TLDR: This research paper explores how Social Identity Theory (SIT) and Social Categorization Theory (SCT) apply to human-agent interaction (HAI). It explains how humans categorize and identify with artificial agents, influencing perceptions and behaviors. The paper outlines current human-oriented activities and future agent-oriented possibilities within social identity work, emphasizing the ethical implications of designing increasingly human-like AI. It introduces the ‘uncanny killjoy’ concept, advocating for caution and transparency regarding an agent’s artificiality to prevent deception and address biases in HAI.
Our daily lives are increasingly intertwined with artificial agents, from chatbots powered by advanced language models to social robots. As these technologies become more sophisticated and human-like, understanding how we interact with them socially becomes crucial. A recent research paper, “Social Identity in Human-Agent Interaction: A Primer”, by Katie Seaborn, delves into this complex area by applying established social science theories to the world of human-agent interaction (HAI).
Understanding Social Identity
The paper introduces the Social Identity Approach (SIA), which includes Social Identity Theory (SIT) and Social Categorization Theory (SCT). These theories explain how people understand themselves and others in social contexts, connecting individuals to groups through psychological mechanisms. Essentially, we categorize ourselves and others into ‘in-groups’ (groups we belong to) and ‘out-groups’ (groups we don’t). This process influences our attitudes and behaviors.
Key concepts within SIA include personal identities (how we define ourselves individually) and social identities (how we define ourselves as group members). There’s also the idea of a ‘superordinate identity,’ like ‘human,’ which encompasses many other identities. The paper raises the question of whether ‘artificial agent’ could also be a superordinate category.
How We Engage with Social Identities
SIA outlines several activities we undertake, consciously or not, related to social identity:
- Social Categorization: Placing ourselves and others into groups.
- Social Identification: Aligning our self-concept with a particular group.
- Social Comparison: Comparing our groups to others, which can lead to social mobility (leaving a group), social change (challenging the status quo within a group), social competition (raising our group’s status), or social creativity (finding new ways to value our group).
These activities have consequences, such as depersonalization (losing individuality within a group), individualization (highlighting personal identity), group polarization (attitudes influenced by the majority), group cohesiveness (unity based on group prototypicality), stereotyping (using explanatory models about groups), conformity (internalizing group identity), mutual influence, weak in-group bias, and positive group distinctiveness (defining our group positively).
Applying Social Identity to AI
The paper then extrapolates how SIA can apply to artificial social agents, acknowledging that not all human models will directly transfer. It proposes several axioms:
- Artificial agents are not human, though they might become ‘people’ in a science fiction sense.
- Anthropomorphic agents (human-like in form) will always trigger social identity processes in human users.
- Human reactions to agents may not always be as expected.
- Social identities are complex, dynamic, and potentially unpredictable.
- Not all human social identity consequences will be found in HAI contexts.
- Not all social identities will matter all the time; saliency and context are key.
Currently, humans are the primary drivers of social identity work in HAI. We categorize agents, sometimes leading to ‘identity passing’ (agents being perceived as group members, even human) or ‘identity rejection’ (rejecting an agent’s social cues). We might also identify with agent groups, or engage in social comparison with them, such as competition or creativity.
Looking to the future, the paper envisions a scenario where agents could possess their own personal identities, reject human-imposed categories, show favoritism, exhibit creativity in self-perception, influence human identities, and even make sacrifices for fellow artificial beings.
Research and Ethical Considerations
For researchers, the paper suggests focusing on how people perceive an agent’s identity and their own identity in relation to agents. It also highlights the importance of understanding how different agent embodiments (physical robots, virtual agents) influence social identity processes, noting that factors like physical presence, proxemics (personal space), and even size can play a role.
The author emphasizes the need to expand the scope of social identities explored beyond common factors like gender, to include neurodiversity, body characteristics, religion, fandoms, and more, recognizing their permeability, stability, and voluntariness.
Crucially, the paper issues a “clarion call” for an ethical orientation, introducing the concept of the “uncanny killjoy.” This involves being extraordinarily cautious and embracing the ‘eeriness’ of machines to prevent deception. It argues that obscuring the artificial origins of our creations must be done with great care. Ethical considerations include addressing power imbalances, avoiding superficial and potentially biased identity detection mechanisms, and designing agents that subtly cue their artificiality. The paper also reminds us that most scientific knowledge, including social identity theories, is often Western-centric and may not apply universally, advocating for culturally sensitive approaches.
Also Read:
- Beyond Prompts: The Emergence of Social Identity in AI Agent Societies
- Unpacking AI’s Influence: How Language Models Persuade and Amplify Bias
Conclusion
As artificial agents become more integrated into our social fabric, understanding the implications of social identity is paramount. The paper serves as a vital primer, urging designers, researchers, and users to approach human-agent interactions with awareness, ethical responsibility, and a critical eye, ensuring that our creations enhance, rather than compromise, our social world.
