Media, legal experts discuss angst of individuals over appropriate use of powerful technology as institutions, culture scramble to set boundaries
Sarah Newman is collecting secrets. Specifically, she’s collecting secrets about AI.
At conferences and other events, she puts out a wooden box and invites passersby to anonymously disclose the ways they’re using the powerful, mushrooming technology.
“I’ve written a speach [sic] with AI on a topic I don’t have a clue about,” wrote one anonymous confessor.
“Using AI to write official policy language without necessarily disclosing,” wrote another.
Newman, director of art and education at metaLAB at Harvard, sees a lot of uncertainty and angst out there. The aim of her project is to promote discussion about the personal use of AI at a time when its spread has already outstripped the ability of businesses and institutions to define what is appropriate in every instance.
“People don’t know what the rules are,” she said. “There are not established shared norms around both AI use — what’s appropriate and what’s not, across contexts — including when you should disclose use.”
“People don’t know what the rules are.”
Sarah Newman
Since the 2022 release of the popular Large Language Model ChatGPT, businesses, schools, and universities as well as newspapers and publishing houses have been trying to sort it out.
But those efforts have been complicated by the technology’s defining property — its ability to mimic human output. This means it can be difficult, if not impossible, for institutions to police whether users are cheating.
And there have been recent high-profile accusations of misuse.
In May, the British literary magazine Granta responded to allegations that one of its prize-winning short stories, Jamir Nazir’s “The Serpent in the Grove,” was AI-generated. (Nazir has denied using AI to write the piece.)
Separately, the author of a book about truth in the age of AI acknowledged using the technology in writing his book after The New York Times identified misattributed or fabricated quotes in the text.
In addition, news outlets and social media have produced a steady drip of stories about the use of AI in wedding vows, dating app messages, or eulogies.
The eventual societal fallout of all this uncertainty is unclear. But fields like journalism already have been affected.
“What we’re actually seeing is audiences assuming that real media is AI-generated, and that’s a really concerning trend to me,” said Andrew Deck, a reporter who covers AI in the industry for Harvard-affiliated Nieman Journalism Lab. “It indicates that [for] some audiences, in some ways, their default assumption these days is that something is false rather than the other way around.”
In truth, Deck said, very little content at reputable news outlets is generated using AI. Reporters mainly use AI for things like transcribing meetings, scrounging up sources, or iterating title ideas. But they’re doing so amid historically low trust in news organizations.
Pew reported last year that about 56 percent of American adults had “a lot of” or “some” trust in information from national news outlets, down 20 percentage points from 2016. AI, with its reputation for spouting hallucinations with certitude, presents a challenge for an industry grounded in accuracy and pushing back against what Deck called a “trust crisis.”
AI use labels, Deck said, “feel like a way to indicate to readers or audiences that maybe they should have an additional layer of skepticism.”
But disclosure might not always be a clean solution. “AI-assisted” could mean, in theory, anything from a quick source search on Perplexity to an editorial pass that was more like a rewrite.
Issues over uncertain standards appear to have even filtered down to families.
Leah Plunkett, Meyer Research Lecturer on Law at Harvard Law School and faculty associate at the Berkman Klein Center for Internet and Society, is an expert in digital technology, youth, and families.
In her work with K-12 schools, she’s noticed that kids are on high alert for signs that adults in their lives have used AI when they themselves have been forbidden it.
“They have an antenna, as kids and teens — particularly teens — for potential hypocrisy or double standards,” she said. “It can be hard to get them to understand that actually different roles in a school or community ecosystem have always been permitted, and at times required, different tools and resources, because they do different activities.”
Newman of metaLAB has been fascinated by secrets, and why we feel the need to keep them, for a long time. A 2016-2021 project examining why we share so much with our digital devices, “The Future of Secrets,” was exhibited internationally. She moderated a metaLAB panel on AI, secrecy, and disclosure this spring.
Newman believes that norms for AI use will become clearer in the coming months and years. She added, however, “We shouldn’t expect norms to just emerge. We should reflect on our own use and then help shape the norms we want to see.”
Like many people, Newman said she has a love-hate relationship with AI. But, she said, “People shouldn’t feel a need to lie. We should be curious about — even intrigued by — our own mixed feelings, and work toward normalizing appropriate, responsible AI use.”
Newman credits her former research assistant, Olivia Tai ’22, with the idea for the AI secrets project.
Tai is an artist and facilitator who has published her own collection of anonymous confessions. She hopes the project can encourage people to talk more openly about their use of AI and be more curious and understanding of how others view the issue.
“Like with all taboo topics, the more we shame each other, the less productive our conversations become,” she said. “We stop hiding when we feel some acceptance, understanding, and recognition from others.”