All the News That’s Fit to Fabricate: AI-Generated Text as a Tool of Media Misinformation

Abstract

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.

TruthfulQA: Measuring How Models Mimic Human Falsehoods

“…Evans et al ( 2021) refine the concept of truthfulness and draw distinctions between truthfulness and honesty. Truthfulness is relevant to many applications including generating news stories (Kreps et al, 2020;Zellers et al, 2019), summarization (Gabriel et al, 2021;Maynez et al, 2020;Stiennon et al, 2020;Wang et al, 2020), conversational dialog (Shuster et al, 2021;Roller et al, 2021), and question answering (Dou et al, 2021;Krishna et al, 2021;Logan IV et al, 2019). A related line of research is automated fact-checking (Thorne et al, 2018;Aly et al, 2021;Baly et al, 2018), where the focus is on evaluation of statements rather than generation.…”

Linguistic Markers of Inherently False AI Communication and Intentionally False Human Communication: Evidence from Hotel Reviews

“…Other work has indeed observed a positivity bias in AI "smart replies" compared to standard messaging (Mieczkowski et al, 2021). Despite these differences, humans still cannot distinguish AIgenerated text from human-generated text with great accuracy (E. Clark et al, 2021;Köbis & Mossink, 2021;Kreps et al, 2022).…”

Harnessing Artificial Intelligence for Health Message Generation: The Folic Acid Message Engine

“…Advances in natural language processing have made it increasingly possible to generate coherent messages [46]. Although enthusiasm and skepticism about using computers for text generation have waxed and waned for decades [47], the advances in the past decade have been particularly impressive as the quality of computer-generated texts is now at a level that makes it often indiscriminable from human-written text [48,49].…”

“…In parallel, we drew a random sample of 30 tweets from a pool of >10,000 real-life tweets. This strategy was chosen as it is not feasible to evaluate thousands of tweets and as it most likely mimics how practitioners would use such a system [49]. Thus, this procedure yielded 2 sets of 30 tweets each-30 AI-generated messages that came from a pool of 60 randomly drawn samples and 30 human-generated messages from Twitter.…”

“…Regarding broader implications and risks, recent events in the political domain have highlighted the danger of algorithmic bots deployed to create or spread misinformation [90][91][92][93]. Several malicious actors seem to be using natural language generators to produce fake or divisive messages; thus, several empirical studies have examined the dangers of using NLG technology to create content that is harmful to society [49,94]. The same problems arise concerning the marketing of products that might harm health or use bots to promote certain brands [95][96][97].…”