Introduction

Artificial Intelligence (AI) is transforming industries worldwide, and political polling is no exception. As traditional polling methods encounter challenges such as low response rates and potential inaccuracies, AI offers a new frontier by synthesizing public opinion from various data sources and creating nuanced models. By using AI, political campaigns, policy analysts, and the media gain powerful tools for gauging public sentiment, often in real-time. But while these technologies promise greater insights, they also raise concerns. Many people feel AI may deepen biases, influence poll results, and potentially distort democratic decision-making by subtly steering public opinion or framing issues in a particular light. This article explores AI’s potential role in polling, the challenges it might face, and the broader implications for public trust and news accuracy in the digital age.

Initial Use Cases for Polling

AI-powered polling tools go beyond static surveys. By analyzing data from social media, online articles, and other platforms, AI tools can capture ongoing public opinion and generate more dynamic insights. These models can even simulate responses by demographic subgroups, predicting shifts in opinion on new policies or controversial topics. AI agents, for example, can be “trained” to represent individuals with specific demographic and ideological characteristics. This allows campaigns and analysts to tailor their queries to “poll” specific segments of society. For instance, AI polling might enable campaigns to ask about the likelihood of support for a candidate’s stance on healthcare policy among young, urban voters with college degrees who lean liberal. This responsive system addresses a key limitation of traditional polling: the slow and costly process of surveying diverse populations, which often fails to capture rapid opinion changes.

Preliminary studies suggest that AI polling results can align closely with traditional polling data, but there are notable challenges. Experiments by researchers at Brigham Young University (BYU) have shown that AI models, when equipped with demographic details, can yield responses that mirror actual human polling results, sometimes accurately reflecting past election outcomes. For instance, BYU’s research found a high correlation between AI-generated responses and real voting patterns from the 2012, 2016, and 2020 U.S. elections, suggesting the potential for AI to replicate human voting behaviors. However, there are limits. AI models have been shown to occasionally produce outdated or inaccurate results if not trained on recent data, as seen in certain model errors related to the U.S. response to the Ukraine crisis. To achieve consistency, AI training data must be regularly updated to keep pace with the ever-evolving political landscape.

Figure 1

An example of one way that GenAI can influence the thought process of voters and the public’s perception of certain events is by changing the consumed information

Source: Trend Research

Will AI have a Permanent Role

AI polling could become a permanent fixture in electoral strategy if it proves both accurate and ethically sound. Unlike traditional methods, AI can quickly overcome barriers such as low engagement rates and response biases, offering real-time insights that allow for agile, data-driven decision-making. This combination offers a promising blend of computational power and traditional validation, though it may also bring its own set of issues. If AI polling becomes a staple in election strategies, transparency, and ethical standards will be crucial to prevent misuse, ensuring that AI’s role enhances rather than detracts from democratic processes. Researchers at institutions like Harvard’s Kennedy School of Government argue that the rise of AI polling will require public accountability measures such as organizations being upfront about the limitations of these technologies,  and periodic validation against human-led surveys, to ensure AI’s outputs accurately represent the public’s views. 

With AI becoming integral to polling, campaigns may cut costs and streamline polling processes, gaining flexibility to conduct in-depth analyses. Additionally, smaller campaigns and organizations could use AI to access insights that would typically require vast resources. On a broader level, this democratization of polling could deepen our understanding of public sentiment across various regions and demographics. However, a persistent reliance on AI also raises ethical questions.

Will campaigns and media outlets use AI polling data responsibly, or will they weaponize it to serve partisan agendas? Trust and accountability are paramount, and achieving these will likely involve balancing human oversight with AI-driven insights to foster informed, inclusive decision-making​ for all those involved.

Figure 2

The concern of voters that AI can influence their voting behavior

Source: Data for Progress

Conclusion

Trust in AI polling hinges on transparency in how data is gathered, analyzed, and presented. AI models, while powerful, are not immune to errors or biases, especially when relying on incomplete or outdated training data. Additionally, some people mistrust AI-driven news sources, often doubting the impartiality of machine-generated analyses. Studies have found that people are increasingly skeptical of AI-influenced news due to concerns that it may reinforce biases or manipulate public opinion. Overall, as AI becomes a larger part of everyday life, it may help in future elections but in its current form, it will hurt more than it will help.

References and Sources

Felix M. Simon, K. M. (2024, September 3). Ai’s impact on elections is being overblown. MIT Technology Review. https://www.technologyreview.com/2024/09/03/1103464/ai-impact-elections-overblown/

The illusion of choice: Uncovering electoral deceptions in the age of ai. Trend Micro (US). (n.d.). https://www.trendmicro.com/vinfo/us/security/news/cybercrime-and-digital-threats/the-illusion-of-choice-uncovering-electoral-deceptions-in-the-age-of-ai?utm_source=trendmicroresearch&utm_medium=smk&utm_campaign=092024_Elections&gad_source=1

Springs, A. (2024, February 8). Voters overwhelmingly believe in regulating deepfakes and the use of Artificial Intelligence. Data For Progress. https://www.dataforprogress.org/blog/2024/2/8/voters-overwhelmingly-believe-in-regulating-deepfakes-and-the-use-of-artificial-intelligence

Stahle, T. (2023, November 14). Can ai predict how you’ll vote in the next election? News. https://news.byu.edu/intellect/can-ai-predict-how-youll-vote-in-the-next-election-byu-study-says-yes

Using AI for political polling. Ash Center. (2024, June 10). https://ash.harvard.edu/articles/using-ai-for-political-polling/