Semantic Effects of AI Humanizers: Evaluation and Improvement Strategies
When AI rewrites, does the meaning stays the same?
We explore how AI humanizers change the meaning of text, compare popular tools, and design a new approach that keeps writing natural while preserving its original intent.
Why Study AI Humanizers?
AI humanizers are tools that rephrase AI-generated text to make it sound more natural and evade detection. While widely used, their impact on meaning is often overlooked. Humanization can introduce semantic drift, where the original intent of the text subtly shifts. This raises concerns for contexts like academic writing, research, and journalism, where preserving meaning is critical.
Our Objectives
Evaluate Meaning Preservation
Benchmark commercial AI humanizers and measure how well they maintain semantic fidelity.
Test Across Conditions
Analyze performance across different tones, lengths, and text structures.
Rank Current Tools
Compare and highlight strengths and weaknesses of leading humanizers.
Develop a Prototype
Build a semantic-first humanizer that prioritizes meaning while staying undetectable.
About Us
Charaspreedalarp Warit
BEng CompSc (3036085202)
warit.cha@connect.hku.hk
Karina Nathalie
BEng CompSC (3036037267)
karinath@connect.hku.hk
Eliza Lesmana
BEng CompSC(30360327279)
eliza25@connect.hku.hk
Under the supervision of
Dr. Benjamin C. M. Kao
Department of Computer Science
kao@cs.hku.hk