In 2017, Weitz-Shapiro and Winters conducted research on information credibility in Brazil, finding that citizens unilaterally punished corrupt politicians in the polls and could widely discern between credible and unreliable information, and further, that the “politically sophisticated” were more able to tell the difference. 2017 is now nearly a decade in the past. Since then, we have seen a global rise in social media use, elections where social media has been used to polarize and misinform voters, and the introduction of generative AI. It is that last development that I want to talk about here. I argue that even politically sophisticated citizens are losing their ability to discern credible sources from unreliable ones in light of AI, and that this upheaval will harm vertical accountability structures in democratic nations. 

Vertical accountability—the part of democratic government wherein elected officials are held responsible by those that vote for them—is an integral part of democracy. It rests on a few basic principles. First, there must be information about politician behavior and action available to the general public. Second, that information must be credible, meaning it comes from well-vetted sources. Third, citizens must be able to identify credible sources and distinguish them from unreliable or biased information. Moving forward, I discuss how these tenents will be harmed by the development of AI. 

First, younger and older generations alike are struggling to distinguish AI-generated content from reality. This is not simply an issue of education or age, but one of human perception and judgment. AI-generated images, videos, and text are becoming increasingly convincing, making it more difficult for the average person to determine whether something is authentic. As this technology continues to develop, it is reasonable to expect that fabricated sources and content will become even harder to distinguish from legitimate journalism. A fake story about corruption could be accepted as fact, while real reporting about corruption could be dismissed as “fake news”. In either case, citizens are left making political judgments (in how or even if they vote for instance) based on information that may not accurately reflect reality.

Second, younger generations, including well-educated and politically sophisticated citizens, are increasingly relying on AI to complete work and assist with critical thinking. The growing tendency to outsource research and analysis raises concerns about the development of independent critical thinking skills. Distinguishing credible information requires more than receiving an answer; it requires questioning where that information came from, considering alternative explanations, and evaluating whether the evidence supports the conclusion. If citizens increasingly rely on AI to perform these tasks for them, they risk becoming less capable of evaluating information independently. Political sophistication may still matter, but it may no longer provide the same protection against misinformation when the information environment itself becomes more difficult to navigate.

As Weitz-Shapiro and Winters argue, political accountability requires both the availability of credible information about politician behavior and the ability of citizens to identify that information as credible. The underlying principle is that governments are more likely to face consequences for their actions when citizens have an abundance of reliable information with which to evaluate their performance. However, an abundance of information does not automatically produce accountability. Citizens may have access to hundreds of articles, videos, and social media posts about a politician, but that access means very little if they cannot determine which sources are trustworthy. AI further dilutes an already crowded information environment, potentially making credible reporting harder to identify among fabricated stories, misleading claims, and politically motivated content. Citizens may punish politicians for actions they never committed, excuse corruption that has been credibly documented, or fail to recognize poor government performance altogether.

Some may argue that AI could actually make credibility easier to establish by allowing citizens to fact-check claims, evaluate sources, and identify misinformation. However, this assumes that AI systems are themselves reliable arbiters of credibility. Different systems are developed and trained differently, and their responses depend on the information available to them and the ways they are designed to operate. Asking AI whether a source is credible does not eliminate the need for critical thinking; it simply moves the problem to another source. Citizens must still evaluate whether an AI-generated answer is accurate, supported by evidence, and free from misleading assumptions. Accepting these answers without question could replace one unreliable source with another.

Others may argue that AI will create a new plurality of media-like sources, fulfilling the need for plentiful news coverage. The problem, however, is that multiple AI systems do not necessarily represent multiple independent sources of credible information. AI systems draw on existing information, meaning their outputs may reproduce or combine information from the same underlying sources. Furthermore, if citizens increasingly turn to AI instead of traditional news outlets, those outlets may lose the audiences and revenue necessary to sustain original reporting. With fewer journalists independently investigating corruption and documenting government behavior, AI systems could become increasingly dependent on information from social media, political campaigns, and other potentially biased sources. In this case, more does not necessarily mean better.

Ultimately, vertical accountability depends on citizens being able to make informed judgments about their elected officials. If credible information becomes harder to identify, citizens may struggle to determine whether their government is performing well, acting corruptly, or failing to meet its responsibilities. Worse, citizens might be intentionally misled by government actors. This does not mean AI will inevitably destroy vertical accountability, nor does every use of AI make citizens less informed. Rather, it raises a serious question about whether citizens can continue to distinguish credible information from unreliable information in an increasingly AI-saturated environment. The question is not whether citizens will have access to political information in the age of AI. They will. The question is whether they will know which information deserves their trust; which information is real. If they cannot make that distinction, their ability to hold elected officials accountable will weaken.

What does this mean for the policyscape? I have three recommendations for policymakers and government leaders today. First, the prioritization of the regulation of AI at an international level. This issue goes beyond partisanship and beyond the United States. It is an international tech race, one that can only be tampered with international action. Second, in those international conversations, we need to prioritize regulation and accountability for individual AI agents. In the current climate, individual teams and companies are not necessarily legally responsible for the things that their agents and bots espouse. If we continue down this path, there will be little if any urgency to fix issues that the general public is unaware of (like misinformation). Third, I highly recommend that we prioritize our local newspapers and the pursuit of journalism, alongside a bolstering of the reading comprehension programs in schools across the country. 

In the end, there are many ways to handle this issue, and I know that what I have proffered is not exhaustive. What is most important now is that we act, that we move forward in some substantive capacity to protect vertical accountability.