Tending to the Digital Commons: Examining the Potential of Artificial Intelligence to Detect and Respond to Toxic Speech (Toda Peace Institute)

In order to prevent toxic speech online, the present study investigates the possibilities of artificial intelligence (AI), specifically large language models (LLMs). The study focuses on hate speech classification and counterspeech generation, highlighting advancements in supervised, unsupervised, and generative approaches while recognizing important limitations, such as the amplification of biases from training data, implicit hate, and difficulties in capturing nuance. Reviewing efforts to create AI-powered counterspeech tools, the focus is on the challenges of creating human-like, constructive responses to offensive content. In order to ensure ethical and successful deployment in addressing online harms, the research finds that LLMs hold promise for counterspeech applications and provides guidelines for developers and policymakers.

https://toda.org/policy-briefs-and-resources/policy-briefs/report-253-full-text.html