The political world is venturing into the next frontier: artificial intelligence-enhanced polling.
A number of Washington heavyweights have recently leaned into AI to help them gauge public opinion. D.C. public affairs shops are developing in-house tools to help clients analyze polling data or test messaging. And campaigns are flirting with virtual “focus groups” comprised of bots.
Club for Growth president David McIntosh said his group has even used trained AI agents to “poll” battleground districts.
“You’re creating agents that would reflect the views of a population, like in a congressional district or a state, and then you can poll those agents: Who would you be more likely to vote for? What do you think about these issues?” McIntosh, whose group has invested six figures into exploring AI-enhanced polling, told reporters recently. “It looks promising.”
This approach — dubbed “silicon sampling” — would allow pollsters to use models like ChatGPT to imitate voters based on demographic data, creating voters’ “digital twins” and eliminating the need for people to answer the phone or fill out online surveys.
In the long run, proponents argue, these methods could save a lot of time and money, either as a replacement of traditional polling or a companion of it. But the move toward AI-infused polling has already drawn ire for its potential inaccuracies and its inability to be, well, human.
Last month, Axios was criticized for an article that stated “a majority of people trust their own doctors and nurses,” without clarifying the finding was based on AI simulation research. (Axios issued an editor’s note correcting the error.) Last week, a pair of academics launched a broadside against silicon sampling in a New York Times guest essay: “If we do not slam the brakes … we could see a significant undermining of trust in public opinion work and social science research more broadly,” they wrote.
Data guru Nate Silver argued that silicon sampling will make traditional polling even more important, since it will up the need for collecting original data. One partner at a major Republican firm in Washington, who spoke on the condition of anonymity, said his company recently installed a policy declaring AI will have no place in drawing a sample, creating a questionnaire or conducting interviews.
Still, Bradley Honan, a longtime Democratic pollster who has worked for the Clintons and DNC, said targeted AI use could transform an aging industry. His firm, Honan Strategy Group, is experimenting with AI systems conducting phone interviews of humans, which makes the process cheaper and faster.
“If I can get you 250 interviews done by tomorrow, that is where I think the power of AI should be focused,” Honan said.
Others are taking more aggressive approaches. Recently, a donor approached Club for Growth with a time-sensitive campaign query. Tom Schultz, the vice president for campaigns, reached out to Aaru, an AI simulation firm, with a questionnaire and instructions for the target demographic. Within six hours, Aaru had results in hand from a survey of 10,000 AI agents.
“I don’t pitch it as a replacement for polling,” Schultz said. “Rather, it’s a supplement in our toolbox of things we can use for more immediate answers, and for more flexible creative testing.”
Aaru’s strategy, a company spokesperson said, is “anchoring our simulations in real-world behavior” that reflects “how populations actually behave, adapt, and evolve.”
Think Big, a DC-based comms shop, argues that AI can help clients make crisis comms decisions, backfilled with polling data, online chatter, news reports or other data relevant to the target demographic.
“We’re leveraging these … AI tools, so you can act fast, act more quickly, better summarize the data that’s coming in and cross reference it against what your candidate or the company or the organization you’re representing is really focused on,” said Lewis Muller, CEO of Think Big.
But even though AI can help synthesize existing public opinion, there’s no accounting for “an exogenous shock to the system,” argued Democratic pollster Anna Greenberg, a senior partner at GQR.
“What happens if a 9/11 happens? ” she said. “Things happen, and it changes how people think. And you don’t always know how it’s going to affect how people think.”
Politico: The AI revolution coming for campaigns
The political world is venturing into the next frontier: artificial intelligence-enhanced polling.
A number of Washington heavyweights have recently leaned into AI to help them gauge public opinion. D.C. public affairs shops are developing in-house tools to help clients analyze polling data or test messaging. And campaigns are flirting with virtual “focus groups” comprised of bots.
Club for Growth president David McIntosh said his group has even used trained AI agents to “poll” battleground districts.
“You’re creating agents that would reflect the views of a population, like in a congressional district or a state, and then you can poll those agents: Who would you be more likely to vote for? What do you think about these issues?” McIntosh, whose group has invested six figures into exploring AI-enhanced polling, told reporters recently. “It looks promising.”
This approach — dubbed “silicon sampling” — would allow pollsters to use models like ChatGPT to imitate voters based on demographic data, creating voters’ “digital twins” and eliminating the need for people to answer the phone or fill out online surveys.
In the long run, proponents argue, these methods could save a lot of time and money, either as a replacement of traditional polling or a companion of it. But the move toward AI-infused polling has already drawn ire for its potential inaccuracies and its inability to be, well, human.
Last month, Axios was criticized for an article that stated “a majority of people trust their own doctors and nurses,” without clarifying the finding was based on AI simulation research. (Axios issued an editor’s note correcting the error.) Last week, a pair of academics launched a broadside against silicon sampling in a New York Times guest essay: “If we do not slam the brakes … we could see a significant undermining of trust in public opinion work and social science research more broadly,” they wrote.
Data guru Nate Silver argued that silicon sampling will make traditional polling even more important, since it will up the need for collecting original data. One partner at a major Republican firm in Washington, who spoke on the condition of anonymity, said his company recently installed a policy declaring AI will have no place in drawing a sample, creating a questionnaire or conducting interviews.
Still, Bradley Honan, a longtime Democratic pollster who has worked for the Clintons and DNC, said targeted AI use could transform an aging industry. His firm, Honan Strategy Group, is experimenting with AI systems conducting phone interviews of humans, which makes the process cheaper and faster.
“If I can get you 250 interviews done by tomorrow, that is where I think the power of AI should be focused,” Honan said.
Others are taking more aggressive approaches. Recently, a donor approached Club for Growth with a time-sensitive campaign query. Tom Schultz, the vice president for campaigns, reached out to Aaru, an AI simulation firm, with a questionnaire and instructions for the target demographic. Within six hours, Aaru had results in hand from a survey of 10,000 AI agents.
“I don’t pitch it as a replacement for polling,” Schultz said. “Rather, it’s a supplement in our toolbox of things we can use for more immediate answers, and for more flexible creative testing.”
Aaru’s strategy, a company spokesperson said, is “anchoring our simulations in real-world behavior” that reflects “how populations actually behave, adapt, and evolve.”
Think Big, a DC-based comms shop, argues that AI can help clients make crisis comms decisions, backfilled with polling data, online chatter, news reports or other data relevant to the target demographic.
“We’re leveraging these … AI tools, so you can act fast, act more quickly, better summarize the data that’s coming in and cross reference it against what your candidate or the company or the organization you’re representing is really focused on,” said Lewis Muller, CEO of Think Big.
But even though AI can help synthesize existing public opinion, there’s no accounting for “an exogenous shock to the system,” argued Democratic pollster Anna Greenberg, a senior partner at GQR.
“What happens if a 9/11 happens? ” she said. “Things happen, and it changes how people think. And you don’t always know how it’s going to affect how people think.”
Recent Post
Archives