How AI Can Help Fix Polling

As wars go, America’s most recent one is quite unpopular: public support for President Donald Trump’s attack on Iran consistently hovers around 30 to 40 percent.

But when it comes to negotiations with Iran, polling offers very little guidance. In a CBS/YouGov poll, a majority of respondents indicated they would not accept a peace that left the Iranian regime in place. Another poll found two-thirds of Americans prioritized low gas prices—but the same proportion said it was important to prevent Iran from acquiring nuclear weapons.

Pollsters are used to such mixed signals. Yet the problem is not that American voters are somehow irrational or uninformed. The problem lies in the way we tend to measure public opinion. Individual beliefs on political questions are far more complicated than the answer to a simple yes-or-no question. Because our understanding of public opinion is often based on ambiguous survey results, policymakers make decisions without really knowing what their constituents think.

This is an inherent flaw of modern democracy. In the republic’s early days—an era shaped by small towns and close-knit communities—representatives could engage personally with many of the voters they represented. As president, Abraham Lincoln spent much of his time talking to voters, who lined the halls outside his White House office. But population growth, urbanization and the rise of a professional bureaucratic class fractured the bond between representatives and their constituents.

Modern polling emerged at least in part to fill this gap. Yet polls have always been an imperfect substitute for direct citizen engagement. Polling offers snapshots rather than narratives, but it rarely captures competing tradeoffs and priorities. Americans may rationally want both regime change and a quick end to the war, both lower gas prices and a denuclearized Iran.

To make matters worse, small shifts in wording or framing can produce wildly different polling results. For example, Americans are more likely to support “military action” than the “war” in Iran. This kind of ambiguity allows politicians like Trump to cherry-pick polling results and justify their actions after the fact.

This is where artificial intelligence comes in. AI will never replace human public opinion gathering, despite what proponents of “silicon sampling” may believe. But AI can help us synthesize complicated sets of data as reliably and much more efficiently than humans ever will.

In contrast to pollsters, AI can digest long-form, open-ended responses that reveal underlying narratives and map the complex tradeoffs Americans are willing to make. It can also scale processes of democratic deliberation and identify areas of majority consensus. There are already real-world examples of this. For over a decade, Taiwan has relied on an AI platform to surface citizen proposals and forge consensus on anything from taxi licensing to pension reform.

In fact, we already sit atop a treasure trove of real-time public opinion data—the internet.

Granted, social media gives us a distorted mirror of public opinion; it is notoriously vulnerable to manipulation and misinformation and tends to amplify viral, emotional and polarizing content. And yet, social media trends and narratives provide important insights on what the public thinks. Companies are using sentiment analysis for stock prediction and consumer trends, and the U.S. State Department reportedly monitors social media trends among foreign publics. It would be a mistake to ignore social media trends at home.

AI can also help us better understand, not just how Americans think about current events, but how they could react to future crises. Researchers frequently rely on war games and scenario planning exercises to map how U.S. policymakers would respond to future contingencies—from a crisis over the Taiwan Strait to Russian attacks on a NATO member. But such wargames almost always limit participation to elite actors; they rarely help us understand public sentiment. AI drastically lowers the cost of scenario planning. By opening access to representative members of the U.S. public, AI-enabled war games could help policymakers better anticipate public reactions to future crises.

Of course, public opinion polls must remain a core part of every policymaker’s toolkit. But rather than replace them, AI should be used to augment the insights of traditional surveys.

Reading a poll can feel a bit like watching a cloud of smoke from afar. We know that something is going on, but we don’t quite know what will happen next and why. AI can help us lift some of that fog. If left unchecked, AI will fuel misinformation and confusion. When used wisely, it can inform more prudent policymaking and rebuild a lost connection between representatives and their constituents.

In the absence of reliable public opinion data, foreign policy making is easily captured by elites and special interests that have direct access to elected officials. But Democrats and Republicans can only ignore their voters for so long. As both parties look to realign U.S. grand strategy with public preference, AI tools can help fix a broken link.

Johannes Lang researches AI and foreign policy at the Harvard Kennedy School. He is a John F. Kennedy Scholar and a McCain Global Leader.