Dartmouth Convenes Landmark ‘AI at 70’ Conference
The three-day AI at 70: What Must Remain Human? conference at Dartmouth included more than two dozen panel discussions, keynote speeches, and other presentations. The talks featured pioneering researchers, leaders in artificial intelligence, and scholars who study AI’s potential effects on society. Here is a sampling of reports from some of the sessions.
Artificial intelligence is increasing the speed and scale of activity related to national security, two panelists said Oct. 29 during a session on AI in International Security and Conflict co-sponsored by the Davidson Institute for Global Security.
“There are two things that AI models, foundation models, are doing that fundamentally change the national security environment,” said Sasha Baker ’05, OpenAI’s head of national security policy and a former acting undersecretary of defense for policy. “One is they are increasing the rate of change…whether those are in science or in cybersecurity or what have you. Everything is happening faster, and the second is the ability to scale. I think most militaries, most governments, are still grappling with what the implications of that are for the way that they do business.”
Moderator Erica Barks Ruggles, the Obenshain Family Distinguished Visitor and Diplomat in Residence at the Dickey Center for International Understanding, followed up with a related question about how governments should regulate these emerging technologies layered with AI capabilities.
“I’ve been studying the proliferation of drones and increasingly autonomous weapons for the last decade, and everything points to a global shift in power, and that affects the way in which we need to look at national security policy and the wars that we choose to wage,” said James Patton Rogers, a lecturer in security and technology in the Department of War Studies at Kings College London and an expert on uncrewed combat systems and war.
He noted that Iran’s high-tech response to the United States’ Operation Epic Fury—deploying drones and precision missile technologies over ever greater distances with ever greater levels of precision—is “making the United States and its allies reconsider its footprint in those regions.”
Adding AI data processing to these technologies would result in a “global shift in power due to the proliferation of these new technologies in real time, and that certainly should give us a pause for thought.”
Baker said Ukraine’s use of drones and other advanced tools illustrates how urgent circumstances can accelerate innovation.
Participants discussed national and international safety standards, export and proliferation controls, accountability for autonomous systems, and the challenge of holding non-state actors responsible. Patton Rogers noted that the United Nations has discussed lethal autonomous weapons for a decade, with limited progress on regulation.
Baker cited a U.S. Department of Defense directive that requires appropriate human judgment in autonomous weapon use. “The American way of war has always been that at the end of the day, there is a commander who is accountable,” she said.
Sasha Baker ’05, OpenAI’s head of national security policy, speaks at the panel discussion on AI in International Security and Conflict. (Photo by Lars Blackmore)
Regarding safety, Patton Rogers added that advanced AI could potentially help personnel apply international humanitarian principles.
He also stressed that military systems in general, including aircraft carriers, may evolve rather than disappear due to changing technology.
An audience member asked about parallels with regulating nuclear weapons.
“A lot of the sort of nuclear nonproliferation analogies break down with AI pretty quickly because nuclear weapons are physically scarce, and the material that you need to build a weapon and the knowledge that you need to build a weapon can effectively be controlled,” Baker said. “We’re probably already past that point with AI.”
Teaching, Learning, and Pedagogy
A conversation about Teaching, Learning, and Pedagogy in the Age of AI on Oct. 30 in the Hanover Inn highlighted some of the related challenges facing academia and efforts to help students understand and possibly use AI in a way that enhances their studies.
James Dobson, associate professor of English and creative writing, ticked off some of AI’s negative consequences: It has “drained much of the color” from the challenges and joys of learning; made assessing student learning difficult, if not impossible; and levied an emotional toll on those trying to navigate fast-evolving norms for its use in academia, he said.
Dobson also noted some of the opportunities AI offers.
“Language models and computer vision models are enabling research and teaching activities across fields and disciplines,” he said. “This moment, too, provides an occasion to rethink our pedagogical commitment.”
James Dobson, associate professor of English and creative writing, and Roopika Risam, Dartmouth AI Fellow and chair of the Department of Film and Media Studies, lead a conversation on Teaching, Learning, and Pedagogy. (Photo by Sophia Scull ’25)
For several years, Dobson said, he has used his teaching “as a test bed for thinking about AI literacy, what students need to know about this technology, and how to engage with generative AI in a scholarly context.”
In his course Critical AI, students are tasked with coming to a historical, political, and technical understanding of neural networks and with using the technology in a way that is personally meaningful. Dobson hopes that analyzing generated stories or pop songs, comparing learned moral frameworks across different LLMs, or constructing predictive models for politics will help students “better understand their potential uses and limitations.”
Roopika Risam, Dartmouth AI Fellow and chair of the Department of Film and Media Studies, shared the results of a months-long effort to define AI literacy and what it means at Dartmouth. A subset of the Faculty Leadership Group on AI tackled the question by first asking, “What is it that we need to teach our students, even if we put aside AI?”
They agreed on four core competencies, starting with knowledge in their fields of study and critical inquiry skills. Next was the ability to apply ethical and sociopolitical awareness to understand what shapes the work they do as thinkers and learners. Finally, students should be able to synthesize those three components to act with “responsible agency,” employing “informed, principled decision-making skills,” Risam said.
In their model, AI literacy is a subset of the core competencies, each with its own considerations for whether or how students might decide to use AI.
Risam gave examples for each level, starting with the first.
“If you don’t have knowledge in an area in which you are prompting a large language model to give you outputs, you have no way of really knowing if those outputs are actually useful for you,” she said.
Risam noted that this year, the Faculty Leadership Group on AI is engaging in a number of community conversations about norms and shared principles, and invited students, educators, and alums in the audience to share their thoughts.
“What should we be thinking about, in relation to generative AI, as we strive to support our students and give feedback to the policymakers, give feedback to our faculty, and give feedback to our students?”
Her introduction to the conversation drew laughter throughout the room.
“Above all, what we need to be able to do is to have open and honest conversations about generative AI with each other,” she said. “And it’s really hard to do because this is a topic where we have people who are wildly enthusiastic and we have people who don’t want to hear the letters A and I ever again.”
A fundamental logical flaw is driving over-hyped claims about AI’s capabilities—and investment in the technology at the expense of richer, more human academic work—speaker Molly Crockett said in their talk, The History of AI and the Future of Human Knowing, Oct. 30 in Moore Theater.
Crockett, a professor in Princeton University’s Department of Psychology and University Center for Human Values, took the audience on a tour through the history of AI, highlighting where researchers have fallen prey to a “well known risk of modeling in science.”
The illusion of generalizability “arises when we inappropriately generalize from the part to the whole,” Crockett said, and while AI has long been put forward as a model of human thinking, computer tasks can actually represent only a portion of human ability.
“Is it really accurate to say that this kind of a model can predict and capture human cognition? Does it really explain our capacity to climb mountains, cure cancer, explore outer space, create captivating art?” they said, pushing back on recent claims about a new AI model designed to predict and simulate human behavior. “When we hear that an AI model can simulate human performance on cognitive tasks, I think this says more about the limitations of what those tasks can measure than it says about the capabilities of the model.”
Princeton professor Molly Crockett says the “future of human knowing might depend on asking less of our machines and more from ourselves and each other.” (Photo by Sophia Scull ’25)
Throughout the talk, Crockett acknowledged many technical achievements in AI research while cautioning against “mistaking its singular voice for the broader chorus of human talents that we value in each other.”
Looking ahead, the “future of human knowing might depend on asking less of our machines and more from ourselves and each other,” said Crockett, who also cited the philosopher Briana Toole’s comparison of communities of knowers to an orchestra playing a symphony.
“Just as each instrument adds beauty and texture to a wall of sound, knowing the world is a group project, and each knower can contribute something unique and different,” they said. “In community, we can know more as a collective than we can individually know ourselves. It’s up to us if the symphony plays on.”
Can We Trust Visual Imagery Anymore
Hany Farid, a professor of computer science and a pioneer in the field of digital forensics, weighed in on the question: Can We Trust Visual Imagery Anymore?
Farid, whose research focuses on developing new tools to identify manipulated digital images, a phenomenon accelerated by generative AI, said that while there are technical solutions that can be developed, “trust is a very human thing.”
“One of the problems in our society—and this is not uniquely American, but I think we have a particular problem here—is that we have eroded trust in institutions,” Farid said. “So, when a media outlet that is respected says something, it’s very easy to demonize them; when an academic says something, it’s easy to demonize them, and that’s a very human problem that we need to resolve.”
An Oct. 29 panel on Arts, Ethics and AI began with audience members creating an original song with Nona Hendryx’s AI-powered avatar.
The legendary singer-songwriter and producer, and fellow artist-scholars Paul D. Miller, who performs as DJ Spooky, and professor Jacqueline Wernimont each shared some of their work before the group turned to the questions AI raises for the arts. ProfessorRoopika Risam, chair of the Department of Film and Media Studies and a Dartmouth AI Fellow, moderated the discussion.
DJ Spooky addresses Roth Visiting Scholar Nona Hendryx and Jacqueline Wernimont, Distinguished Chair, Digital Humanities and Social Engagement, during the Arts, Ethics, and AI panel at Dartmouth’s AI at 70: What Must Remain Human? conference. (Photo by Kata Sasvari)
Attendees scattered across Loew Auditorium to scan a QR code and open Cyboracle Prime, Hendryx’s “digital twin” trained on her music and her understanding of Afrofuturism, the aesthetic that blends African diaspora history and Black culture with science fiction and technology.
“It was important for me to create something that was centered in my history and engagement with the world, and also make it a creative expression,” said Hendryx, who built the conversational AI experience with the technology studio EyeJack and developed it with Dartmouth students as a 2026 Roth Visiting Scholar.
Echoing the African musical tradition of call and response, Cyboracle’s prompts shape the song it builds. “The one thing that stays with a human being until the last moment is movement and melody,” she said, pointing to studies of people with Alzheimer’s disease and dementia.
Miller, whose collaborators run from Metallica and Chuck D of Public Enemy to the composer Steve Reich and Yoko Ono, made the talk itself a demonstration of his method: a collage of disparate sources. “I am remixing my talk as I go,” he said. “It’s a riff.”
He moved from the origins of the internet, the subject of his multimedia concert Quantopia, to the Iron Rider, a bicycle he designed this year with Santa Cruz Bicycles, its frame patterned after Marcel Duchamp’s readymade Bicycle Wheel, to his forthcoming book Digital Fiction, on how algorithms change the way we tell stories. A common thread, he said, is that nothing is ever finished. “We are in an era of the parasocial,” he said. In a post-truth world of “computational propaganda,” he asked, how do we think about the authentic self? “AI,” he said, “is LLMs remixing all of us.”
Wernimont, the Distinguished Chair of Digital Humanities and Social Engagement, works in what she calls “data visceralization,” turning data into something you can touch, to give back to bodies “what has often been stolen from them.” She pointed toEnergy Pools, her fiber installation in the Irving Institute atrium: its colorful hanging strands stand for the energy the U.S. uses, with interior strands representing the scale of green energy Americans need to embrace in order to reach international climate goals by 2030. Visitors can reach in and braid those strands out by hand, pulling that future into the present.
She also invited attendees to handle a towering woven cloth at the front of the auditorium from A Table Weaving, a collaborative project that turns data on the growth of cloud storage into fabric. “An extraordinary amount of energy is used to store our photos, our music on our public cloud,” she said, noting the real figures on that usage aren’t public. “I wanted to spend time thinking about it, in a way that felt real and accessible.”
Risam turned the discussion to what AI means for artists’ work.
“Buckle up. It’s going to get wild,” Miller said. “It’s going to get weirder because the bots are making other bots.”
“The idea of critical thinking is becoming a lost art,” he added. But imagination remains “the ultimate renewable resource.”
The talk turned to how AI prizes the finished product over the process of making it. Wernimont pushed back. “I don’t know how we learn anything without the process,” she said. “Muscle memory is really important. I think that’s how we know.”
Hendryx put it in a similar way. “It’s the making, the process,” she said. “We build. I think that’s what is satisfying for us as beings.”
She described the exchange with Cyboracle as “an embodied process that reaches across time and space,” a stranger meeting her history to make something together.
Around the Green: AI Solutions
For one of the last panels, several Dartmouth researchers presented their ongoing work as short talks.
“We spent much of this afternoon wrestling with some big questions around AI—where it can help, where we need to be cautious, and where we still do not know,” said Peter Chin, AI at 70 Conference faculty chair and professor of engineering, in his introduction.
“We’re going to shine a light on some amazing work that our colleagues here and at other institutions are already doing to tackle some of these really difficult challenges,” Chin said.
“Think back to your first year of college, not the big moments, but the little moments, like the night that you skipped dinner because there was a big problem set you hadn’t done yet,” said Geisel School of Medicine professor Lisa Marsch, who opened the session talking about Evergreen, a generative AI platform being built by Dartmouth students and faculty to provide students with personalized wellness support.
“For most of us, those are moments that passed, but for some students, these are events that quietly stack up,” said Marsch, who leads the project as the founding director of the Center for Technology and Behavioral Health and the Andrew G. Wallace Professor of Psychiatry and Biomedical Data Science.
Evergreen is intended to help students navigate those stressors before they compound into more serious problems by providing “personalized support precisely when it might be needed,” Marsch said. The interactive platform is currently in a randomized controlled trial that has enrolled 200 Dartmouth undergraduates.
“So much of the support that we offer to young people is designed for the moment that someone is already in crisis. So, the question is, what if we could start earlier?” Marsch said. “We know that every student deserves support before a crisis and not just after one.”
Joe Catrino, executive director of the Center for Career Design, says the center “is helping students prepare for jobs that don’t yet exist.” (Photo by Sophia Scull ’25)
Joe Catrino, executive director of the Center for Career Design, talked about how Dartmouth is preparing students for life after graduation by emphasizing durable skills such as judgment, curiosity, identifying and solving unexpected problems, and follow-through.
“The Center for Career Design is helping students prepare for jobs that don’t yet exist,” Catrino said. “Seventy years ago here on this campus, we got the name of something new. Seventy years later, the machines can answer almost everything. But the career question of what is worth doing with a life is still ours to answer and to ask.”
Jacqueline Wernimont talked about the lack of information on the energy consumed by high-performance computing, including AI. Wernimont and co-presenter Carly Bobak, director of research engagement and data science in Information, Technology and Consulting, conducted a campus-based study to learn if Dartmouth’s Living Lab model could be used to better understand where this information can be found, where the gaps are, and steps to develop a campus monitoring system to make computing more sustainable.
“This was a story about a problem—an absent or missing dataset problem. We can do better, we know we can,” Wernimont said. “That’s a thing we can do and we can be a leader in higher ed in that space.”
Bijan Mazaheri, an assistant professor of engineering, laid out the limitations of AI to generate data and what makes human-derived information unique. AI struggles with concepts such as anomalies, cause and effect for unusual events, and counterfactuals (such as predicting the next year of growth for a plant in a dark closet while ignoring that plants need sunlight).
“Our mistakes are what make us human and our obsessions with them, diving into those mistakes,” Mazaheri said. “Until we can get AI to work at all the different levels of this hierarchy, from abstract to fine-grain, AI machines are just going to be average machines that are built on the anomalies of human minds.”
Mikhail Gronas, an associate professor of Russian, opened the presentation on the Human-AI Co-Creativity Consortium, which included his collaborators Anna Rumshisky from the University of Massachusetts Lowell and Roger Beaty at Penn State.
The multi-institutional group of “cognitive alarmists” works toward collaboration between AI and humans that defends human cognition and creativity, Gronas said. The group developed benchmarks for measuring, encouraging, and training people in co-creativity.
“We’re basically alarmed by cognitive offloading, cognitive atrophy, and creative atrophy,” Gronas said. “One of the central questions of today’s world is what will humans do when AI does what we do better than us?”
“Humans explore in a variety of different directions. If you drop humans out of this process, the models will do something different. The models are optimized to achieve a reward,” Rumshisky said.
“If this happens and humans are taken out of the equation, our sciences and our knowledge base will continue to develop, but we won’t actually know what we lost … and which pathways we have not explored,” she said.
-Steve Hartsoe, Aimee Minbiole, Harini Barath, Simone Silverbush, and Morgan Kelly contributed to this report.
