Constituents and nature of verbal communication in the shape of identity and conduct of artificial voice agents and people.
Verbal communication, i.e., spoken or written, plays a fundamental role in shaping, expressing, and reaffirming our identities. As individuals, our choice of words, tonality, and style of speech often reflect not only our personal experiences but also our cultural, social, and historical backgrounds. Howarth (2011) delves deep into the associations between representations, identity, and resistance in communication, highlighting how our spoken words can both construct and challenge dominant societal narratives. According to this perspective, verbal communication is not merely an exchange of information; it is a continuous process of identity negotiation and assertion. Hence, our verbal interactions are deeply tied to our sense of self, our perceived place in society, and our behavioral conduct [1].
Understanding the nature of verbal communication and its associations with identity and conduct becomes crucial when designing interactions with virtual agents, e.g., conversational AI, digital humans or even robots or voice-enabled devices. In order to be effective and relatable, these agents must emulate the complexities of human verbal communication. In Lee et al., (2021), the manner in which these agents communicate greatly influences users’ perceptions of their identity and intent. Just as humans rely on the nuances of verbal cues to convey identity and dictate conduct, virtual agents must be programmed to communicate in ways that resonate with human expectations and social norms. This requires not just the use of appropriate language, but also the mix of cultural nuances, empathetic responses, and context-aware interactions. Designers could then elicit a sense of trust, relatability, and authenticity in the interactions between users and these virtual entities.
Modern state-of-the-art (SoTA) conversational AI models aim to emulate the richness and complexity of human-human conversations. These models, while highly advanced, are not without their flaws. According to Adewumi et al., (2022), some of the significant challenges the surveyed SoTA open-domain conversational AI face, include producing bland and repetitive responses and struggling with understanding and appropriately reacting to figurative language. Moreover, the widespread use of Large Language Models (LLMs) like ChatGPT has evidence several limitations like making up facts, i.e., hallucination, or delayed responses problematic for real-time use. One notable observation is that hybrid models, which combine multiple methods, tend to outperform single-architecture solutions. For instance, hybrid models might combine retrieval and generative methods to ensure that responses are both accurate and contextually appropriate. Despite advancements, achieving human-like performance in open-domain conversations remains an uphill task, especially when considering low-resource languages, which are underrepresented in current AI models. Additionally, an emerging area of ethical concern is the gender representation in conversational AI, with a notable prevalence of female-gendered chatbots, raising questions about gender biases and stereotypes [3].
This research project then explores how humans associate their identity with verbal communication, how this identity shapes behavior, and how the advent of digital voices simulating our own can influence and potentially change our conduct. This framework will guide the development and study of conversational agents following up on SoTA findings. The goal is to approach the current challenges in verbal interactions with artificial agents to find solutions to them.
[1] Howarth, C. (2011). Representations, Identity, and Resistance in Communication. In: Hook, D., Franks, B., Bauer, M.W. (eds) The Social Psychology of Communication. Palgrave Macmillan, London. https://doi.org/10.1057/9780230297616_8 [2] Lee, S. K., Kavya, P., & Lasser, S. C. (2021). Social interactions and relationships with an intelligent virtual agent. International Journal of Human-Computer Studies, 150, 102608. doi:10.1016/j.ijhcs.2021.102608 [3] Adewumi, T.; Liwicki, F.; Liwicki, M. State-of-the-Art in Open-Domain Conversational AI: A Survey. Information 2022, 13, 298. https://doi.org/10.3390/info13060298
