YouTube AI Mistakes Black And White In Chess For Racism

Computer science experts at Carnegie Mellon University (CMU) began investigating the AI issue after a popular YouTube chess channel was blocked for having “malicious and dangerous” content last June.

This is the content channel of Croatian chess player Antonio Radic whose nickname is Agadmator. It is also the most popular chess channel on YouTube in the world with over 1 million subscribers.

On June 28, 2020, Radic was blocked from YouTube while playing a chess game with Grandmaster Hikaru Nakamura. No specific reason was given by the Googles’-owned video platform.

Radic’s channel was revived after 24 hours, causing the chess champion to suspect that he was temporarily banned for introducing “black and white” content, even though he was talking about chess at the time.

YouTube’s censorship system is based on both humans and AI algorithms, meaning that any AI system can misinterpret comments if it lacks proper training to understand the context.

Ashiqur KhudaBukhsh, a project scientist at CMU’s Institute of Language Technology said that if YouTube relied on artificial intelligence to detect racist language, this kind of accident would entirely be possible.

He tested this theory using a modern speech classifier to sift through more than 680,000 comments collected from 5 popular chess-focused YouTube channels.

After manually assessing and selecting 1,000 comments classified by AI as hostile, the team discovered that 82% of them were misclassified due to using words like “black,” “white,” “attack,” and “threat.” All of these words are used frequently in chess.

The report was presented at the Association for the Advancement of AI’s annual conference held this month.

>>> Former World Chess Champion Warns Against Unemployment Due To AI