Why AI-Generated Text Detection Fails: Evidence from Explainable AI Beyond Benchmark Accuracy

1. 1Investigate the performance of detection systems beyond benchmark accuracy
2. 2Analyze the interpretability of detection models using explainable AI techniques
3. 3Identify dataset-specific artefacts that may be exploited by detectors
4. 4Develop more robust detection systems that genuinely identify machine authorship

AI engineers and researchers benefit from understanding the limitations of current AI-generated text detection systems, as it informs the development of more robust and reliable detectors

💡 Current detection systems may not be as reliable as reported, and their performance may be inflated by exploiting dataset-specific artefacts

🚨 AI-generated text detection fails in real-world settings due to over-reliance on dataset artefacts 🚨