Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

J-Guard: Journalism Guided Adversarially Robust Detection of AI-generated News (2309.03164v1)

Published 6 Sep 2023 in cs.CL and cs.AI

Abstract: The rapid proliferation of AI-generated text online is profoundly reshaping the information landscape. Among various types of AI-generated text, AI-generated news presents a significant threat as it can be a prominent source of misinformation online. While several recent efforts have focused on detecting AI-generated text in general, these methods require enhanced reliability, given concerns about their vulnerability to simple adversarial attacks. Furthermore, due to the eccentricities of news writing, applying these detection methods for AI-generated news can produce false positives, potentially damaging the reputation of news organizations. To address these challenges, we leverage the expertise of an interdisciplinary team to develop a framework, J-Guard, capable of steering existing supervised AI text detectors for detecting AI-generated news while boosting adversarial robustness. By incorporating stylistic cues inspired by the unique journalistic attributes, J-Guard effectively distinguishes between real-world journalism and AI-generated news articles. Our experiments on news articles generated by a vast array of AI models, including ChatGPT (GPT3.5), demonstrate the effectiveness of J-Guard in enhancing detection capabilities while maintaining an average performance decrease of as low as 7% when faced with adversarial attacks.

The paper "J-Guard: Journalism Guided Adversarially Robust Detection of AI-generated News" addresses the critical issue of detecting AI-generated news, which poses a significant threat due to its potential to disseminate misinformation. Recognizing that existing AI text detectors often succumb to adversarial attacks and produce false positives—particularly in the context of the nuanced journalistic style—this work introduces a novel framework named J-Guard.

J-Guard is developed through an interdisciplinary approach, combining insights from both computer science and journalism to enhance the performance and robustness of AI-generated text detection methods specifically tailored for news articles. The framework focuses on incorporating journalistic stylistic cues, which are distinctive attributes observed in professional news writing, to train AI detectors more effectively. These stylistic cues are crucial in differentiating authentic journalistic content from AI-generated text.

The authors conducted extensive experiments involving news articles generated by a variety of AI models, including ChatGPT (GPT3.5). The findings demonstrate that J-Guard significantly enhances detection capabilities. One of the noteworthy achievements of this framework is its ability to maintain an approximate performance decrease of only 7% under adversarial conditions. This indicates that J-Guard offers a resilient solution immune to simple adversarial attacks without compromising on detection accuracy.

In summary, J-Guard presents a promising direction for the detection of AI-generated news by embedding journalistic principles into the detection process, ultimately safeguarding the credibility of news organizations and mitigating the spread of misinformation online.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (8)
  1. Tharindu Kumarage (21 papers)
  2. Amrita Bhattacharjee (24 papers)
  3. Djordje Padejski (1 paper)
  4. Kristy Roschke (1 paper)
  5. Dan Gillmor (1 paper)
  6. Scott Ruston (3 papers)
  7. Huan Liu (283 papers)
  8. Joshua Garland (35 papers)
Citations (9)