Can you meaningfully consent in eight seconds? Identifying Ethical Issues with Verbal Consent for Voice Assistants

William Seymour, Mark Coté, Jose M. Such

TL;DR

This paper critically examines the "voice-forward consent" (VFC) mechanism used by voice assistants like Alexa, arguing it undermines informed consent principles due to time pressure, limited information, and interface design. The authors propose potential solutions and pose key research questions to foster more ethical and usable consent practices for conversational AI.

What Problem Does This Paper Address?

The proliferation of voice assistants (VAs) has introduced new challenges in managing user consent for data sharing with third-party "skills" (applications). Traditionally, users would navigate complex permission menus on companion smartphone apps. Amazon's "voice-forward consent" (VFC) aims to streamline this by allowing users to grant permissions verbally mid-conversation. However, this paper argues that VFC, despite its convenience, compromises core tenets of informed consent. The authors highlight how the rapid, conversational nature of VAs, coupled with the intangible interface, blurs lines of responsibility and can pressure users into granting access without full understanding. This is particularly concerning given the increasing ubiquity of VAs and the sensitive personal data they can access.

Methods

This paper is a "provocation paper" that uses a critical analysis of existing literature from Human-Computer Interaction (HCI) and Ubiquitous Computing (Ubicomp), alongside key data protection regulations like GDPR. The authors analyze the current implementation of VFC for Alexa, drawing on examples and user experience principles to identify ethical shortcomings. They then propose potential solutions and frame open research questions for the community. The methodology is primarily analytical and conceptual, rather than empirical or experimental.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Time Pressure is a Consent Killer: The default 8-second timeout for VFC responses, combined with the "take it or leave it" nature of consent within an ongoing interaction, creates undue pressure, potentially negating the "freely given" aspect of informed consent.
  • Speech is a Poor Medium for Detail: The inherent limitations of spoken language and conversational turn-taking make it difficult to convey the necessary depth of information about data usage, third-party developers, and user rights, which is crucial for informed consent.
  • Interface Symmetry is Broken: Shifting consent granting to the voice interface, while revoking it through a separate app interface, creates an asymmetry that makes it harder for users to understand how to manage their permissions over time.
  • System vs. Skill Confusion: The lack of audible distinction between the VA's operating system and third-party skills can lead users to mistakenly believe the VA itself is vetting or controlling the data usage, rather than a potentially less trustworthy third party.
  • Consent as a Living Practice: The paper suggests that consent shouldn't be a one-time event but an ongoing process, with VAs potentially checking in with users periodically to reaffirm or revoke permissions.

Key Results

Not applicable, as this is a conceptual paper identifying issues and posing questions, rather than presenting empirical results.

Limitations and Future Work

The authors acknowledge that their work is an early-stage analysis and that further expert opinion and empirical research are needed. They propose several avenues for future work, framed as open questions:

  • User Experience of VFC: How can VFC be designed to be both usable and ethically sound, avoiding user habituation to dismissing consent prompts? This includes user studies and co-design.
  • Prioritizing VFC Elements: Given the constraints of VFC, which aspects of informed consent, dynamic consent, and legal grounds for data collection are most relevant and should be prioritized?
  • Multi-modal Consent: How can VFC be enhanced by integrating other modalities like screens, augmented reality, and web apps to provide richer, optional information and controls?
  • Platform API Integration: How can better VFC practices be integrated into existing platform APIs and flows, potentially allowing for more granular consent options and developer justifications?
  • Platform Architecture and Regulation: How will evolving VA norms, technology, and regulatory landscapes necessitate changes in platform architecture and consent frameworks?

Summary updated

Abstract

Determining how voice assistants should broker consent to share data with third party software has proven to be a complex problem. Devices often require users to switch to companion smartphone apps in order to navigate permissions menus for their otherwise hands-free voice assistant. More in line with smartphone app stores, Alexa now offers "voice-forward consent", allowing users to grant skills access to personal data mid-conversation using speech. While more usable and convenient than opening a companion app, asking for consent 'on the fly' can undermine several concepts core to the informed consent process. The intangible nature of voice interfaces further blurs the boundary between parts of an interaction controlled by third-party developers from the underlying platforms. This provocation paper highlights key issues with current verbal consent implementations, outlines directions for potential solutions, and presents five open questions to the research community. In so doing, we hope to help shape the development of usable and effective verbal consent for voice assistants and similar conversational user interfaces.
BibTeX
@inproceedings{seymour2022can,
  author = {William Seymour and Mark Coté and Jose M. Such},
  title = {{Can you meaningfully consent in eight seconds? Identifying Ethical Issues with Verbal Consent for Voice Assistants}},
  booktitle = {CUI},
  year = {2022},
  url = {https://openreview.net/forum?id=lpMJ54taJi},
}
Figure