Google’s AI-Powered ‘Inclusive Warnings’ Feature Is Very Broken

Author: Samantha Cole

Publisher: Vice

Publication Year: 2022

Summary: The following article discusses how earlier this year, Google created a writing assistant tool that works similarly to Grammarly but checks for inclusive language in addition to making suggestions regarding style and tone. In theory, this product has good intentions, but in practice, it has performed poorly at times. For example, the word “motherboard” was flagged with an inclusivity warning for having a gender-specific word even though there is not a substitute for the word “motherboard” in the English language. This implies that there are places in the code where the tool should make exceptions to general rules. The author also found that a transcribed interview with Ku Klux Klan leader David Duke did not receive any notes despite being filled with racial slurs and hateful language. This implies that the tool does not detect hate speech.


Posted

in

, ,

by