Online misinformation has become a constant; only the way actors create and distribute that information is changing. Advances in artificial intelligence (AI) such as GPT-2 mean that actors can now synthetically generate text in ways that mimic the style and substance of human-created news stories. We carried out three original experiments to study whether these AI-generated texts are credible and can influence opinions on foreign policy. The first evaluated human perceptions of AI-generated text relative to an original story. The second investigated the interaction between partisanship and AI-generated news. The third examined the distributions of perceived credibility across different AI model sizes. We find that individuals are largely incapable of distinguishing between AI- and human-generated text; partisanship affects the perceived credibility of the story; and exposure to the text does little to change individuals’ policy views. The findings have important implications in understanding AI in online misinformation campaigns.
While mass vaccination has blunted the pandemic in the United States, pockets of vaccine hesitancy remain. Through a nationally representative survey of 1027 adult Americans conducted in February 2021, this study examined individual misconceptions about COVID-19 vaccine safety; the demographic factors associated with these misconceptions; and the relationship between misconceptions and willingness to vaccinate. Misconceptions about vaccine safety were widespread. A sizeable minority (40%) believed that vaccine side effects are commonly severe or somewhat severe; 85% significantly underestimated the size and scale of the clinical trials; and a sizeable share believed either that the vaccines contain live coronavirus (10%) or were unsure (38%), a proxy for fears that vaccination itself may cause infection. These misconceptions were particularly acute among Republicans, Blacks, individuals with lower levels of educational attainment, and unvaccinated individuals. Perceived side effect severity and underestimating the size of the clinical trials were both significantly associated with vaccine hesitancy.
1 August 2013 in American Political Science Review
What determines when states adopt war taxes to finance the cost of conflict? We address this question with a study of war taxes in the United States between 1789 and 2010. Using logit estimation of the determinants of war taxes, an analysis of roll-call votes on war tax legislation, and a historical case study of the Civil War, we provide evidence that partisan fiscal differences account for whether the United States finances its conflicts through war taxes or opts for alternatives such as borrowing or expanding the money supply. Because the fiscal policies implemented to raise the revenues for war have considerable and often enduring redistributive impacts, war finance—in particular, war taxation—becomes a high-stakes political opportunity to advance the fiscal interests of core constituencies. Insofar as the alternatives to taxation shroud the actual costs of war, the findings have important implications for democratic accountability and the conduct of conflict.
Traditionally, writing assistance systems have focused on short or even single-word suggestions. Recently, large language models like GPT-3 have made it possible to generate significantly longer natural-sounding suggestions, offering more advanced assistance opportunities. This study explores the trade-offs between sentence- vs. message-level suggestions for AI-mediated communication. We recruited 120 participants to act as staffers from legislators’ offices who often need to respond to large volumes of constituent concerns. Participants were asked to reply to emails with different types of assistance. The results show that participants receiving message-level suggestions responded faster and were more satisfied with the experience, as they mainly edited the suggested drafts. In addition, the texts they wrote were evaluated as more helpful by others. In comparison, participants receiving sentence-level assistance retained a higher sense of agency, but took longer for the task as they needed to plan the flow of their responses and decide when to use suggestions. Our findings have implications for designing task-appropriate communication assistance systems.
Advances in machine learning have led to the creation natural language models that can mimic human writing style and substance. Here we investigate the challenge that machine-generated content, such as that produced by the model GPT-3, presents to democratic representation by assessing the extent to which machine-generated content can pass as constituent sentiment. We conduct a field experiment in which we send both handwritten and machine-generated letters (a total of 32,398 emails) to 7132 state legislators. We compare legislative response rates for the human versus machine-generated constituency letters to gauge whether language models can approximate inauthentic constituency voices at scale. Legislators were only slightly less likely to respond to artificial intelligence (AI)-generated content than to human-written emails; the 2% difference in response rate was statistically significant but substantively small. Qualitative evidence sheds light on the potential perils that this technology presents for democratic representation, but also suggests potential techniques that legislators might employ to guard against misuses of language models.