AI Text Detection via Stylometric Redundancy Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to accurately distinguish between human-authored texts and those generated by generative artificial intelligence (AI) conversational agents, particularly in cases where AI assistance is used in human writing.
Innovation Solution
A computer-implemented method that combines stylometric features and redundancy analysis to differentiate between human and AI-generated texts, using Bayes factors to evaluate the evidence and determine the likelihood of human or AI authorship.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stylometric features are used to detect AI-generated texts, then detection capability is improved, but reliability deteriorates because AI agents can be instructed to apply specific styles making discrimination difficult
Solution Approach 1:
The patent segments the text analysis into multiple independent stylometric features (word length, sentence length, paragraph length, N-gram frequencies, punctuation patterns) rather than relying on a single feature. This segmentation allows the system to capture diverse aspects of writing style that are harder for AI to uniformly replicate across all dimensions.
Solution Approach 2:
The patent transforms the detection approach by changing from analyzing individual stylometric features to analyzing the joint distribution and correlations of multiple features simultaneously. By examining how features co-occur and relate to each other (e.g., the relationship between sentence length and N-gram patterns), the system detects subtle inconsistencies in AI-generated texts that differ from natural human writing patterns.
2Productivity
If AI conversational agents are used to generate texts, then productivity is improved, but harmful factors increase due to potential fraudulent uses such as fake news and anonymous criminal letters
Solution Approach 1:
The patent implements a feedback mechanism by providing probabilistic authorship attribution results that can be used to flag potentially AI-generated or fraudulent content. This feedback loop enables users to verify and validate text authenticity, allowing them to take appropriate actions against suspicious content while maintaining the benefits of AI-assisted writing for legitimate purposes.
Solution Approach 2:
The patent introduces an intermediary verification layer between text generation and publication/distribution. The stylometric analysis system acts as a mediator that assesses text authenticity before content reaches its final destination, enabling platforms and users to filter or flag potentially harmful AI-generated content without completely blocking productive AI writing assistance.
3Ease of operation
If generative AI is used to create documents, then ease of operation is improved, but loss of information increases because AI agents understand neither the meaning of training texts nor the meaning of messages they produce
Solution Approach 1:
The patent performs preliminary stylometric analysis on reference texts with known human authorship before comparing them to test texts. By establishing baseline stylometric profiles from verified human writings in advance, the system creates a reference framework that helps identify when a text lacks the nuanced semantic understanding characteristic of human authorship, thereby detecting AI-generated content that may have been created solely for ease of operation.
Data Source
Figure 1~2
Figure 3~4a
Figure 4b~4c
AI summary
A computer implemented method for determining if a questioned text has been produced by a human or by a generative artificial intelligence based conversational agent, comprising the steps of: retrieving from the test text a feature (y) representing the redundancy of the test text; using this feature for determining if the test text has been produced by a human or by a generative artificial intelligence based conversational agent.