Programs to monitor hate speech mix up the conversations

"Artificial intelligence" does not know the difference between talking about chess and racism

Talking about "black and white" chess pieces may be interpreted by artificial intelligence as "racist talk" that should be withheld.

Archives

After the tremendous success of the TV series "Queen's Gambit", meaning (The Minister's trick), which takes place in the world of world chess tournaments, the game of chess has gained great attention and has become a constant topic of conversation on social media.

Finally, however, a study conducted has shown that programs for monitoring hate speech or dialogues that contain offensive racist expressions may confuse dialogues related to the game of chess with conversations about racism issues.

The study conducted by two researchers from Carnegie Mellon University in the US attributed the reason for this confusion, that talk about the colors of the chess pieces that differ between white and black may be interpreted by artificial intelligence systems for detecting offensive phrases on social media as "racist talk" that should be withheld. Under the rules that operate by these systems.

The TechExplor website, which specializes in technology, indicated that a Croatian chess player named Antonio Radek had been blocked on YouTube to display the videos last year, claiming that it contained "harmful and dangerous" content.

Researcher Ashkar R.

Khodabush, of the Institute of Language Technologies at Carnegie Mellon University, said that talking about "white" versus "black" was what led to the blocking of the channel, because the program used by the website automatically interpreted these phrases as "racist."

He added, "We do not know the program that (YouTube) uses, but if the site relies on an artificial intelligence system to monitor racist expressions, then these types of incidents may occur," indicating that if this accident happened with a famous figure such as Radek, It can happen to others without anyone knowing about them.

Within the framework of the study, Khoudabouch and his colleague Robak Sacer tested two programs to monitor racist expressions on social media sites. More than 680,000 phrases from five chess channels specialized in the game of chess were shown on YouTube on these two programs. Then the two researchers examined 100 phrases that they The two programs classify it as "offensive", but after the study it was found that 82% of these expressions were "not offensive" at all, but rather included words such as black and white, attack, threat, and other common terms in the world of chess.

• 82% of the statements that were rated offensive were never offensive.

• Blocking a chess player channel for allegedly publishing "racist" content.

Follow our latest local and sports news, and the latest political and economic developments via Google news