A critical appraisal of equity in conversational AI: Evidence from auditing GPT-3's dialogues with different publics on climate change and Black Lives Matter

  title={A critical appraisal of equity in conversational AI: Evidence from auditing GPT-3's dialogues with different publics on climate change and Black Lives Matter},
  author={Kaiping Chen and Anqi Shao and Jirayu Burapacheep and Yixuan Li},
Autoregressive language models, which use deep learning to produce human-like texts, have become increasingly widespread. Such models are powering popular virtual assistants in areas like smart health, finance, and autonomous driving, and facilitating the production of creative writing in domains from the entertainment industry to science communities. While the parameters of these large language models are improving, concerns persist that these models might not work equally for all subgroups in… 

Figures and Tables from this paper



How deliberative designs empower citizens’ voices: A case study on Ghana’s deliberative poll on agriculture and the environment

  • Kaiping Chen
  • Political Science
    Public understanding of science
  • 2020
Understanding of how to effectively foster public deliberation among populations in the Global South is contributed to and the nuances of expertise and public reasoning on science are measured.

Black Lives Matter Coverage: How Protest News Frames and Attitudinal Change Affect Social Media Engagement

Abstract This article investigates the relationship between digital news coverage of protests and the audience’s willingness to engage with a story about Black Lives Matter on social media. Using a

The Science of Open (Communication) Science: Toward an Evidence-Driven Understanding of Quality Criteria in Communication Research

It is argued that communication as a field proceed empirically as it applies open science practices to different subfields in communication and end the essay with pathways forward for a science of open (communication) science.

Enhancing human agency through redress in Artificial Intelligence Systems

It is argued that AI policy should introduce rights for users to swiftly contest or rectify an AI-enabled decision, which would empower individual autonomy and strengthen fundamental rights in the digital age.

Not one, but many “publics”: public engagement with global development in France, Germany, Great Britain, and the United States

New panel data from the Aid Attitudes Tracker (2013–18) provide evidence of both aggregate and individual-level change in engagement over time but with an important distinction: respondents in less engaged groups are less likely to move out of these groups and tend to stay unengaged.

Inequality is Always in the Room: Language & Power in Deliberative Democracy

Deliberative democracy has the potential to legitimize collective decisions. Deliberation's legitimating potential, however, depends on whether those who deliberate truly enter as equals, whether

Teaching Computer Ethics: A Deeply Multidisciplinary Approach

A curricular experiment at Stanford University focused on teaching computer ethics, where three faculty instructors, from philosophy, political science, and CS, each bring their respective lens to four related course modules: algorithmic decision-making, data privacy and civil liberties, AI and autonomous systems, and the power of platform companies.

Beyond Adversary Democracy

""Beyond Adversary Democracy" should be read by everyone concerned with democratic theory and practice."-Carol Pateman, "Politics" "Sociologists recurrently complain about how seldom it is that we

Effects of Persuasive Dialogues: Testing Bot Identities and Inquiry Strategies

It is shown that the perceived identity of the chatbot had significant effects on the persuasion outcome and interpersonal perceptions and interaction effects among perceived identities and inquiry strategies.

Threatening communication: A qualitative study of fear appeal effectiveness beliefs among intervention developers, policymakers, politicians, scientists, and advertising professionals

A study of the reasons for this persistent wide application of threatening communication in behavior change intervention development in The Netherlands finds that participants most closely involved with the actual intervention development were generally convinced that threatening information was to be prevented, but often did not understand the exact processes involved.