Grammarly Accused of Exploiting Identities Without Consent

A Controversial Feature at Grammarly

Grammarly's "expert review" feature, recently highlighted by Wired, has sparked outrage due to its unauthorized use of expert identities to provide writing advice. This feature, which claims to draw inspiration from influential figures, even includes recently deceased professors. During a test of this feature, a user discovered that their own boss, the editor-in-chief of The Verge, was listed as an expert without having given consent.

The AI-generated suggestions from Grammarly included comments attributed to Nilay Patel, editor-in-chief of The Verge, as well as other writers like David Pierce, Sean Hollister, and Tom Warren. None of them had consented to the use of their names for this feature. Grammarly launched this feature in August, promising to refine users' messages with relevant industry insights.

Renowned Experts Without Permission

Among the cited personalities are iconic figures such as Stephen King, Neil deGrasse Tyson, and Carl Sagan. The Verge also identified numerous tech journalists mentioned, such as Casey Newton, Joanna Stern, and Monica Chin. Inaccuracies in the descriptions of the experts, such as outdated job titles, highlight Grammarly's lack of verification and updating of information.

The Verge also found many other tech journalists mentioned in this feature, including former Verge writers like Casey Newton and Joanna Stern, as well as Monica Chin, Lauren Goode from Wired, Mark Gurman and Jason Schreier from Bloomberg, Kashmir Hill from The New York Times, Kaitlyn Tiffany from The Atlantic, Wes Fenlon from PC Gamer, Raymond Wong from Gizmodo, Richard Leadbetter, founder of Digital Foundry, Mark Spoonauer, editor-in-chief of Tom's Guide, Katharine Castle, former editor-in-chief of Rock Paper Shotgun, and Kat Bailey, former news director at IGN. The descriptions of some experts contain inaccuracies, such as outdated job titles, which could have been accurately updated if Superhuman had sought permission from these individuals to reference their work.

Response from Superhuman

Alex Gay, Vice President of Product and Corporate Marketing at Superhuman, Grammarly's parent company, stated that the Expert Review agent does not claim direct endorsement from the experts. It provides suggestions inspired by their work, encouraging users to explore these influential voices. However, Gay clarified that the experts appear in this feature because their works are publicly available and widely cited.

When asked if Superhuman had considered notifying the individuals named in its AI feature or seeking their permission, Gay responded: "The experts in Expert Review appear because their published works are publicly available and widely cited."

Reliability of Sources in Question

The feature presents challenges in exploring the works of the experts. The suggestions often link to spammy copies of legitimate sites or incorrect archives, failing to accurately reflect the original sources. Some suggestions are even based on the work of others, raising questions about the authenticity of the advice provided.

The work of the experts has proven difficult to "explore further." The feature frequently crashed, and its "sources" linked to spammy copies of legitimate sites or other archived copies that were not the true source page.

Some sources even linked to completely unrelated links that were not written by the person whose work was supposedly an example, which could indicate that the suggestions offered by Grammarly's AI with a person's name may be based on someone else's work. This is only apparent if users click "see more" to expand the suggestions and then click the "source" button at the end of the suggestion.

Confusion in the Presentation of Suggestions

In Google Docs, Grammarly's suggestions appear as comments from real users, simulating interaction with the expert imitated by the AI. For example, a suggestion attributed to Sean Hollister from Verge proposed a redundant parenthesis, while the real Sean Hollister would likely have removed this suggestion. This demonstrates that AI, while it can mimic writing style, cannot replace human editorial judgment.

One suggestion from Grammarly's AI "inspired by" senior Verge writer Sean Hollister concerned adding a parenthesis with context already included elsewhere. The only problem is that I was actually edited by the real Sean Hollister, who prefers to avoid repetitive or unnecessary explanations while using clear language and organization.

Had I followed this advice and submitted it to him, the real Sean would likely have removed the parenthesis suggested by Grammarly. An AI can ingest vast amounts of a person's writing and learn to imitate it, certainly, but that same strategy cannot teach an AI how to edit in the way that person would, solely based on the writings they have published, even if you give the bot a checkmark logo and call it an "expert."