Academic

NOTAI.AI: Explainable Detection of Machine-Generated Text via Curvature and Feature Attribution

arXiv:2603.05617v1 Announce Type: new Abstract: We present NOTAI.AI, an explainable framework for machine-generated text detection that extends Fast-DetectGPT by integrating curvature-based signals with neural and stylometric features in a supervised setting. The system combines 17 interpretable features, including Conditional Probability Curvature, ModernBERT detector score, readability metrics, and stylometric cues, within a gradient-boosted tree (XGBoost) meta-classifier to determine whether a text is human- or AI-generated. Furthermore, NOTAI.AI applies Shapley Additive Explanations (SHAP) to provide both local and global feature-level attribution. These attributions are further translated into structured natural-language rationales through an LLM-based explanation layer, which enables user-facing interpretability. The system is deployed as an interactive web application that supports real-time analysis, visual feature inspection, and structured evidence presentation. A web interf

O
Oleksandr Marchenko Breneur, Adelaide Danilov, Aria Nourbakhsh, Salima Lamsiyah · · 1 min read · 7 views
#cs.CL

arXiv:2603.05617v1 Announce Type: new Abstract: We present NOTAI.AI, an explainable framework for machine-generated text detection that extends Fast-DetectGPT by integrating curvature-based signals with neural and stylometric features in a supervised setting. The system combines 17 interpretable features, including Conditional Probability Curvature, ModernBERT detector score, readability metrics, and stylometric cues, within a gradient-boosted tree (XGBoost) meta-classifier to determine whether a text is human- or AI-generated. Furthermore, NOTAI.AI applies Shapley Additive Explanations (SHAP) to provide both local and global feature-level attribution. These attributions are further translated into structured natural-language rationales through an LLM-based explanation layer, which enables user-facing interpretability. The system is deployed as an interactive web application that supports real-time analysis, visual feature inspection, and structured evidence presentation. A web interface allows users to input text and inspect how neural and statistical signals influence the final decision. The source code and demo video are publicly available to support reproducibility.

Executive Summary

NOTAI.AI introduces an innovative explainable framework for detecting machine-generated text by synergizing curvature-based signals with neural and stylometric features within a supervised XGBoost classifier. By integrating 17 interpretable features—including Conditional Probability Curvature, ModernBERT detector score, readability metrics, and stylometric cues—the system offers a layered detection mechanism. The inclusion of SHAP for feature attribution and LLM-based translation into natural-language rationales enhances user interpretability. The deployment as an interactive web application with real-time analysis and visual inspection capabilities further strengthens its practical applicability. The open-source nature and availability of a demo video support transparency and reproducibility.

Key Points

  • ▸ Integration of curvature-based signals with neural and stylometric features
  • ▸ Use of SHAP for feature attribution and LLM-based rationales for user interpretability
  • ▸ Deployment as an interactive web application for real-time analysis

Merits

Enhanced Transparency

NOTAI.AI’s SHAP-based attribution and natural-language rationales provide clear, user-friendly explanations, improving trust and usability.

Comprehensive Feature Integration

Combining 17 interpretable features across multiple domains (curvature, readability, stylometry) offers a more holistic detection model than traditional single-signal approaches.

Demerits

Complexity Overhead

The inclusion of multiple interpretable features and translation via LLM may introduce computational overhead or latency, particularly for real-time deployment.

Generalizability Concern

Performance may vary across domains or languages where stylometric or curvature patterns diverge from training data distributions.

Expert Commentary

NOTAI.AI represents a significant advancement in the field of AI text detection by bridging the gap between algorithmic accuracy and user interpretability. The fusion of curvature-based signals—a novel dimension in detection—with traditional neural and stylometric indicators demonstrates a sophisticated understanding of the multidimensional nature of machine-generated text. Moreover, the LLM-mediated translation of technical attributions into structured rationales marks a pivotal shift toward democratizing access to algorithmic reasoning, particularly for non-technical users. While computational efficiency concerns are valid, the trade-off between interpretability and performance is justified in contexts where accountability and user agency matter. This system sets a benchmark for future explainable detection frameworks and may inspire similar innovations across other domains requiring algorithmic transparency.

Recommendations

  • ✓ 1. Explore lightweight LLM alternatives or fine-tuned embedding models to mitigate computational overhead without compromising interpretability.
  • ✓ 2. Conduct comparative benchmarks across diverse language corpora and domain-specific datasets to validate generalizability and adapt feature weighting dynamically.

Sources

Related Articles