E-contact facilitated by conversational agents reduces interethnic prejudice and anxiety in Afghanistan

Intergroup contact, while crucial for social harmony, often leads to conflict due to differing interests, values, and cultures. The intergroup contact hypothesis posits that increased contact between groups can reduce prejudice and hostility, though Allport's original hypothesis focused on direct, in-person contact. Increasingly, research acknowledges the role of indirect contact channels, including electronic contact (e-contact), in improving intergroup relations. However, concerns remain about the potential for e-contact to exacerbate prejudice, especially in contexts marked by segregation and conflict. This study addresses this gap by examining the impact of a conversational agent (CA) – an AI program engaging in human-like conversations – as a facilitator of e-contact in Afghanistan, a country with a long history of interethnic conflict and segregation. The CA acts as a neutral authority figure, moderating online discussions and promoting equal status interaction. The hypothesis is that CA-facilitated e-contact will reduce intergroup prejudice and anxiety more effectively than unfacilitated e-contact in this challenging context. Afghanistan’s diverse population, with a history of Pashtun dominance and conflict involving Tajiks and Hazaras, provides a particularly relevant setting to test this hypothesis.

The intergroup contact hypothesis, while widely accepted, has limitations. Studies have shown that intergroup contact can sometimes exacerbate prejudice, particularly in contexts of segregation and conflict. Moreover, the transferability of positive contact effects beyond the immediate participants is questionable. The role of indirect contact, especially via the internet, has gained increased attention. E-contact offers advantages such as secure environments, reduced anxiety, and overcoming geographical barriers. However, research on optimal conditions for e-contact, particularly in conflict zones like Afghanistan, remains scarce. Existing research emphasizes the need for structured, supervised intergroup contact to prevent negative outcomes. This study builds upon existing literature by introducing the use of a conversational agent as a novel approach to facilitate positive and productive e-contact, offering a potential solution to issues of deficient supervision in online intergroup interactions.

This study employed a randomized controlled trial (RCT) design with 128 Afghan participants (Pashtun, Tajik, and Hazara) recruited through an online agency. Participants, aged 23-37 with Bachelor's or Master's degrees and sufficient English proficiency, were randomly assigned to four conditions: a control group with no CA facilitation and three experimental groups with different CA settings (normal CA, CA prompting for issues, CA prompting for ideas). The study involved two discussion sessions: a two-hour synchronous session focusing on a proposal for Afghanistan's frozen assets and a three-day asynchronous session focused on upgrading informal settlements. Both sessions utilized D-Agree, an online discussion platform incorporating the CA. Intergroup prejudice was measured using Bogardus' social distance scale, and intergroup anxiety was measured using a modified version of Stephan & Stephan's Intergroup Anxiety Scale. Measurements were taken at three time points: before the intervention (T1), after the synchronous discussion (T2), and after the asynchronous discussion (T3). Data were analyzed using repeated measures ANOVA, with condition and time as factors. Bayesian analyses were also conducted to validate null findings. The study also assessed the impact of CA facilitation on social engagement by measuring the number of ideas generated and the length of opinions.

The study found no significant baseline differences in intergroup prejudice or anxiety among the three ethnic groups. Participants in the CA facilitation conditions demonstrated significantly higher social engagement (more ideas and longer opinions) compared to the control group. Repeated measures ANOVAs revealed significant Condition × Time interactions for both intergroup prejudice and anxiety. Specifically, the CA condition showed a significant decrease in intergroup prejudice from T1 to T3, while the control group showed a slight increase. Similarly, the CA condition exhibited significant reductions in intergroup anxiety from T1 to T2 and T1 to T3, whereas the control group showed no significant change. No significant differences in intergroup prejudice or anxiety reduction were found across the different CA facilitation styles. Effect sizes were characterized as small to medium, comparable to those found in broader meta-analyses on intergroup contact.

The findings support the study's hypothesis that CA-facilitated e-contact reduces interethnic prejudice and anxiety in Afghanistan. The increased social engagement observed in the CA conditions suggests that the CA effectively stimulated meaningful participation and fostered a more inclusive discussion environment. The lack of significant differences among CA facilitation styles indicates that the CA's presence as a neutral moderator was the primary driver of positive outcomes. These findings have practical implications for online conflict management and internet policing strategies. The study highlights the potential of CAs as cost-effective tools for reducing online prejudice, particularly in contexts where direct intervention is challenging. The moderate effect sizes, though obtained in a high-conflict context, are comparable to results from broader intergroup contact research, reinforcing the potential of this approach.

This study demonstrates the effectiveness of CA-facilitated e-contact in reducing interethnic prejudice and anxiety in Afghanistan. The CA's role as a neutral authority figure promoting equal status interaction and structured discussion appears key. Future research should explore the generalizability of these findings to other contexts and investigate the specific mechanisms through which CAs achieve these effects. The ethical considerations surrounding CA use in intergroup contact, such as bias mitigation and transparency, require further attention.

The study's limitations include the unequal sample sizes across conditions and the potentially unrepresentative sample due to participants' higher socioeconomic background. The controlled experimental setting may not perfectly reflect real-world online interactions. Future studies should address these limitations by using larger, more representative samples and exploring the implementation of CAs in less controlled settings.