Should ChatGPT Vote for You? Automated Voting and Artificial Intelligence

• In Event: Knowing and Deciding Responsibly in Democracy

Abstract

Recent advances in artificial intelligence (AI) systems generally and large language models (LLMs) more narrowly have reinvigorated concerns regarding how emerging digital technologies impact contemporary democracies. Optimists anticipate innovations in civic technologies that may reinvigorate local-level civic engagement. Critics, in contrast, warn of a deluge of AI-generated dis/misinformation propagating rampantly across the digitized public sphere. This paper is concerned in particular with proposals for automated voting systems that would allow citizens to delegate their voting choices to AI assistants. Advocates (e.g. Susskind 2018) contend that voluntary systems for automated voting can increase the inclusiveness of elections by removing motivational and information barriers to voting. Further, algorithmically assisted voting in some form appears inevitable as little can easily be done to prevent citizens from consulting ChatGPT or other AI systems on their smartphones within voting booths.

I ultimately argue against the use of automated voting systems on the basis that such technologies would incentivize political elites to prioritize communicating with algorithms over humans. Before reaching that conclusion, however, I demonstrate that common objections to delegating decisions to algorithms do not hold here. Thus, in this paper, I adopt an interdisciplinary approach bridging critical algorithmic studies, democratic theory, and Americanist political science to offer three contributions.

First, I demonstrate that on a simple technical level automating voting systems are potentially practical. Simple algorithms that recommend voting choices have existed for years (e.g., https://voteforpolicies.org.uk/). Likewise, various countries and subnational units have experimented with systems to permit citizens to vote online, and large majorities of Europeans desire online voting options (Jonsson and Luca de Tena 2021). Hence, automated voting would entail removing humans from in between already existing technologies, rather than radical feats of novel engineering.

Second, I show that automated voting systems are not a significant departure from current practices within contemporary democracies. For instance, the pervasive political ignorance and civic disengagement among the mass public suggests that delegating voting decisions to algorithms would change little for how average citizens interact with their political systems. Additionally, paper ballots in states that allow party-line voting with a single mark can be understood as an already widespread form of algorithmic voting. Further still, common concerns regarding delegating judgments to AI systems including inscrutability, hallucinations, biases, etc. also apply to voters delegating their judgments to human peers.

Third, I contend that automated voting systems ought to still be rejected, even if the common objections to delegating decisions to algorithms do not hold here. I specifically argue that when an algorithm becomes an obligatory passage point for politicians seeking to gain and retain elected office, then appeasing that algorithm takes priority over communicating to humans. Hence, the problem resides not within the technologies in themselves, but instead in the anticipated reactions of political elites.

Author

• Matthew Lucky, Indiana University Bloomington