Summary of Esperanto: Evaluating Synthesized Phrases to Enhance Robustness in Ai Detection For Text Origination, by Navid Ayoobi et al.
ESPERANTO: Evaluating Synthesized Phrases to Enhance Robustness in AI Detection for Text Origination
by Navid Ayoobi, Lily Knab, Wen Cheng, David Pantoja, Hamidreza Alikhani, Sylvain Flamant, Jin Kim, Arjun Mukherjee
First submitted to arxiv on: 22 Sep 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a novel technique called back-translation to evade AI-generated text detection systems, highlighting the vulnerability of current methods to manipulation. The approach involves translating AI-generated text into multiple languages before back-translating it to English, creating a manipulated version that retains original semantics while reducing true positive rates (TPRs) of existing detectors. The authors evaluate this technique on nine AI detectors, including open-source and proprietary systems, demonstrating their susceptibility to back-translation manipulation. To improve robustness against this form of manipulation, the paper presents a countermeasure. Additionally, the authors create a large dataset containing 720k texts from eight different large language models (LLMs), including both human-authored and LLM-generated texts in various domains and writing styles. This dataset is publicly shared to benefit the research community. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper shows how big language models can be used for good or bad things, like cheating on tests or spreading false information. To stop this from happening, researchers created special tools to detect AI-written text. But these tools are not perfect and can be tricked into thinking fake text is real. The new method, called back-translation, makes it even harder for these detection tools to tell the difference between real and fake text. The authors tested their method on nine different detectors and showed that they all struggled with the new way of manipulating the AI-written text. |
Keywords
» Artificial intelligence » Semantics » Translation
