Can LLMs advance democratic values? (2410.08418v2)

Abstract: LLMs are among the most advanced tools ever devised for analysing and generating linguistic content. Democratic deliberation and decision-making involve, at several distinct stages, the production and analysis of language. So it is natural to ask whether our best tools for manipulating language might prove instrumental to one of our most important linguistic tasks. Researchers and practitioners have recently asked whether LLMs can support democratic deliberation by leveraging abilities to summarise content, as well as to aggregate opinion over summarised content, and indeed to represent voters by predicting their preferences over unseen choices. In this paper, we assess whether using LLMs to perform these and related functions really advances the democratic values that inspire these experiments. We suggest that the record is decidedly mixed. In the presence of background inequality of power and resources, as well as deep moral and political disagreement, we should be careful not to use LLMs in ways that automate non-instrumentally valuable components of the democratic process, or else threaten to supplant fair and transparent decision-making procedures that are necessary to reconcile competing interests and values. However, while we argue that LLMs should be kept well clear of formal democratic decision-making processes, we think that they can be put to good use in strengthening the informal public sphere: the arena that mediates between democratic governments and the polities that they serve, in which political communities seek information, form civic publics, and hold their leaders to account.

Can LLMs Advance Democratic Values? An Expert Overview

In their preprint, Lazar and Manuali investigate the potential for LLMs to support democratic values in deliberation and decision-making. Recognizing the critical role of language in democracy, the authors explore whether LLMs, as advanced linguistic tools, can enhance this important societal function. They conclude that while LLMs offer potential benefits, their role in formal democratic decision-making processes should be approached with caution.

Key Findings

The authors review several research initiatives exploring the use of LLMs in democratic settings. These initiatives typically leverage LLMs for summarization, opinion aggregation, representing voter preferences, facilitating public deliberation, and even implementing democratic principles. The efficacy of these approaches is varied and context-dependent.

1. Summarization: Utilizing LLMs to synthesize public feedback during consultations could streamline bureaucratic processes. However, challenges include potential biases and inaccuracies that might distort democratic participation and accountability.
2. Aggregation and Representation: LLMs are employed to propose consensus-building statements and predict voter sentiments. Yet, concerns regarding transparency and the models' representational fidelity suggest limitations in their application for direct democratic decision-making.
3. Facilitation: LLMs have facilitated dialogue in online deliberative settings but often lack the ability to foster genuine face-to-face democratic engagement, which is vital for compromise and preference transformation.
4. Implementation: The use of LLMs in executing decisions—such as implementing constitutions in deliberative settings—raises significant challenges in terms of control, authorization, and effective governance, underlining the need for careful oversight.

Theoretical and Practical Implications

Lazar and Manuali emphasize the importance of democratic values—equality, authorisation, civic virtue, conciliation, transformation, and collective intelligence—and evaluate how well LLM-based initiatives serve these goals. They argue that:

• The non-instrumental values of democracy, such as equality and civic engagement, could be undermined if LLMs replace human participation.
• Instrumental benefits like collective intelligence might be advanced through better information processing, but only if LLMs' biases and errors are rigorously managed.

Future Directions and Considerations

The authors urge caution against deploying LLMs in roles that could compromise democratic norms, arguing instead for their application in enhancing the informal public sphere. LLMs have potential to improve information retrieval, content moderation, and societal engagement online. By fostering a more informed digital public sphere, LLMs could indirectly support democratic health.

Given their findings, researchers and policymakers are encouraged to focus on LLMs' potential to augment the public sphere without substituting for critical democratic processes. Future AI systems should prioritize transparency, accountability, and fairness to enable real democratic engagement.

Conclusion

Lazar and Manuali's analysis clarifies that while LLMs hold promise, they are not a panacea for democratic challenges. Instead, their role is best limited to supporting democratic infrastructure in ways that complement rather than replace human judgment and participation. By focusing on enhancing the informal public sphere, LLMs could indeed contribute to a more resilient democracy.