Whose Voice Counts? Mapping Stakeholder Perspectives on AI Through Public Submissions to the U.S. Government

Problem
This preprint addresses the gap in understanding stakeholder perspectives on AI, particularly how these perspectives are represented in public submissions to the U.S. government during the consultation for the Trump Administration’s AI Action Plan. The authors highlight the lack of comprehensive analysis on the diverse viewpoints of various stakeholders, such as individuals, academia, and the private sector, in shaping AI policy and its societal implications.

Method
The authors introduce a corpus cleaning pipeline to preprocess the public submissions dataset. They employ topic modeling and frequency analysis to identify and categorize the predominant themes expressed by different stakeholder groups. The analysis focuses on comparing the concerns raised by individuals against those articulated by institutional stakeholders, particularly the private sector. The methodology includes the use of Natural Language Processing (NLP) techniques to extract topics and quantify their prevalence, although specific algorithms or hyperparameters are not disclosed.

Results
The findings reveal a stark contrast in the concerns of different stakeholders. Individuals predominantly express apprehensions regarding the societal impacts of AI, while private sector stakeholders focus on issues related to AI development, security, and policy frameworks. The analysis indicates that the AI Action Plan primarily reflects the priorities of the private sector, with individual concerns being underrepresented. The authors do not provide specific quantitative metrics or effect sizes but emphasize the qualitative differences in topic prevalence across stakeholder groups.

Limitations
The authors acknowledge several limitations, including the potential bias in the dataset due to the self-selection of respondents who chose to submit letters. They also note that the analysis is constrained by the scope of the submissions, which may not capture the full spectrum of public opinion on AI. Additionally, the reliance on topic modeling may oversimplify complex sentiments and nuances in stakeholder perspectives. An obvious limitation not flagged by the authors is the temporal context of the submissions, which may not reflect current sentiments or developments in AI policy.

Why it matters
This work has significant implications for the development of AI governance frameworks. By mapping stakeholder perspectives, it highlights the necessity for inclusive dialogue in AI policy-making that considers diverse viewpoints, particularly those of individuals who may be adversely affected by AI technologies. The findings suggest that current policy frameworks may inadequately address public concerns, potentially leading to societal discord. This research can inform future studies on stakeholder engagement and contribute to more equitable AI governance practices.

Authors: Alina Karakanta, Alex Christiansen, Tomás Dodds, Bissie Anderson, Matteo Fuoli, Marcus Perlman, Aletta G. Dorst
Source: arXiv:2605.22650
URL: https://arxiv.org/abs/2605.22650v1