Software:UndetectableAI

From HandWiki

UndetectableAI is a software developed based on the concept of artificial intelligence content detection. Its of primary function is to modify AI-generated text to reduce recognizability by software designed to detect such content.[1][2]

History

The UndetectableAI software was designed by Bars Juhasz, a PhD student from Loughborough University, who previously worked alongside the Royal Air Force to research unmanned aircraft system operations in denied command and control environments. The online deployment of UndetectableAI co-developed by Christian Perry and Devan Leos, was officially released in May 2023.[3][4]

Reception and analysis

UndetectableAI has been the subject of discussion within the technology and academic communities. Articles in mainstream technology news outlets such as Mashable,[5] and TechTudo[6] have discussed the use and ethical implications of the software.[7]

Experts in AI and machine learning have weighed in on the potential impact of such software on the integrity of academia and online content.[8][9][10]

Academic concerns

In July of 2023, in a peer reviewed research paper, titled "Modern threats in academia: evaluating plagiarism and artificial intelligence detection scores of ChatGPT," researchers from Magna Græcia University (Andrea Taloni et. al) tested UndetectableAI against generative-text and plagiarism detection softwares.[11]

The findings concluded that while the detection softwares used were 95% accurate in detecting standard instances of AI-generated scientific texts (specifically those generated by GPT-4), when processing AI-generated and plagiarized texts through UndetectableAI they became significantly harder to detect. The research paper suggested that the functionality of UndetectableAI demonstrated the limitations of detecting text generated by LLM's, and described its functionality as having ability to enable "malicious attempts to circumvent AI detection."

On November 4, 2023, Erik Piller, an academic at Nicholls State University, published a paper titled "The Ethics of (Non)disclosure: Large Language Models in Professional, Nonacademic Writing Contexts," addressing ethical concerns in artificial intelligence. In the paper, Piller critically examined the UndetectableAI software, questioning the moral foundation and the underlying intention of its deployment in various contexts, while expressing skepticism of the software having a positive application.[12]

As of January 18th 2024, Johns Hopkins University lists UndetectableAI as a known challenge to detecting AI-generated text.[13][14][15]

Potential to affect data quality

On August 14, 2023 Dr. Christoph Bartneck, an associate professor at the University of Canterbury, published a joint research paper (Bartneck et. al), titled "Detecting The Corruption Of Online Questionnaires By Artificial Intelligence", which investigated the challenges posed by Undetectable.AI to data quality control in online questionnaires.

The paper noted that the UndetectableAI software was ability to bypass conventional AI detection systems, and how it could raise concerns about the integrity of data collected from online studies. The study found that while AI detection systems demonstrated the ability to identify ChatGPT-generated text, they failed to identify text obfuscated by Undetectable.AI; however the paper concluded that human judgment was ultimately more successful in distinguishing between human and AI-generated content.[16][17]

Cultural impact

On November 30th, 2023 EarthWeb used UndetectableAI's content analysis function alongside GPTZero to scan text of apologies posted by celebrities, some of whom got accused of writing their apologies with AI, based on the results.[18]

On January 15th 2024, an article published to Life & Style (magazine) drew parallels between the perception of using AI, versus actually having actually used it. The article proposed that even when used ethically, the "stigma and perception" of using AI could serve as a possible motivation for obfuscation. The article used UndetectableAI's traffic volume (sourced from Similarweb), while citing recent celebrity scandals involving accusations of using AI, alongside the aspect of ChatGPT becoming widespread to justify this opinion.[19]

Mechanism

In machine learning, the main function of UndetectableAI is ontologically adversarial, but the exact details of how UndetectableAI's functions is not publicly known. However, observations made in the research published by Taloni et. al,[11] documented that while the software slightly increased readability of AI-generated texts, a reduction in the correctness of grammatical structure was noted (such as by omitting commas or hyphens).[20]

References

  1. ↑ "Using Undetectable AI for Bypassing AI Detection - Grit Daily News" (in en-US). 2023-07-11. https://gritdaily.com/using-undetectable-ai-for-bypassing-ai-detection/. 
  2. ↑ Staff, Knewz (2023-11-01). "New AI Mimics Real Writing — 'No One Can Tell'" (in en-US). https://knewz.com/new-ai-mimics-real-writing-no-one-can-tell/. 
  3. ↑ Staff, OK! (2023-11-28). "Researcher Working With Royal Air Force Created an 'Undetectable' AI" (in en-US). https://okmagazine.com/p/researcher-working-royal-air-force-created-undetectable-ai/. 
  4. ↑ "Undetectable AI helps emulate 'human' side to AI" (in en-US). 2024-01-18. https://www.kget.com/video/undetectable-ai-helps-emulate-%E2%80%98human%E2%80%99-side-to-ai/8670091/. 
  5. ↑ Staff, Editorial (September 25, 2023). "Undetectable AI Now Supports 50+ Languages (Including Norwegian)". Mashable. https://nl.mashable.com/artificial-intelligence-1/9117/undetectable-ai-now-supports-50-languages-including-norwegian. 
  6. ↑ "Undetectable.AI: como saber se um texto foi escrito pelo ChatGPT" (in pt-br). 2023-10-10. https://www.techtudo.com.br/dicas-e-tutoriais/2023/10/undetectableai-como-saber-se-um-texto-foi-escrito-pelo-chatgpt-edsoftwares.ghtml. 
  7. ↑ Staff, Radar (2023-11-10). "Alan From Mighty Med Condemns AI Cheats — Then Explains How To Cheat With AI" (in en-US). https://radaronline.com/p/alan-from-mighty-med-condemns-ai-cheats-then-explains-how-to-cheat-with-ai/. 
  8. ↑ "Many AI researchers think fakes will become undetectable". The Economist. ISSN 0013-0613. https://www.economist.com/science-and-technology/2024/01/17/many-ai-researchers-think-fakes-will-become-undetectable. 
  9. ↑ Knibbs, Kate. "Researchers Tested AI Watermarks—and Broke All of Them" (in en-US). Wired. ISSN 1059-1028. https://www.wired.com/story/artificial-intelligence-watermarking-issues/. Retrieved 2024-01-18. 
  10. ↑ Steere, Elizabeth. "The Trouble With AI Writing Detection" (in en). https://www.insidehighered.com/opinion/career-advice/teaching/2023/10/18/faculty-should-know-tools-students-use-beat-ai-detection. 
  11. ↑ 11.0 11.1 Taloni, Andrea; Scorcia, Vincenzo; Giannaccare, Giuseppe (2023-08-02). "Modern threats in academia: evaluating plagiarism and artificial intelligence detection scores of ChatGPT" (in en). Eye: 1–4. doi:10.1038/s41433-023-02678-7. ISSN 1476-5454. PMID 37532832. https://www.nature.com/articles/s41433-023-02678-7. 
  12. ↑ Piller, Erik (November 4, 2023). "The Ethics of Nondisclosure". Rupkatha Journal. ISSN 0975-2935. https://rupkatha.com/V15/n4/v15n402.pdf. 
  13. ↑ Bardard, Neil (2024-01-17). "Promise or Reality? An era of Undetectable AI" (in en-US). https://capitolhilltimes.com/2024/01/17/promise-or-reality-an-era-of-undetectable-ai/. 
  14. ↑ "Generative AI Tools – Johns Hopkins Engineering CLDT" (in en-US). https://engineering.jhu.edu/cldt/chatgpt/ai-tools/. 
  15. ↑ "AI Resources – Johns Hopkins Engineering CLDT" (in en-US). https://engineering.jhu.edu/cldt/chatgpt/ai-resources/. 
  16. ↑ Lebrun, Benjamin; Temtsin, Sharon; Vonasch, Andrew; Bartneck, Christoph (2023-08-14). "Detecting The Corruption Of Online Questionnaires By Artificial Intelligence". arXiv:2308.07499v1 [cs.HC].
  17. ↑ Detecting The Corruption Of Online Questionnaires By Artificial Intelligence - Benjamin Lebrun, Sharon Temtsin, Andrew Vonasch, Christoph Bartneck [PDF]
  18. ↑ Buckler, Nicole (2023-11-20). "Top 10 Celeb Apologies Accused of Being Written by AI: Matty Healy, Doja Cat, Joe Rogan, and Elon Musk" (in en-AU). https://thechainsaw.com/business/ai-detectors-used-to-accuse-celebs-of-faking-their-apologies/. 
  19. ↑ Stojan, Jon (2024-01-15). "ChatGPT Craze Fueled Public Skepticism and Celeb Scandals" (in en-US). https://www.lifeandstylemag.com/posts/chatgpt-craze-fueled-public-skepticism-and-celeb-scandals/. 
  20. ↑ Bardard, Neil (2024-01-17). "Promise or Reality? An era of Undetectable AI" (in en-US). https://capitolhilltimes.com/2024/01/17/promise-or-reality-an-era-of-undetectable-ai/. 
This page will be placed in the following categories if it is moved to the Article namespace.
Categories:


Retrieved from "https://handwiki.org/wiki/index.php?title=Software:UndetectableAI&oldid=3214446"