Political polls have long been essential in shaping public opinion and guiding election strategies. They predict outcomes, help politicians understand voter concerns, and inform the public. However, AI has increasingly influenced political polling, particularly in the 2024 presidential election. By rapidly processing vast datasets, AI can identify patterns in voter behavior, analyze social media engagement, and enhance polling accuracy. Yet, AI-driven polling is highly dependent on data quality and context. Without sufficient or unbiased data, results can be skewed, potentially misleading political strategies and public perception.
Jared Sharpe, a visiting assistant professor of accounting and MIS at the University of Delaware’s Alfred Lerner College of Business and Economics, highlights the growing role of Generative AI in political polling and media. As AI technology advances, he predicts a shift toward more dynamic and adaptive content, which could transform not only polling methodologies but also the way political narratives are shaped and disseminated. This evolution underscores the need for transparency and responsible AI integration to ensure accurate and ethical political analysis.
Lerner: What are the pros and cons of polling, in your opinion?
Sharpe: In general, polling is a necessary part of most business practices. It is certainly a better way of understanding what your clientele may be interested in, rather than making a blind decision or assuming the past will repeat itself in perpetuity. That said, the age-old conundrum is that you only get responses from people who enjoy or even will respond to polls, which creates a great deal of bias in the result.
Without getting into too much of the politics of it, political polls are especially confounding because the results of previous polls are often well-known to future potential poll responders and may influence their decision to respond to the next one they encounter.
Lerner: How do you think AI will play a role in the future of polls?
Sharpe:I think Generative AI will continue to play a role in all forms of media: audio, visual, and textual. How that finds its way into elections is hard to say specifically – that will depend on the intentions of the entity deploying it. While it is true that the ability to quickly create any kind of content autonomously has greatly improved both in quality and quantity, we have had spam for a long time, and on the whole, people pay little mind to it.
I am skeptical that we will see a large increase in false news stories or deep-faked images. I think we will see more targeted content at specific issues to specific people in their most consumed form (text, images, short videos, etc.) about issues that are linked to them somehow; parents may see political content about education, for example. I think the content will be more dynamic as well; we won’t be seeing the same content over and over again, but small variations of it: different phrasing, different colors, different voice-overs, scripts, etc., simply because it’s easy to create with Generative AI and more likely to hold onto people’s attention if it’s something they haven’t seen before.
Lerner: In your opinion, what are the advantages and disadvantages of using AI in regards to polling and elections?
Sharpe:Technically speaking, general Artificial Intelligence is already at work in some ways, such as automating information transfer tasks, image recognition, digitizing text, and the like, but this is not new. I think the automation of these mundane tasks are a great positive of AI in the entire election process and business in general.
That said, much like how people have formed an opinion in their minds about a candidate or an issue, a political party, news station, their neighbor, and the driver of the car in front of them on 95, I think people will or already have formed an opinion about both AI and AI-generated content. This opinion is perhaps another hurdle for ‘pros’ of generative AI in polling and elections. While generative AI makes it easier to get surveys and polling in front of a more diverse set of people (people with ready access to the internet anyway), responders will likely be biased by their opinion of the technology. Is that any worse than how it is/was before? Hard to say.
Lerner: How do data and analytics play a role in interpreting polls?
Sharpe: Interpreting one poll is not that difficult, but it is also not very informative. Taking the results of several polls and making them informative is more difficult. Many of the methods that I’ve seen for extrapolating or attempts to remove response-count issues from polls rely on similar assumptions to the original poll: the measured correction comes from a known distribution in the polled population, such as the percentage of people in an age bracket and the percentage of the poll respondents that were in that bracket. As you can imagine, it is very easy to over-emphasize the responses from a few hundred people to a few million.
Even if the poll is magically perfectly corrected for the bias of those who respond to polls, taking the result of the poll to the election adds the additional assumption that those who responded to the poll are representative of those who will cast votes. And that’s just for one election year. In analytics, we tend to like to use as much data as we have access to, but bringing data from previous election years forward adds more assumptions or the need for more data. We now need to know how each of these measurements has changed over time. This is not an easy task. To make it more difficult, we are not measuring predictable, physical changes in the world: we’re polling people. Every two years, everyone is two years older, and in 3-4 presidential elections, 12-16 years, everyone is in a different stage of their life – some of their political leanings may have changed, their likelihood of going to vote may have changed, or they may have moved.
Lerner: Do you think analyzing polling data is more important in certain cases, such as for this most recent presidential election? Why or why not?
Sharpe: I think analyzing polling data is important in all years, but I also think we absolutely need to continue to advance how we conduct our analysis and how its results are presented. Polling analytics needs to keep up with technology in both execution and analysis.
Lerner: How do you think data and analytics will develop in the future? How will this affect polling for future elections?
Sharpe: I think we will see a lot more automated polling; I would not be surprised if some entity is able to better monetize the process of collecting responses in a way that reaches more people, especially as a larger percentage of the population has lived with computers and the internet for most of their lives.