Thought-encourager: Designing social interactions via AI agents responsibly

Abstract

Recent studies have been shown the potential of AI agents in daily social interactions. Google Duplex, which is new technology within an AI assistant to conduct natural conversations to complete specific tasks such as making a reservation over the phone without the user directly speaking. Gmail's smart replies, which are AI-generated responses based on the email, and an email sender can choose one of several responses generated by AI to edit or send it immediately. While AI can improve human communication abilities, it can negatively impact how humans build and maintain relationships. It is crucial to assess the AI agent's social and ethical implications in interpersonal relationships. AI practitioners and researchers need to develop a foundational empirical understanding of AI agents' impact on human relationships and behaviors to design social interactions via AI agents responsibly. Lane states that responsible technological innovation is significant design challenges of the 21st century. Microsoft launched Responsible Innovation Toolkits to help AI practitioners identify potential negative impacts of AI technologies on people and design technologies with more responsibilities. One of the toolkits is Judgment Call, a tool for industry product teams considering new technology from multiple perspectives and identifying ethical concerns based on AI principles from Microsoft. The limitations of Judgment Call include lack of understanding of various stakeholders who might be impacted by new technology, a short amount of time for participants to have critical thinking on misuse of new technology, and not proposing solutions to prevent unintended negative results. Moreover, no previous studies have been done to anticipate and address potential negative unintended social interactions via AI agents. This study developed Thought-encourager to identify potential negative unintended results of the social interactions via AI agents. Thought-encourager is designed to guide AI design researchers and practitioners to have critical thinking processes based on the user reviews developed from user scenarios of 6 expectations on the future social interactions via AI agents. Through practicing critical thinking and reflective processes, implications for the responsible design of social interactions via AI agents are proposed.