How AI is Reshaping the Polling Business
As artificial intelligence works its way into the business of politics, some pollsters are turning to the emerging technology for new insights.
AI-assisted polling is still in its infancy. But a handful of companies are exploring ways to put it to work on everything from crafting survey questions to segmenting and analyzing massive amounts of data. In some cases, companies like OpenResearch, a nonprofit funded by OpenAI, are experimenting with AI chatbots as part of the polling process.
In a recent interview with Campaigns & Elections, Amir Kanpurwala, the co-founder of the quantitative research platform Outward Intelligence and an alum of The Harris Poll, called AI the next “evolution in how market research and polling” is conducted, arguing that the tech could be used to eliminate – or at least mitigate – survey bias and generate insights for clients in a fraction of the time that it would take traditional pollsters.
Of course, there are risks, Kanpurwala said. Bad actors could use AI bots and agents to skew data and erode trust in polling as an institution, he said. The polling business is also susceptible to the same pressures that AI is putting on other industries, and some people will likely lose their jobs as AI takes on a greater role, Kanpurwala said.
Yet he also argued that there are reasons to be excited about AI-assisted polling, noting that it comes at a cheaper cost and could help pollsters elicit fuller, more relevant answers from respondents.
The following interview has been edited for length and clarity.
C&E: Set the stage for us a little bit. Where are we at right now with AI and polling? What does that space look like?
Kanpurwala: We’re at this very, very nascent place with how AI is affecting the market research world. There’s the optimism of it and there’s the pessimism of it, and I’ll start with the pessimism.
Polling is incredibly important. It dictates how we talk about things and how we perceive the world around us. It’s used in understanding how other people think about issues and whether or not you’re on the right or wrong side of that issue. In politics, it helps us understand who’s ahead, and that can really help a campaign or hurt a campaign.
AI can throw a wrench in that to a certain extent. If you’re collecting data and some of that is human and some of that is coming from AI bots or agents, it can call into question a lot of the data that is being collected. And so, I think we are very early on from that pessimistic stage of how AI is cascading through the data that is being collected.
But I do think the AI piece can be extremely beneficial for every part of the process in how surveys are conducted and the information is disseminated. And people don’t understand how complex that process can be. So you start with a question: I want to better understand how voters perceive these different candidates or how consumers think about my company. But there’s many steps in how you go from that question to actually answering that question. You have to write a survey and you have to make sure those questions are worded properly. So we’re using AI to help create questions that are minimizing bias.
Then there’s the question of how does that survey get to people? At Outward Intelligence, we do all our surveys online. And that could be either on someone’s desktop or on a tablet or on a phone. But we need to be able to take that written survey and design that into something that’s programmed – that has logic behind it. We’re using our AI to go from here’s a list of questions to making sure they’re all connected logically.
We’re also using AI to collect data that’s extremely representative. If you look at some of the prior polling cycles for campaigns, people will say that the representativity is off – the amount of Democrats vs. Republicans in this study, or the number of liberals and conservatives, isn’t accurate… so we’re using AI to ensure that we’re getting highly representative data at every single segment level.
C&E: How do people respond to this idea of having an AI ask questions or craft questions or guide a conversation?
Kanpurwala: I think the goal for us is to make it as seamless and as intuitive as possible. So for our company, OI, when folks are taking surveys, they’re doing that in their browser, they’re doing that in their homes, on lunch breaks, during their commute. The onus is on us to make that survey-taking experience not feel antiquated, not feel like a drag – to make it feel actually kind of delightful to be probed about your thoughts in a way that makes you feel valued.
It’s also got to feel intuitive. So the questions are asked in a way that doesn’t seem repetitive or redundant or not misleading or biased. The AI really has to fade into the background and allow folks to just relay their thoughts. To that extent, it’s about navigating the right questions to the right respondents and not inundating them with the same repetitive topics.
In a political survey, you might have a bunch of different questions about favorability or different policies or ballot measures. We know the ZIP Code you’re in. We know, by matching your data to voting files, whether or not you participated in prior elections. And so the AI should be able to adjust all this information about the respondent and then, in real time, be able to ask questions that are relevant to them.
They should never know that there’s an AI that’s dictating this. All that should matter is that the questions that they’re answering are pertinent to them and what they care about.
C&E: To what extent is this technology going to impact the industry and the folks that work in it?
Kanpurwala: We talk to a lot of folks day in and day out – people that work at these companies – and you can see that there are job losses that are occurring in this space right now. And that’s because AI is making it easier than ever across every stage of the survey-making process. Anecdotally, I am seeing that there are job losses that are happening, unfortunately. This is a period of stage.
C&E: You mentioned earlier that there are reasons for optimism and reasons for pessimism when it comes to AI in polling. What are you most worried about with this technology and, on the flip side of that, what are you most encouraged by?
Kanpurwala: I’ll start with what worries me the most, and that is the ability for bad actors, both without a state-based agenda and with a state-based agenda, to affect the data that we’re seeing.
We have this thing, this reflection of the social fabric, that is polling. And we I think have faith, as a country, in that data, maybe not for individual polls, but at an aggregate level. So what happens when folks try to manipulate that? What can you believe? How can you be entirely sure that the data that you have is not being corrupted in some way or fashion?
Our company has built a bunch of resources and tools to be able to detect AI-based agents as they’re taking a survey. But it’s a cat-and-mouse game, and folks are going to invest more and more in this space. And it may not always be for some totally nefarious thing… but even if it’s not nefarious, it may have the same impact in so far as reducing the trust folks have in polling. Those are the things that keep me up at night.
As far as how I’m optimistic about AI, we’re able to provide this service – polling – at a much more accessible price point and much faster. I’ve done studies for clients in the past where they wanted to be able to see how perceptions of their companies are changing on a daily basis. And prior to the advent of AI and some of these modern tools, it could take as long as 90 days, which is an eternity, for corporations to be able to make a decision on stale data. And so, using these tools we’re able to provide people back their data, not in the span of 90 days, but in as little as one day.
I think AI can help allow decision makers to understand what’s happening and react to that in a very short amount of time and at a fraction of the price point that they’ve been paying, historically.
