How AI Fails: An Interactive Pedagogical Tool for Demonstrating Dialectal Bias in Automated Toxicity Models

TL;DR

This paper quantifies bias in toxicity detection models against African-American English and introduces an interactive tool to demonstrate how biased thresholds operationalize discrimination.

Contribution

It provides a quantitative benchmark of bias in toxicity models and presents a pedagogical tool to illustrate the impact of bias in AI moderation.

Findings

The toxicity model scores African-American English as 1.8 times more toxic.

The model's bias is more pronounced in identity hate detection, 8.8 times higher.

The interactive tool demonstrates how human-set thresholds can perpetuate discrimination.

Abstract

Now that AI-driven moderation has become pervasive in everyday life, we often hear claims that "the AI is biased". While this is often said jokingly, the light-hearted remark reflects a deeper concern. How can we be certain that an online post flagged as "inappropriate" was not simply the victim of a biased algorithm? This paper investigates this problem using a dual approach. First, I conduct a quantitative benchmark of a widely used toxicity model (unitary/toxic-bert) to measure performance disparity between text in African-American English (AAE) and Standard American English (SAE). The benchmark reveals a clear, systematic bias: on average, the model scores AAE text as 1.8 times more toxic and 8.8 times higher for "identity hate". Second, I introduce an interactive pedagogical tool that makes these abstract biases tangible. The tool's core mechanic, a user-controlled "sensitivity threshold," demonstrates that the biased score itself is not the only harm; instead, the more-concerning harm is the human-set, seemingly neutral policy that ultimately operationalises discrimination. This work provides both statistical evidence of disparate impact and a public-facing tool designed to foster critical AI literacy.