and Affiliations
Contributions
DZ conceived the experiment; SJ, SL, and KG performed the experiment and collected the data; DZ analyzed the data; DZ and RG wrote the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Ethical approval
All procedures performed in this study were in accordance with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The local ethics committee approved the experimental protocol.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhang, D., Shen, J., Li, S. et al. I, robot: depression plays different roles in human–human and human–robot interactions. Transl Psychiatry 11, 438 (2021). https://doi.org/10.1038/s41398-021-01567-5
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41398-021-01567-5