Kai-Cheng Yang

Kai-Cheng Yang, Filippo Menczer

Large language models (LLMs) exhibit impressive capabilities in generating realistic text across diverse subjects. Concerns have been raised that they could be utilized to produce fake content with a deceptive intention, although evidence thus far remains anecdotal. This paper presents a case study about a Twitter botnet that appears to employ ChatGPT to generate human-like content. Through heuristics, we identify 1,140 accounts and validate them via manual annotation. These accounts form a dense cluster of fake personas that exhibit similar behaviors, including posting machine-generated content and stolen images, and engage with each other through replies and retweets. ChatGPT-generated content promotes suspicious websites and spreads harmful comments. While the accounts in the AI botnet can be detected through their coordination patterns, current state-of-the-art LLM content classifiers fail to discriminate between them and human accounts in the wild. These findings highlight the threats posed by AI-enabled social bots.

Matthew R. DeVerna, Harry Yaojun Yan, Kai-Cheng Yang et al.

Fact checking can be an effective strategy against misinformation, but its implementation at scale is impeded by the overwhelming volume of information online. Recent artificial intelligence (AI) language models have shown impressive ability in fact-checking tasks, but how humans interact with fact-checking information provided by these models is unclear. Here, we investigate the impact of fact-checking information generated by a popular large language model (LLM) on belief in, and sharing intent of, political news headlines in a preregistered randomized control experiment. Although the LLM accurately identifies most false headlines (90%), we find that this information does not significantly improve participants' ability to discern headline accuracy or share accurate news. In contrast, viewing human-generated fact checks enhances discernment in both cases. Subsequent analysis reveals that the AI fact-checker is harmful in specific cases: it decreases beliefs in true headlines that it mislabels as false and increases beliefs in false headlines that it is unsure about. On the positive side, AI fact-checking information increases the sharing intent for correctly labeled true headlines. When participants are given the option to view LLM fact checks and choose to do so, they are significantly more likely to share both true and false news but only more likely to believe false headlines. Our findings highlight an important source of potential harm stemming from AI applications and underscore the critical need for policies to prevent or mitigate such unintended consequences.

Kai-Cheng Yang, Filippo Menczer

Search engines increasingly leverage large language models (LLMs) to generate direct answers, and AI chatbots now access the Internet for fresh data. As information curators for billions of users, LLMs must assess the accuracy and reliability of different sources. This paper audits nine widely used LLMs from three leading providers -- OpenAI, Google, and Meta -- to evaluate their ability to discern credible and high-quality information sources from low-credibility ones. We find that while LLMs can rate most tested news outlets, larger models more frequently refuse to provide ratings due to insufficient information, whereas smaller models are more prone to making errors in their ratings. For sources where ratings are provided, LLMs exhibit a high level of agreement among themselves (average Spearman's $ρ= 0.79$), but their ratings align only moderately with human expert evaluations (average $ρ= 0.50$). Analyzing news sources with different political leanings in the US, we observe a liberal bias in credibility ratings yielded by all LLMs in default configurations. Additionally, assigning partisan roles to LLMs consistently induces strong politically congruent bias in their ratings. These findings have important implications for the use of LLMs in curating news and political information.

Stefano Cresci, Kai-Cheng Yang, Angelo Spognardi et al.

Research on social bots aims at advancing knowledge and providing solutions to one of the most debated forms of online manipulation. Yet, social bot research is plagued by widespread biases, hyped results, and misconceptions that set the stage for ambiguities, unrealistic expectations, and seemingly irreconcilable findings. Overcoming such issues is instrumental towards ensuring reliable solutions and reaffirming the validity of the scientific method. Here, we discuss a broad set of consequential methodological and conceptual issues that affect current social bots research, illustrating each with examples drawn from recent studies. More importantly, we demystify common misconceptions, addressing fundamental points on how social bots research is discussed. Our analysis surfaces the need to discuss research about online disinformation and manipulation in a rigorous, unbiased, and responsible way. This article bolsters such effort by identifying and refuting common fallacious arguments used by both proponents and opponents of social bots research, as well as providing directions toward sound methodologies for future research.

Jason Alexander, Priyal Nanda, Kai-Cheng Yang et al.

The proliferation of misleading visualizations online, particularly during critical events like public health crises and elections, poses a significant risk. This study investigates the capability of GPT-4 models (4V, 4o, and 4o mini) to detect misleading visualizations. Utilizing a dataset of tweet-visualization pairs containing various visual misleaders, we test these models under four experimental conditions with different levels of guidance. We show that GPT-4 models can detect misleading visualizations with moderate accuracy without prior training (naive zero-shot) and that performance notably improves when provided with definitions of misleaders (guided zero-shot). However, a single prompt engineering technique does not yield the best results for all misleader types. Specifically, providing the models with misleader definitions and examples (guided few-shot) proves more effective for reasoning misleaders, while guided zero-shot performs better for design misleaders. This study underscores the feasibility of using large vision-language models to detect visual misinformation and the importance of prompt engineering for optimized detection accuracy.

Matthew R. DeVerna, Harry Yaojun Yan, Kai-Cheng Yang et al.

Artificial intelligence is fundamentally changing how health content is encountered and acted upon across both the information and healthcare ecosystems. AI systems now generate claims, curate information, interpret symptoms, synthesize evidence, and guide decisions, with significant opportunities and risks for the public. Potential benefits include improvements in access, comprehension, and continuity of care. At the same time, AI can introduce inaccurate or manipulative content that is difficult to distinguish from reliable guidance, and encourage automated decisions that affect care with little transparency or recourse. We introduce a four-stage framework to examine how these opportunities and risks unfold as the public moves through the information environment and into formal healthcare.

Kai-Cheng Yang, Danishjeet Singh, Filippo Menczer

Recent advancements in generative artificial intelligence (AI) have raised concerns about their potential to create convincing fake social media accounts, but empirical evidence is lacking. In this paper, we present a systematic analysis of Twitter (X) accounts using human faces generated by Generative Adversarial Networks (GANs) for their profile pictures. We present a dataset of 1,420 such accounts and show that they are used to spread scams, spam, and amplify coordinated messages, among other inauthentic activities. Leveraging a feature of GAN-generated faces -- consistent eye placement -- and supplementing it with human annotation, we devise an effective method for identifying GAN-generated profiles in the wild. Applying this method to a random sample of active Twitter users, we estimate a lower bound for the prevalence of profiles using GAN-generated faces between 0.021% and 0.044% -- around 10K daily active accounts. These findings underscore the emerging threats posed by multimodal generative AI. We release the source code of our detection method and the data we collect to facilitate further investigation. Additionally, we provide practical heuristics to assist social media users in recognizing such accounts.

Kai-Cheng Yang

AI-powered search systems are emerging as new information gatekeepers, fundamentally transforming how users access news and information. Despite their growing influence, the citation patterns of these systems remain poorly understood. We address this gap by analyzing data from the AI Search Arena, a head-to-head evaluation platform for AI search systems. The dataset comprises over 24,000 conversations and 65,000 responses from models across three major providers: OpenAI, Perplexity, and Google. Among the over 366,000 citations embedded in these responses, 9% reference news sources. We find that while models from different providers cite distinct news sources, they exhibit shared patterns in citation behavior. News citations concentrate heavily among a small number of outlets and display a pronounced liberal bias, though low-credibility sources are rarely cited. User preference analysis reveals that neither the political leaning nor the quality of cited news sources significantly influences user satisfaction. These findings reveal significant challenges in current AI search systems and have important implications for their design and governance.

Matthew R. DeVerna, Kai-Cheng Yang, Harry Yaojun Yan et al.

Large language models (LLMs) have raised hopes for automated end-to-end fact-checking, but prior studies report mixed results. As mainstream chatbots increasingly ship with reasoning capabilities and web search tools -- and millions of users already rely on them for verification -- rigorous evaluation is urgent. We evaluate 15 recent LLMs from OpenAI, Google, Meta, and DeepSeek on more than 6,000 claims fact-checked by PolitiFact, comparing standard models with reasoning- and web-search variants. Standard models perform poorly, reasoning offers minimal benefits, and web search provides only moderate gains, despite fact-checks being available on the web. In contrast, a curated RAG system using PolitiFact summaries improved macro F1 by 233% on average across model variants. These findings suggest that giving models access to curated high-quality context is a promising path for automated fact-checking.

Zening Duan, Anqi Shao, Yicheng Hu et al.

While researchers often study message features like moral content in text, such as party manifestos and social media, their quantification remains a challenge. Conventional human coding struggles with scalability and intercoder reliability. While dictionary-based methods are cost-effective and computationally efficient, they often lack contextual sensitivity and are limited by the vocabularies developed for the original applications. In this paper, we present an approach to construct vec-tionary measurement tools that boost validated dictionaries with word embeddings through nonlinear optimization. By harnessing semantic relationships encoded by embeddings, vec-tionaries improve the measurement of message features from text, especially those in short format, by expanding the applicability of original vocabularies to other contexts. Importantly, a vec-tionary can produce additional metrics to capture the valence and ambivalence of a message feature beyond its strength in texts. Using moral content in tweets as a case study, we illustrate the steps to construct the moral foundations vec-tionary, showcasing its ability to process texts missed by conventional dictionaries and word embedding methods and to produce measurements better aligned with crowdsourced human assessments. Furthermore, additional metrics from the vec-tionary unveiled unique insights that facilitated predicting outcomes such as message retransmission.

Mohsen Sayyadiharikandeh, Onur Varol, Kai-Cheng Yang et al.

Malicious actors create inauthentic social media accounts controlled in part by algorithms, known as social bots, to disseminate misinformation and agitate online discussion. While researchers have developed sophisticated methods to detect abuse, novel bots with diverse behaviors evade detection. We show that different types of bots are characterized by different behavioral features. As a result, supervised learning techniques suffer severe performance deterioration when attempting to detect behaviors not observed in the training data. Moreover, tuning these models to recognize novel bots requires retraining with a significant amount of new annotations, which are expensive to obtain. To address these issues, we propose a new supervised learning method that trains classifiers specialized for each class of bots and combines their decisions through the maximum rule. The ensemble of specialized classifiers (ESC) can better generalize, leading to an average improvement of 56\% in F1 score for unseen accounts across datasets. Furthermore, novel bot behaviors are learned with fewer labeled examples during retraining. We deployed ESC in the newest version of Botometer, a popular tool to detect social bots in the wild, with a cross-validation AUC of 0.99.

Jisung Yoon, Kai-Cheng Yang, Woo-Sung Jung et al.

Graph embedding techniques, which learn low-dimensional representations of a graph, are achieving state-of-the-art performance in many graph mining tasks. Most existing embedding algorithms assign a single vector to each node, implicitly assuming that a single representation is enough to capture all characteristics of the node. However, across many domains, it is common to observe pervasively overlapping community structure, where most nodes belong to multiple communities, playing different roles depending on the contexts. Here, we propose persona2vec, a graph embedding framework that efficiently learns multiple representations of nodes based on their structural contexts. Using link prediction-based evaluation, we show that our framework is significantly faster than the existing state-of-the-art model while achieving better performance.

Kai-Cheng Yang, Onur Varol, Pik-Mai Hui et al.

Efficient and reliable social bot classification is crucial for detecting information manipulation on social media. Despite rapid development, state-of-the-art bot detection models still face generalization and scalability challenges, which greatly limit their applications. In this paper we propose a framework that uses minimal account metadata, enabling efficient analysis that scales up to handle the full stream of public tweets of Twitter in real time. To ensure model accuracy, we build a rich collection of labeled datasets for training and validation. We deploy a strict validation system so that model performance on unseen datasets is also optimized, in addition to traditional cross-validation. We find that strategically selecting a subset of training data yields better model accuracy and generalization than exhaustively training on all available data. Thanks to the simplicity of the proposed model, its logic can be interpreted to provide insights into social bot characteristics.