Friend, Foe, or Bot? Exploring Intergroup Dynamics in Hybrid Human-Bot Teams
This is a strong CHI paper because it shows that transparency in hybrid teams is socially consequential in a non-obvious way. Hidden bots can help teams coordinate, yet they also enable unfair spillover punishment; revealed bots reduce cohesion and become socially downgraded, with blame redirected to humans. The result is a nuanced warning against simplistic “always disclose” or “never disclose” design rules for AI teammates.
Axes Lens
Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.
Contribution shape
- Knowledge form
- causal knowledge typical · 31/268
- Novelty type
- empirical finding typical · 68/268
- Abstraction level
- practice typical · 85/268
- Generalization target
- field argument typical · 55/268
- Validation mode
- controlled experiment typical · 47/268
Evidence profile
- Evidence strength
- strong typical · 158/268
- Claim alignment
- strong typical · 231/268
- Overclaim risk
- low typical · 53/268
Review Summary
This paper’s main value is that it takes a question many HCI researchers would frame narrowly—whether AI teammates help or hurt collaboration—and shows that the more important issue may be how AI teammates reshape relations between teams. The experiment is carefully structured around bot visibility, and that manipulation turns out to matter a great deal. When bots are hidden, humans appear to coordinate around them in a tacit way, often compensating for aggressive bot behavior rather than copying it. That is already interesting because it runs against straightforward reciprocity and imitation expectations. But the more important finding is the cost of that coordination: retaliation can spill over onto partners who did not commit the original act. When bots are revealed, the pattern changes again. Bots lose standing as full teammates, their actions carry less direct accountability, and conflict is redirected toward human partners. In other words, disclosure does not simply improve trust or fairness; it reorganizes who counts as a socially meaningful actor. Methodologically, the paper is persuasive because it does not rely on a single anecdotal effect. It uses a 240-participant between-subjects experiment, analyzes multiple behavioral and post-game outcomes, and interprets the results through reciprocity, imitation, and spillover mechanisms. The claims are also reasonably bounded by the authors’ own limitations section. This is still a simplified online game, with one human and one bot per team, no verbal communication, short-term interaction, and a Western participant pool. So the right reading is not that the paper discovers a universal law of hybrid teamwork, but that it establishes a robust and important pattern in a controlled setting: bot transparency changes the social distribution of cohesion, blame, and retaliation. That is a meaningful contribution for HCI because many real systems will face exactly this design tension.
What Changed
Canon before
The dominant baseline assumption in prior HCI research is that artificial teammates primarily influence collaboration within teams and that social identity theories developed for humans apply similarly to bots in team settings. Additionally, prior work often assumes that transparency of AI teammates either uniformly improves collaboration or has no drastic effect on intergroup biases in competitive contexts.
Departure from common sense
Contrary to standard expectations of reciprocity and imitation, participants often compensated for bot behavior rather than mirroring it. Hidden bots supported tacit within-team coordination but also produced spillover retaliation onto partners, while revealed bots were treated as less accountable and shifted conflict toward human teammates.
Actual novelty
The paper offers a novel empirical account of how bot-identity transparency changes both intrateam coordination and intergroup bias in hybrid human-bot teams. Using a controlled competitive game, it shows that concealment can improve coordination while enabling asymmetric retaliation, whereas disclosure marginalizes bots as secondary teammates and redirects accountability toward humans.
Evidence
The evidence comes from a between-subjects online experiment with 240 participants in hybrid dyads competing in StarHarvest under aware versus unaware bot-identity conditions. The paper reports beta regression analyses for reciprocity, imitation, and spillover, plus post-game evaluations, bonus allocations, and correlation analyses. The strongest support is for the claim that bot awareness systematically changes the direction and target of social responses. The authors also provide an explicit limitations section that bounds generalization to a simplified, short-term, Western-sample game setting without teammate communication.
“ Outgroup interactions have received relatively less attention. A few studies show that artificial agents embodying outgroup identities (e.g., in VR [110]) can trigger human-like prejudice, or that association with AI can alter perceptions of human outgroups [53, 68]. Even fewer works examine ingroup–outgroup dynamics in hybrid teams”
actual novelty · 1 Introduction · confidence 0.97
“ Instead, humans often reacted in the opposite direction, for example, compensating for their bot teammates’ actions rather than imitating them. Awareness shaped these dynamics: when bots were hidden, players regulated interactions within their own team but also spilled retaliation from humans onto bots; when bots were rev”
departure from common sense · 4.2 Bot Awareness Reshapes the Cooperation – Accountability Tradeoff (RQ1) · confidence 0.98
“There are multiple ways in which this study can be improved for future research. First, our study includes findings from a limited demographic of the Western population, thus interpretations should be confined to their specific cultural and social norms [3]. It is based on a controlled online game setting with simplified rules and a fixed set of bot strategies. While it s”
limitation · 6 Limitations · confidence 0.99
“nteract with each other. We conducted a between-subjects experiment with 240 participants recruited through Prolific [89], paired them into dyads that competed against each other with bot teammates (human–bot vs. human–bot). Each dyad was assigned one of four bot types and placed in one of”
validation scope · 3 Methodology · confidence 0.95
Limits
Method limits
The study is bounded by a controlled online game with simplified rules, fixed bot strategies, one human and one bot per team, short-term interaction, and no verbal communication. The authors also note that the bot strategies did not always manifest clearly, which may have limited participants’ ability to perceive bot personalities.
Deployment limits
The findings should not be transferred directly to real deployments with larger teams, richer communication, repeated interaction, or more adaptive AI agents. Practical settings may involve different accountability norms, negotiation opportunities, and social interpretations than those available in this constrained competitive game.
Boundary conditions
The reported effects depend on a cooperative-competitive StarHarvest setting with dyadic hybrid teams, manipulated bot visibility, and preconfigured bot behaviors. Results are especially tied to whether participants believed bots were human or knew they were bots, and may differ across cultures, longer-term relationships, and contexts with communication or mixed team compositions.
Position in field
This paper extends HCI work on human-AI teaming by shifting attention from within-team collaboration alone to intergroup dynamics between hybrid teams. Its contribution is field-relevant because it shows that transparency is not a simple good: it changes cohesion, blame, and retaliation patterns in ways that complicate standard social-identity expectations and common design intuitions about responsible AI disclosure.