Listening to the Voices: Describing Ethical Caveats of Conversational User Interfaces According to Experts and Frequent Users

Document Title

Listening to the Voices: Describing Ethical Caveats of Conversational User Interfaces According to Experts and Frequent Users

Document Information

• Research Domain: Ethical issues in conversational user interfaces (CUI) within Human-Computer Interaction (HCI)
• Keywords: CUI, conversational user interface, conversational agent, voice assistant, chatbot, thematic analysis, ethical design, deceptive design patterns, dark patterns

Research Background and Issues

• Identified Problems or Challenges:
  1. The proliferation of conversational user interfaces (CUIs) has brought significant benefits but also entails risks of deceptive use and breaches of user trust.
  2. While "dark patterns" in graphical user interfaces (GUIs) have been extensively studied, how to identify and address similar issues in CUIs remains unclear.
  3. Current understanding of the ethical design potential and challenges of CUIs is limited, particularly in areas such as trust, privacy protection, and managing user expectations.
• Importance of the Issues: Ambiguous designs may mislead users, negatively impacting user experience and potentially placing users in vulnerable situations. As CUIs become more widespread, neglecting intentional or unintentional ethical issues could significantly harm technology acceptance.
• Research Motivation and Related Work:
  1. Identifying potential ethical risks in conversational user interfaces to prevent poor design from influencing user decisions, especially issues akin to "dark patterns" in graphical interfaces.
  2. Advancing technology development while enhancing transparency and user experience in CUIs through effective design frameworks.

Solutions

• Proposed Methods or Solutions:

  1. Conducting semi-structured interviews with 27 participants, including academic researchers, industry practitioners, and frequent users, to explore ethical issues in CUIs.
  2. Developing five design themes that reveal ethical vulnerabilities and proposing a framework called "CUI Expectation Cycle (CEC)".
  3. Integrating characteristics of dark patterns from Mathur et al. and existing theories (e.g., Norman's action cycle and Oliver's expectation confirmation theory).

• Innovative Contributions:

  1. Identifying key ethical challenges and user pain points in CUI design.
  2. Introducing the CUI Expectation Cycle framework to address gaps in existing best practices by focusing on expectation setting, trust-building, and inclusivity.

• Implementation Steps and Techniques:

  1. Conducting in-depth interviews with diverse groups (researchers, practitioners, users).
  2. Applying reflexive thematic analysis to decode and synthesize data, establishing five major themes.
  3. Leveraging existing academic theories and design principles to develop a new framework for guiding future research and practice.

Research Outcomes

• Specific Results:

  1. Five Major Themes:
     • Building Trust and Protecting Privacy: Transparency of information and control over data.
     • Guiding Interaction: Information presentation and feature discovery.
     • Humanization and Harmony: Balancing anthropomorphic design with technological capabilities.
     • Inclusivity and Diversity: Addressing the needs of marginalized users.
     • Setting Expectations: Enhancing transparency to reduce user frustration.
  2. Introducing the CUI Expectation Cycle (CEC) framework, which includes two bridges (execution bridge and evaluation bridge) and two limiting factors (deception boundary and distrust boundary) to narrow the gap between user expectations and CUI systems.

• Comparison with Existing Solutions:

  1. Compared to existing GUI-based design guidelines and frameworks, CEC is better suited for ethical design in CUIs.
  2. It addresses unique challenges such as user diversity, anthropomorphic features, and privacy transparency.

• Experimental and Evaluation Results:

  1. Privacy and trust issues were commonly reported among the 27 interviewees.
  2. The CEC framework was proven effective in addressing user frustration and misleading design issues during experiments.

• Limitations and Future Directions:

  1. Limited sample size and diversity of participant backgrounds (e.g., most user interviews focused on voice assistants rather than chatbots).
  2. Further exploration is needed to refine humanization design, particularly quantifying its impact on users.
  3. Future research is recommended to expand into other interface types, such as text-based chatbots, and explore broader industry application scenarios.