Grammarly has pulled a controversial tool that used artificial intelligence to simulate written feedback “inspired” by academics and public intellectuals – both dead and alive – without their consent as it faces a class action lawsuit.

The popular grammar-checking software endorsed by several universities has faced a growing backlash against its “expert review” tool which launched last year.

It allowed paying users to paste their work into the software and get AI-generated feedback from what it called “subject matter experts”, spanning from well-known authors like Stephen King, Neil deGrasse Tyson and Steven Pinker to more obscure academics. None of those listed appear to have been asked for their permission.

A complaint, filed by journalist Julia Angwin on 11 March, alleges that Superhuman, the company behind Grammarly, had violated her privacy by using her identity and was thus preventing her from “controlling the commercial use” of her name. 

ɫ��ֱ��

“In this action, Julia Angwin, an award-winning journalist and editor, challenges Grammarly’s misappropriation of the names and identities of hundreds of journalists, authors, writers, and editors to earn profits for Grammarly and its owner, Superhuman,” it says in the lawsuit, which was 

Superhuman announced earlier on 11 March that it was disabling the feature over the critical feedback it had received. “The agent draws on publicly available information from third-party [large language models] to surface writing suggestions inspired by the published work of influential voices,” CEO Shishir Mehrotra . “Over the past week, we received valid critical feedback from experts who are concerned that the agent misrepresented their voices.”

ɫ��ֱ��

In a statement to Times Higher Education, Mehrotra said the feature had “missed the mark on what both experts and users expect out of us”.

“We announced that expert review was being taken down for a redesign before the claim was filed, and in its short lifespan it had very little usage. We are sorry, and we will rethink our approach going forward.

“We have reviewed the lawsuit, and we believe the legal claims are without merit and will strongly defend against them. Regardless, there is a better approach to bringing experts onto our platform and we are working on a version that will provide significantly more benefit to both users and experts.”

Many universities in the UK and US have a Grammarly subscription and actively direct students to use it. On its website, it lists Stanford University, Boston University and the University of Pennsylvania among its many university clients.

ɫ��ֱ��

The University of Birmingham has previously published specific guidance on Grammarly use, although it does not address the “expert review” tool and it is unclear when these rules were issued. Times Higher Education has contacted the university for comment.  

John Kaag, the chair and professor of philosophy at the University of Massachusetts Lowell, said scholars and teachers appearing in the Grammarly tool “implies endorsement or participation that was never agreed to”.

Kaag has created his own chatbot for the company Rebind after spending 30 hours recording his own thoughts, insights and answers to questions about Henry David Thoreau’s memoir Walden.

This was a remarkably different approach to Grammarly’s, Kaag explained, as it allowed users to have a guided conversation about the book, drawing directly from his real commentary.

ɫ��ֱ��

“In Rebind, we get and contract original commentary about great books, hours and hours of original commentary from some of the world’s best teachers and readers,” he said, adding that this is then drawn verbatim into conversation with users as they work their way through a book and there is attribution to the commentary, which has been contracted.

“Grammarly’s position is that expert review does not claim direct participation from those experts – the suggestions are described as inspired by their work rather than produced by them. This is just garbage,” Kaag said. “Grammarly is using named scholarly authority as a design choice to make the AI feedback feel more credible and domain-specific.”

ɫ��ֱ��

seher.asaf@timeshighereducation.com