Consequently, CUI design imperatives are increasingly moving beyond traditional performance metrics such as speed, accuracy, and task success, so as to embrace more socio-technical dimensions governing successful human-AI coexistence and acceptance.

In this lecture I will thus discuss multi-disciplinary research across several independent studies, establishing an empirical foundation for designing socially competent AI. We will look into the findings of an expert-led investigation that resulted in a number of distinct characteristics of social intelligence required for next-generation agents.

Furthermore, we will explore the concept of authenticity, and talk about features such as transparent purpose, learning from experience, and conversational coherence

as vital requirements for fostering genuine user interactions and mitigating theperception of CUIs as mere work simplifiers. Importantly, we will also examine the user’s role and societal impact where empirical results demonstrate that individual personality traits - specifically the general propensity to trust and affinity for technology - significantly influence trust perception with CUIs and respective intelligent agents.

We will also address the ethical design challenge of gender, by looking at a study

that showed how the inclusion of a nonbinary voice can successfully disrupt negative gender stereotypes, even as it presents new challenges regarding user likability. Finally, we will explore human-chatbot interaction in high-disclosure settings, revealing counter-intuitive data on embodiment, where certain human-like avatars surprisingly facilitated, rather than inhibited, the breadth and depth of user self-disclosure.

Collectively, I hope these inputs will trigger some discussions on CUI development that prioritizes ethical design, trustworthiness, and socially intelligent interaction as potential core metrics for next-generation AI systems.