Roblox introduces automatic paraphrasing of profanity in in-game chats

Roblox, the online game, has launched a new AI-based feature capable of automatically altering or paraphrasing inappropriate language in conversations within the game. Engadget reported on Sunday that Roblox had previously used an AI filter to block words or phrases that violated community policies. However the system only replaced censored messages with a hash symbol (####), which often made conversations hard to understand.

With this latest feature, words or phrases deemed inappropriate will be replaced by more polite expressions by AI, so the flow of conversation remains comprehensible to players.

Roblox’s Chief Safety Officer Rajiv Bhatia said the initial phase of implementing this feature focuses on rude words. For example, if a player writes “Hurry TF up”, the system will replace it with “Hurry up”.

Each altered message will be accompanied by a note that the sentence has been paraphrased by the system. The sender can also see the portion of language that has been edited. Nevertheless, players who continue to use rude words can still be sanctioned for violating community policies.

The paraphrase feature is currently applied to conversations between users whose ages have been verified and who are within similar age groups, and it is available in various languages supported by Roblox’s translation system.

In January, Roblox also introduced an age-verification system after reports suggesting the platform had safety issues for children, including claims that adult players used the game to approach children inappropriately.

To prevent this, users under 13 are no longer able to use the in-game chat feature, while other players can only communicate with users within the same age range.

Nevertheless, the move has not fully quelled criticism. The Los Angeles County government filed a lawsuit in February alleging that Roblox knew the platform could expose children to paedophiles. A similar lawsuit was also filed by the Louisiana attorney-general, accusing the platform of not adequately protecting child users from paedophiles.