Now loading...
Grammarly’s new AI tool for writing feedback draws on supposed insights from prominent figures in various fields, even pulling in names from the tech journalism world without their consent, according to a recent investigation. As highlighted in a Wired report, the feature also references insights inspired by long-deceased academics, raising questions about its approach to expertise.
In testing the system, a journalist discovered an unexpected inclusion: suggestions attributed to her own editor-in-chief at The Verge, Nilay Patel. The tool generated advice seemingly influenced by Patel, along with contributions from editor-at-large David Pierce and senior editors Sean Hollister and Tom Warren. None of these individuals authorized Grammarly to use their styles or personas in this way.
Debuted back in August, the Expert Review function in Grammarly aims to refine users’ text by applying viewpoints from relevant professionals. Activating it in the app’s interface prompts an analysis of the content, yielding tips modeled after experts in the topic at hand. The lineup features well-known authors and thinkers such as Stephen King, Neil deGrasse Tyson, and the late Carl Sagan, among a broad selection.
Further examination revealed a roster heavy with tech reporters, encompassing ex-Verge staffers like Casey Newton, Joanna Stern, and Monica Chin; Wired’s Lauren Goode; Bloomberg contributors Mark Gurman and Jason Schreier; New York Times writer Kashmir Hill; The Atlantic’s Kaitlyn Tiffany; PC Gamer’s Wes Fenlon; Gizmodo’s Raymond Wong; Digital Foundry’s Richard Leadbetter; Tom’s Guide editor Mark Spoonauer; ex-Rock Paper Shotgun leader Katharine Castle; and former IGN news head Kat Bailey. Several profiles include errors, such as obsolete professional titles, which might have been corrected through direct outreach for approval.
Superhuman, the company behind Grammarly, addressed concerns in a statement to The Verge. Alex Gay, vice president of product and corporate marketing, explained that the Expert Review agent avoids claiming actual support or involvement from these figures. Instead, it offers guidance drawn from their publicly accessible writings and directs people to key influencers for further study. For details on the feature, users can check Grammarly’s official guide.
Gay noted that no efforts were made to inform or seek approval from those listed, citing the open availability and frequent citations of their published material. Yet, delving deeper into the referenced sources proved challenging. The tool often malfunctioned, and its citations led to dubious replicas of legitimate sites or irrelevant archives not originating from the credited authors.
Occasionally, links pointed to entirely mismatched content, hinting that advice linked to one expert might stem from another’s output. Users must dig into expanded suggestions and source buttons to spot these issues.
The presentation of these tips can confuse, especially in platforms like Google Docs, where they mimic genuine collaborative notes, evoking edits from the named specialist. For instance, a recommendation styled after Verge senior editor Sean Hollister urged adding explanatory details already present, clashing with his actual preference for concise, direct prose without redundancy.
While artificial intelligence can absorb and replicate patterns from a person’s published output, it falls short of capturing their nuanced editing judgment. Simply analyzing texts, no matter how extensive, cannot equip a model to advise as the original thinker might, even with branding that suggests authenticity.
