Last fall, Dartmouth College made headlines after announcing the development of a new A.I.-powered tool for student wellbeing, called Evergreen. At a moment when concerns about the use of A.I. for mental health were coming to a head, Evergreen’s emergence signaled a continued interest in employing the technology in this way, in the right way. Already, Dartmouth has amassed $7 million of its $16 million fundraising goal to support Evergreen’s production.

But not everyone is equally ready to embrace the app. Over the last several months, students have expressed skepticism of the emerging resource, which has taken up a notable space in campus discourse. As institutions navigate how best to both regulate and leverage A.I. for students, the response at Dartmouth offers insight into when and how young people seem willing to accept the technology in their lives, as well as the kind of communication they may require from institutions introducing it in new ways.

Evergreen is far from Dartmouth’s first foray into A.I., nor digital health innovation. The Ivy League college has a longstanding tradition in both realms, including as the site where the term “artificial intelligence” was coined 70 years ago. In the spring, The Boston Globe reported on how the college has strived to live up to its history, outpacing its peers in the speed and breath of its rollout of A.I., to mixed reviews. Evergreen is discussed, with a focus on Dartmouth’s recent partnership with the A.I. company Anthropic and potential plans to "integrate the tech into most facets of campus life.”

Meanwhile, since 2011, Dartmouth’s Center for Technology and Behavioral Health (C.T.B.H.) has been ground zero for pioneering research on digital interventions for the treatment of behavioral health issues. Lisa Marsch, the center’s founding director, described the goal of C.T.B.H. as leveraging the accessibility of modern technology, like phones and wearables, to make a broad and meaningful impact on health and wellbeing. Marsch, who specializes in technology and addiction treatment, is known for developing a related mobile tool which became the first F.D.A.-authorized “prescription digital therapeutic.”

"How do we think about the whole student body and ways that we can bring the science of flourishing to students in a way that scales?"

It's now Marsch and several of her fellow experts at C.T.B.H., including psychiatrist Nicholas Jacobson and computer scientist Andrew Campbell, who are spearheading the Evergreen project. It launched as an opportunity to tackle the wellbeing challenges young people are facing nation- and worldwide, Marsch said, as well as fulfill an institutional priority laid out by current President Sian Beilock to better support student mental health at Dartmouth specifically. Unlike the counseling center, aimed at treating or diagnosing mental health conditions, a digital tool could offer everyday, in-your-pocket tips to strengthen wellbeing and ward off more serious crises.

“It's about prevention,” Marsch said of her vision for Evergreen. “It's about, how do we go upstream? How do we think about the whole student body and ways that we can bring the science of flourishing to students in a way that scales?”

In its current form, Evergreen has a few different functions. The interface is an A.I. chatbot, named Evie, through which students can engage with educational modules on different behavioral health topics. The first version of Evergreen’s A.I. isn’t generative, like well-known chatbots ChatGPT or Claude, meaning it doesn’t offer original or spontaneous responses. Instead, all the content comes from pre-scripted, expert-reviewed dialogues that offer insight into college-relevant issues, including sleep hygiene, nutrition, and physical activity. 

Those interested in a more personalized Evergreen experience have that option. By allowing the app to collect passive data on different contextual elements of their life, like step count, heart rate, and sleep patterns, students can receive targeted assistance for their health needs. Someone whose sleep cycle has fallen out of whack, for example, could get a nudge from the tool to review the sleep module. 

Evergreen’s capacities differ from traditional wearables because the app was designed with a Dartmouth-specific experience in mind. If users provide data on upcoming assignments and deadlines, the app can support them with time management. If they tell it about their extracurricular interests, it can point them towards relevant events on campus. If they decide to track their meals, it can show them the nutritional information for that day’s spread in the dining hall. 

As Evergreen takes in individual student data, it will also be able to spit out campus-wide trends. Another of the app’s services will be to present these types of broad insights into user behavior, so students can understand their wellbeing collectively. Throughout the semester, they could track how many hours of sleep their peers are averaging. During exam periods, they could watch what happens to visits to the fitness center or time spent outdoors.

It’s important to note at this point that Evergreen is still far from a live product. A year since its announcement, the app has only just entered an initial round of testing, involving 15 students, although that number is expected to jump to 200 in the fall. A second, more advanced version of Evergreen that does involve generative A.I. is being developed simultaneously and isn’t estimated to start testing until 2027.

By students, for students

The reason that Evergreen has already managed to secure an active role in the minds and lives of Dartmouth students is likely because hundreds of them have been employed to work on the project. Upwards of 250 of these project assistants, as they’re called, have been helping engineer the technology, deliberating on the issues Dartmouth students might need the app’s help navigating, crafting the scripts for the dialogues, and producing and narrating supplementary videos. It was a student’s idea, for example, to integrate the dining hall menu. And it was students who suggested Evergreen provide designated support for those who weren’t admitted to the fraternity or sorority of their choice, or at all.

The assertion of its student-centered ethos seems to have opened Evergreen up to particular scrutiny from, well, students.

It's a key point of pride for Evergreen that it’s “Built by students, for students” (that’s essentially the greeting line on the home page of the app’s website). But ironically, the assertion of its student-centered ethos seems to have opened Evergreen up to particular scrutiny from, well, students. The promise that they’re in the driver’s seat ends up deepening skepticism of the enterprise altogether when they sense that they’re not.

In January, students’ hackles were raised after the editors of The Dartmouth, the student newspaper, revealed the author of an op-ed supportive of Evergreen had received payment from his employers at the app for the time he spent writing. His supervisors at Evergreen also provided feedback on a draft of his essay, which was titled “Why Evergreen.AI Deserves a Chance.” That headline alone suggests reservations about the tool were already brewing. Now there was another reason to imagine students weren’t really in charge at Evergreen. 

Then, later this spring, a similar publicity snafu rocked evergreen when The Dartmouth again turned up some incriminating details about student involvement in the project. Three different student workers had come forward to admit they used A.I. to write dialogues for the Evergreen modules, which were branded as intentionally student — not A.I. — generated. 

In the article, Lisa Marsch affirmed this conduct was against the rules put in place to preserve the student voice of the content. More recently, she amended that, given the extent of the editing process for these scripts, some of the language having been originally A.I.-generated isn’t inherently concerning. “I don't know that it would be that much of a problem because we have such a careful, thorough review of the quality from peers and experts that ultimately nothing's going to make it into this that hasn't gone through that rigorous process,” she said.

The biggest barrier to overcome for Marsch and her team moving forward may be how students feel about Evergreen. These feelings could be mixed up in views on A.I. generally, as well as Dartmouth’s specific efforts to champion it. To Trace Ribble, a rising sophomore, Evergreen is part of his school’s larger mission to “claim” A.I. as its expertise. “We want to be the A.I. school. We want to be on the cutting edge,” he said. By his account, he added, “there's not much student support for it.” 

Ribble was responsible for another op-ed in The Dartmouth, expressing his view that “Evergreen.A.I. Is Not Good for Dartmouth.” He writes from a background in youth mental health advocacy, having founded a nonprofit to make resources accessible in rural areas like his home in Oklahoma. His concerns about Evergreen center on a wariness of the newness and therefore efficacy of A.I. for mental health support. Evergreen, he writes, is a “multi-million dollar cash sink designed to generically ‘support student success.’” The funds raised, he believes, would be better spent on expanding in-person care.

It’s true that people struggling with their mental health have turned to A.I. chatbots and received dangerous guidance, or lack thereof, in recent years. Generic large language models often aren’t trained to recognize warning signs or offer referrals for behavioral issues. Around the time students were first hearing about Evergreen, one major news story featured a teen boy who died by suicide after falling in love with a chatbot and his family’s effort to hold the A.I. company responsible. His parents aren’t the only ones.

Ribble understands Evergreen isn’t positioning itself as therapy, but he worries the slope is slippery. “They're walking this fine line where they're like, ‘Oh, we're not trying to replace traditional therapy.’ But then they're lauding their connection to, I think it's called Therabot, which is just exactly what it sounds like — therapy A.I.,” he said. “I don't know. It's really icky to me.”

The biggest barrier to overcome moving forward may be how students feel about Evergreen.

Therabot is indeed the generative A.I.-therapy chatbot designed by Dartmouth psychiatrists Nicholas Jacobson and Michael Heinz, who are also leaders of Evergreen. In a recently published clinical trial, the first of its kind for this kind of software, users were found to have experienced a 51 percent average reduction in symptoms for depression and 31 percent for anxiety.

Still, Heinz confirmed Evergreen’s work is different. “I don’t think it’s the place for this app to deliver mental health services,” he said. Even the second version of the app that, unlike the first, will have a generative component will be “purpose-built in a way that it knows what it is, it knows its limitations.” So if a student started chatting with the bot about a serious mental health issue, it would be able to refer students to better resources for help. “It would know the off-ramps,” Heinz said. “We're training this in right now, so it knows the resources at Dartmouth, and it's able to connect students to those resources.”

The “off-ramps” are key to Evergreen, Marsch agreed. This digital resource is different from typical online content, she said, because it actually wants users to get off and engage beyond. “Unlike a lot of social media and other online things that people can do, which have algorithms designed to keep your eyes glued to the screen, this is all about the opposite. It's about giving you resources to connect you to the real world.”

Keene Dampal, another mental health advocate at Dartmouth, is a project assistant for Evergreen. He said he knows that the app isn’t “the best perceived” among his peers, but while he’s an enthusiastic supporter, he doesn’t necessarily think the wider hesitance is such a bad thing. “In fact, I think it's a very useful thing,” he said. “It just shows that people are critical before they actually just say yes to something that's new.”

Clarifying the mission of Evergreen will be important, Dampal continued, but in the end, not everyone may want to use it. That’s okay, too, he said, as long it helps someone who does.

You can reach LearningWell Reporter Mollie Ames at mames@learningwellmag.org with comments, ideas, or tips.