Table 1
Study overview of the four experiments.
| EXPERIMENT | GOAL | HYPOTHESES | SUPPORTED | PARTICIPANTS | AGENTS | LEVEL | |
|---|---|---|---|---|---|---|---|
| Validation (Experiment 1) | Validate whether people’s feelings are altered based on the result of the game and the focus (self/other) of the question during our competitive game | 1. | Participants would feel better when they won the game as opposed to when the other player won | Yes | N = 81 | Human | Interpersonal |
| 2. | Participants would feel worse after they themselves lost as opposed to when the other player lost | Yes | |||||
| 3. | Participants would feel better when the other player than themselves lost | Yes | |||||
| Experiment 2 | Verify if intergroup biases would emerge in human-human teams during our competitive game | 1. | Participants would feel relatively more empathy towards ingroup team members than to outgroup members | Yes | N = 37 | Human | Intergroup |
| 2. | Participants would feel more schadenfreude towards the opponents who lost than towards team members who lost. | Yes | |||||
| Experiment 3 | Investigate whether individuals show intergroup biases towards robots in human-robot teams | 1. | See experiment 2 – H1 | Yes | N = 87 | Human & robot (Cozmo) | Intergroup |
| 2. | See experiment 2 – H2 | Yes | |||||
| 3. | More salient intergroup empathy and schadenfreude biases when comparing ingroup and outgroup human players than when comparing the ingroup and outgroup Cozmo robots | No | |||||
| Experiment 4 | Test if findings generalize across robots who differ in human likeness. | 1. | See experiment 2 – H1 | Yes | N = 93 | Human & robot (NAO) | Intergroup |
| 2. | See experiment 2 – H2 | Yes | |||||
| 3. | Increasing tendency to have intergroup biases from the least human-like agent (Cozmo) to a more human-like robot (NAO) and finally the human agent. | No | |||||
Figure 1
Competitive reaction time game. Participants were arbitrarily paired with either a human (Experiment 2), mechanoid Cozmo robot (Experiment 3), or a humanoid Nao robot (Experiment 4) while playing against a similar team in a competitive reaction time game. The game outcome was determined by the average reaction time to the targets per team. The fastest team to respond to the targets won a round and gained five points, while the other team lost two points. We measured participant’s trial-by-trial emotional fluctuations at the level of each player (self, their team member and opponents) for every game outcome (win or lose) while taking into account interpersonal factors (team identification, blame for the result, and score difference). Participant completed two scales that probed trial-by-trial fluctuations in positive affect (feeling good) and negative affect (feeling bad). In Experiment 3 and 4, teams were shuffled every 10 rounds resulting in teams of all possible combinations.
Figure 2
Trial-by-trial ratings of positive and negative reactions to every game outcome for each player. A self-other bias was observed when participants played the competitive reaction time game against one player (A). Participants not only felt better when they won and worse when they lost, but also schadenfreude, they felt better when the other player lost a round. A robust empathy and schadenfreude bias driven by team membership was observed (B and C). Participants felt better when ingroup members won and worse when ingroup members lost (empathy), and felt better when outgroup members lost (schadenfreude). These intergroup schadenfreude and empathy biases were observed when participants formed a team with humans (B) and humanoid (NAO) and mechanoid robots (Cozmo) (C). Data is calculated relative to the self for C. The dots represent the raw data and the beans the density of the responses. The black bar shows the mean with the white rectangle showing the 95% confidence interval.
Figure 3
Team identification consistently increased the intergroup schadenfreude bias. People who identified more with their team compared to the opponent team (difid) felt more schadenfreude, feeling good when the other team lost. This effect was observed for both human–human (A) and human–robot teams (B–C). The points represent the raw data with the linear regression lines of the fitted models with 95% confidence interval around the lines.
Table 2
Trial-by-trial ratings of positive and negative reactions to every game outcome in Experiment 3.
| FEELING GOOD | FEELING BAD | ||||
|---|---|---|---|---|---|
| WIN | LOSE | WIN | LOSE | ||
| Ingroup | |||||
| Human | 0.01 [–0.01, 0.03] | 0.04 [0.02, 0.06] | –0.02 [–0.04, 0.00] | –0.04 [–0.07, –0.01] | |
| Robot | –0.00 [–0.02, 0.02] | 0.06 [0.04, 0.08] | 0.01 [–0.00, 0.03] | –0.05 [–0.08, –0.03] | |
| Outgroup | |||||
| Human | –0.38 [–0.44, –0.33] | 0.35 [0.29, 0.41] | 0.35 [0.30, 0.40] | –0.34 [–0.40, –0.28] | |
| Robot | –0.38 [–0.44, –0.33] | 0.34 [0.28, 0.40] | 0.35 [0.29, 0.40] | –0.35 [–0.41, –0.29] | |
[i] Mean values with 95% confidence intervals are shown.
Table 3
Trial-by-trial ratings of positive and negative reactions to every game outcome in Experiment 4.
| FEELING GOOD | FEELING BAD | ||||
|---|---|---|---|---|---|
| WIN | LOSE | WIN | LOSE | ||
| Ingroup | |||||
| Human | –0.00 [–0.02, 0.02] | 0.01 [–0.00, 0.03] | –0.00 [–0.02, 0.02] | –0.02 [–0.04, 0.01] | |
| Robot | 0.02 [–0.04, –0.00] | 0.06 [0.04, 0.08] | 0.03 [0.01, 0.04] | –0.02 [–0.04, –0.00] | |
| Outgroup | |||||
| Human | –0.46 [–0.51, –0.40] | 0.42 [0.36, 0.48] | 0.40 [0.35, 0.46] | –0.45 [–0.45, –0.32] | |
| Robot | –0.48 [–0.53, –0.43] | 0.42 [0.36, 0.48] | 0.41 [0.36, 0.47] | –0.39 [–0.45, –0.33] | |
[i] Mean values with 95% confidence intervals are shown.