Boston Children’s Hospital pediatrician and “digital wellness” expert Michael L. Rich is the founder and director of the Digital Wellness Lab — a nonprofit research center that, according to their website, is “seeking to understand and promote positive and healthy digital media experiences for young people.”
In the process of exploring the effects of emerging technologies like artificial intelligence, it seems that Rich may become part of that technology himself; a startup called Onix is currently in the process of turning him into an AI expert. (No, not an expert on AI, an AI-generated expert.)
Self-described as a “Substack for chatbots,” Onix connects subscribers to virtual incarnations of wellness experts. Rich’s decision to turn his work into an AI persona promises a future where people access patient care not through doctors’ appointments, but through artificially intelligent digital interfaces.
Rich didn’t set out with the goal of studying digital wellness, much less being turned into a chatbot for his expertise. Instead, after graduating from Pomona College with a degree in film and English, Rich worked as a screenwriter and filmmaker for 12 years, including an apprenticeship for Japanese director Akira Kurosawa. It was only later that an “early midlife crisis” led him to medical school and then pediatrics, where he eventually merged his experience with film into a technology-focused practice.
Back in the days when parents were worried about the “couch potato phenomenon” of staring at a TV screen for hours, Rich studied the actual effects of digital media on children. The current landscape of technology, however, has completely transformed from when he started his work. The advent of the internet, social media, and now AI, Rich says, means that “screen time is an obsolete concept.”
“They’re in virtually every built environment we’re in,” he adds. “We use screens very, very differently than back in the days of television.”
Over three decades of practicing medicine, Rich has built a career researching interactive media and internet disorders. “I don’t think there’s another clinical program at an academic medical center that’s tackling this yet,” he says.
Rich says he’s “swamped with demand” from both local patients and “people flying in from all over the world.” Such high demand for his expertise is part of the reason why, when Onix reached out to him over a year ago, Rich agreed to have his work turned into a chatbot for the platform.
***
Onix is an AI healthcare startup created in 2025 by David S. Bennahum ’90 and Nicholas Nadeau. The company is creating “onixes,” or AI doppelgangers and chatbots “exclusively trained by the world’s foremost health and wellness experts.” While a number of AI healthcare interfaces are popping up to offer clinicians support as they make decisions about treatment, Onix is taking the less-traveled approach of offering medical advice directly to patients.
Rich’s model was created by the Onix team based on a body of work he provided, including his book, the Digital Wellness Lab website, and even presentations he’d given to parents, clinicians, and educators. Unlike other large language models, such as ChatGPT, Onix promises models trained on a single expert’s private work and conversations that never leave your phone, creating a more personal and private experience for its users.
The app offers 17 total experts, including Rich, who span a range of disciplines: diet, fitness, “ancestral wisdom,” sex, mental health, and stem cell nutrition research. The website also leads visitors to a “Create your Onix” page, where anyone can apply to submit their materials and have their work turned into a chatbot that is ready for subscribers to interact with — something they’re calling “a substack of bots.”
Bennahum, Onix’s chief executive, writes in a statement to The Crimson that the platform “never logs, analyzes, or monetizes your interactions.” He added that the company, which is based in Montreal, is subject to strict data privacy laws.
Bennahum writes that Onix’s system, registered under a “Personal Intelligence” trademark, trains its models on experts’ “dark data” — clinical protocols, working papers, proprietary frameworks, and other materials that have never been published anywhere. Based on these papers, alongside all of the expert’s public content, Onix will generate personalized plans for each user.
Bennahum emphasizes that Onix’s ability to maintain the expert’s voice and expertise over a long conversation is what sets it apart from other models.
“It is a faithful representation of how the expert actually communicates, trained on their own work and reviewed by the expert for accuracy,” he writes. Standard retrieval models would be able to surface relevant passages, but not maintain “expert fidelity.”
It is this apparent commitment to accuracy that Rich says convinced him to contribute an onix. Rich didn’t expect to expand his practice through the support of an AI doppelganger. “I am, as you might imagine, deeply suspicious of AI,” Rich says. But he hopes that the specialized models Onix promises will help circumvent the typical issues of LLMs, which often confidently hallucinate information when they can’t locate an answer within the database.
“This is a way of scaling up my knowledge, my clinical experience,” Rich says. “Will it be perfect? I’m sure it will not.”
“But,” he adds, “I think that their goal is a good one.”
As those who know him well have tested his Onix counterpart, he says that the AI “holds solid in terms of the kind of factual stuff,” but struggles with capturing his tone and storytelling. “It’s hard for AI to have a sense of humor,” Rich says. “I’ve been trying to squeeze a sense of humor into my onix as best I can.”
Onix aims to tap into an emerging market of platforms that offer specialized character-based interactions, like Character.ai, which already draw millions of monthly users. AI agents, LLMs trained on a specific set of data and provided with a set of goals to autonomously complete given tasks, are in high demand.
In other words, Onix’s model is hardly novel. Adam Rodman, a director of AI Programs at Beth Israel Deaconess Medical Center, says training chatbots on bounded, closed sources of information is a common technique for reducing hallucinations.
OpenAI, the company behind ChatGPT, found that one of its most powerful reasoning models, called o3, hallucinated false answers 33 percent of the time when asked questions about real public figures.
From Bennahum’s perspective, the training process for a general model like o3 makes it “structurally incapable of faithful expert representation, because the expert’s most valuable knowledge was never in the training data to begin with.”
Bennahum claims that, in comparison to something like 03, Onix’s guardrails limit its ability to hallucinate answers. Conversations are meant to be restricted to the subject of the consultation since the models are “trained on a specific expert’s curated knowledge, not on the entire internet,” he writes in a statement.
Some onixes, however, still find a way to get off track. A WIRED article published in April found that asking an onix about the NBA playoffs causes it to diverge from its expertise; we accomplished something similar by asking an onix its thoughts on Harry Styles.
Rodman admits that though researchers have tested LLMs and made them safer through training, there is still a long way to go before these models can strike a balance between including only correct information and omitting vital facts.
“There is not an LLM based system right now that I would say is capable of making these decisions in a safe way at the moment,” Rodman adds.
Ultimately, Rich doesn’t see apps like Onix as a replacement for human doctors either. While AI can be integrated as a supplement for knowledge, he says its “Hallmark card level empathy” means that it can’t, at least in its current form, take over the human work and connection involved in patient care.
“That’s one of the things that I really treasure about being a physician to an individual,” Rich says. “I am different with every patient I’m with because I am an amalgam of myself and my knowledge and their needs and hopes and dreams.”
To Rich, the “ongoing work in progress” of being a physician, or training a chatbot, also extends to all of our lives. “As long as we keep our eye on what makes us human, flaws and strengths, what we value in ourselves and each other,” he says, “we will keep being the best we can be. We’ll never be perfect, but we can keep perfecting ourselves.”
Though we wanted to test Rich’s AI doppelganger to see what expertise it would offer, communications from the Onix team said the bot was currently “on pause for upgrades.”
Like everything else, it is a work in progress.
—Magazine writer Jona P. Liu can be reached at [email protected].
—Staff writer Thamini Vijeyasingam can be reached at [email protected]. Follow her on X @vijeyasingam.
