TLDR: A research paper by Susan Ariel Aaronson and Michael Moreno examines public consultation processes on AI governance in Australia, Colombia, and the US, finding a significant gap between the promise and practice of participatory governance. Policymakers are largely failing to effectively solicit diverse public input, provide adequate information, widely market calls for comment, or demonstrably incorporate citizen feedback into policy. This “lost in translation” phenomenon undermines public trust and legitimacy in AI governance. The authors recommend improving AI literacy, marketing consultations widely, using innovative engagement strategies, ensuring diverse participation, and being responsive to public concerns, potentially leveraging AI tools to facilitate better dialogue.
A new research paper titled “Lost in Translation: Policymakers are not really listening to Citizen Concerns about AI” by Susan Ariel Aaronson and Michael Moreno reveals a significant disconnect between what citizens want to say about artificial intelligence (AI) and what policymakers actually hear and act upon. The paper highlights that despite governments seeking public input on AI risks and governance, much of the public’s feedback is getting lost, hindering the development of public trust in AI and its regulation.
The Challenge of Participatory AI Governance
The authors compared public consultation processes in Australia, Colombia, and the United States. Their findings indicate that these countries did not effectively engage a diverse range of citizens. Less than one percent of the population responded to calls for comment in each nation, and policymakers often appeared unresponsive to the feedback they did receive. This creates a persistent gap between the ideal of participatory AI governance and its practical implementation.
Democratic governance thrives on dialogue between the government and its people. However, AI’s rapid evolution and complexity often lead policymakers to rely heavily on academics and business leaders—experts who may have vested interests in how AI is regulated. Public polls consistently show deep concerns about AI, particularly regarding employment, human autonomy, democracy, and human rights. While people prefer AI developers to regulate the technology, only a small percentage trust tech companies to self-regulate.
A Look at Three Countries
The study focused on specific consultations: the U.S. National Telecommunications and Information Administration’s (NTIA) AI Open Model Weights Request for Comment, Colombia’s Ministry of Science, Technology and Innovation (MinCiencias) consultation on an ethical and sustainable AI roadmap, and Australia’s Department of Industry, Science, and Resources’ (DISR) “Supporting Responsible AI” discussion paper.
Who Participated?
The composition of respondents varied significantly:
- Australia: Received 510 comments (447 made public), with individuals making up 36%, firms 23.7%, and a balanced mix of think tanks, non-profits, and business associations.
- United States: Received 326 comments, predominantly from individuals (74%), followed by firms (9%) and think tanks/research institutions (7.7%).
- Colombia: Received only 73 comments, with academia dominating (57.5%), followed by the private sector (24.7%) and public entities (11%). This consultation primarily attracted experts rather than a broad cross-section of society.
Overall, none of the three governments successfully engaged a broad and diverse segment of their populations.
Information and Outreach Efforts
While all three governments provided some background materials, these were often insufficient or difficult to access. The U.S. NTIA provided extensive questions that many participants found overwhelming. Colombia released its AI Roadmap just two days before its consultation opened, with a very short window for feedback and limited publicity, making it hard for the public to find and respond to the draft document. Australia offered a range of materials and a longer consultation period, including ministerial roundtables and virtual town halls, which led to greater participation, possibly influenced by the recent RoboDebt scandal that highlighted algorithmic harms.
However, none of the governments made a sustained effort to widely market their calls for comment beyond traditional or digital channels, often assuming a baseline familiarity with AI governance concepts.
Using Public Feedback
The study found that governments largely failed to demonstrate how public feedback influenced policy changes. The U.S. acknowledged concerns in speeches but did not show how comments shaped reports. Colombia claimed feedback “improved” its roadmap, but the final document showed no highlighted modifications. Australia was the most responsive, releasing an interim report that mapped public concerns to actionable measures, including establishing a “Temporary Expert Advisory Group” to design AI guardrails.
Using the IAP2 Spectrum of Public Participation, the authors concluded that all three countries informed, consulted, and acknowledged public input but did little to actively involve, collaborate with, or empower citizens in the policy-making process.
Also Read:
- Bridging Disciplines: The Essential Role of Interdisciplinary Collaboration in AI Mental Health Chatbots
- The Geography of Trust in Foundation Models and Generative AI
Recommendations for a Better Dialogue
To build trust and legitimacy in AI governance, the authors propose eight key recommendations:
- Support AI literacy and provide easy-to-understand background materials.
- Consistently monitor public comment portals and demonstrate responsiveness.
- Widely market calls for comment through diverse media channels and trusted voices.
- Regularly hold online town halls on AI to genuinely hear public concerns.
- Utilize innovative engagement strategies like policy hackathons and crowdsourcing.
- Ensure participation from underrepresented groups, including those without internet access.
- Be responsive to public input, as citizens can offer unique perspectives.
- Leverage AI systems, such as swarm AI for collaborative decision-making and large language models for analyzing feedback, to facilitate broader and more effective engagement.
The paper concludes that policymakers must rethink how they engage citizens in governing AI to ensure that public comments are not lost in translation, thereby sustaining democracy and public trust.
