Two years ago, when Colby used a $30-million gift from the Shelby Cullom Davis Charitable Fund to launch the Davis Institute for Artificial Intelligence, President David A. Greene talked to dozens of reporters. They all wanted an answer to this question: Why would a tiny liberal arts college in northern New England create an academic and research-based institute dedicated to a discipline that sits at the very forefront of science and technology?
Greene’s answers always focused on a core belief that artificial intelligence should never become the province of huge research institutions alone.
“Other sectors of higher education have to have a strong foothold here,” Greene told the Boston Globe.
In the two years since, AI has become the most talked-about scientific innovation since the internet itself. Institutions from the Brookings Institution to the Carnegie Endowment for International Peace to the United Nations have issued warnings about the threats posed by AI.
But she also sees AI as a means to improve the human condition. And if you look deeply at what Colby’s Davis Institute has accomplished, you see how that can happen. The proof is in the results—or, as it were, in the machine-learning models. Research from the earliest work of the Davis Institute will help psychologists diagnose and provide care for children on the autism spectrum. And it will help scientists and economists understand how global warming is changing the geography and economic output of the small island nations that dot the world’s oceans.
The institute’s achievements are possible because of how it works. It isn’t built to speed the development of AI models for their own sake. It’s built to further the potentially life-changing research that comes from academics across the Colby campus, by applying AI tools.
“We don’t do anything on our own,” Stent said. “We do it always with subject matter experts from another discipline. That’s one of the great things about being at a liberal arts college. I literally do not run a hackathon, a course, a seminar, or even a dinner without subject matter experts. The second part of it is really to continually pay attention to what our values are and whether the opportunities that are presented to us and the activities we are engaging in help forward our values or not.”
In short, the Davis Institute for AI stays true to its goal of improving the human condition by helping Colby faculty whose research is designed to improve the human condition.
Each day starts with a ‘stand-up’ meeting
This simple formula works, according to the Colby researchers who have collaborated with the institute, because Stent and her colleagues have built in methodologies that foster collaboration at a level rarely seen in academic institutions.
In her private-sector work developing technologies for the likes of Bloomberg LP, Yahoo! Labs, and AT&T Labs, Stent became steeped in a software-creation method known as “agile” development. In an agile organization, every day begins with what’s called a “stand-up meeting,” where all the parties on the team gather to share what they plan to work on that day. These meetings are called “stand-ups” because no one may sit down.
The lack of comfort moves things along quickly. Agile organizations also require weekly “sprint” meetings, where team members share their accomplishments and lay out their plans for the following week. The group then boils down those weekly goals into larger objectives that are pursued in four-month-long “sprints” of activity.
The structure of the Davis Institute drives this method. The primary crew is small. It includes Stent and the Davis AI Associate Director Amy Poulin, one or two postdoctoral associates, including inaugural associate Tahiya Chowdhury and Michael Yankoski, who joined the institute this fall, a small number of professors who spend their sabbatical years collaborating with the institute to further their own research, and participating student researchers.
Over the past year, those two Colby faculty members were Alejandra “Aleja” C. Ortiz, assistant professor of geology who studies island nations on hundreds of atolls—or coral-reef chains in the oceans—and Veronica “Vero” Romero ’09, assistant professor of psychology who studies how people interact with each other and their environment in varying contexts. Romero is using AI to model the behaviors of children with autism when they interact with psychologists and others.
In typical academic settings, one would rarely see a geologist like Ortiz working closely with a psychologist such as Romero. But inside the Olin Science Building, where the institute is headquartered, the two work together every day.
“Anytime you get people who speak slightly different disciplinary languages in a room, if they’re patient enough to work through their differences, then you get these fascinating sparks,” Stent said. “It’s just been a real pleasure because they were very willing to participate with us in this somewhat unusual structure for academia.”
Romero is enthusiastic about the method.
“We have run into these problems where I’m sitting there going, ‘I’m trying to understand this, but I’m stuck,’” Romero said. “And then Aleja will go, ‘Well, I don’t know what they do in psych, but in geology we do this.’ And then it clicks, and it’s like, ‘Oh, I’ve never done that. But that makes a ton of sense for my specific situation.’ That’s happened multiple times to us. It’s been really great.”
Ortiz told similar stories. She had a database of several thousand islands and was trying to figure out how to collate them.
“In other words, what’s the average size of an atoll island? How do you physically calculate that? Do you use mean? Do you use median? How do you track the error associated with that and propagate it? I was kind of running into these pieces, just dealing with big data and statistical analyses. So I mentioned it, and Vero was like, ‘Oh, yeah. You should be using descriptive statistics.’”
Descriptive statistics are a method of summarizing features within an enormous collection of information.
“We ended up having a whole discussion about what might make the most sense for this particular project and the problem I was having,” Ortiz said. “It really strengthened the analyses I could do, because I could say statistically, ‘Oh, these are discreet populations. The atolls in this region are physically different from the atolls in this other region.’ That happened because we were in the same room, and every day we had a meeting for five to 15 minutes where everyone said, ‘Hey, I’m going to be doing X today, and I’ve had this problem and I need to fix this other thing.’ It enables us to really be collaborative. That has just been absolutely fantastic and really has helped all of my research.”
Ortiz recently coauthored a paper detailing some of her research in the academic journal Frontiers in Earth Sciences.
Applying AI across the curriculum
The collaborative atmosphere of the Davis Institute for Artificial Intelligence is just one force that helps these researchers accelerate their work. The other is its expertise in applying AI models to different disciplines of study. Postdoctoral Associate Chowdhury is a computer engineer with an extensive background in the creation and application of AI models.
“You can use machine learning for many kinds of data,” she said. “It can be financial data, it can be climate change data, self-driving car data. All we want with machine learning is to find patterns that lie in the data. Then, in the future, you can use those patterns. I am particularly interested in the type of data where you have humans involved.”
The human impact of Romero’s work on children with autism is easy to see. In Chowdhury’s work with Ortiz, who focuses on understanding the climate-related changes on coral atolls, the ultimate focus is on how that shifting geography will affect the hundreds of thousands of humans who live on the world’s 400-plus atolls and depend on them for their economic well-being.
“I would say all my projects have overlapped between these three themes—human, machine, and nature,” Chowdhury said. “How they interact with each other, how the interactions impact each other, and how we can find a space where human, machine, and nature can coexist without harming each other.”
Ortiz referred to herself as a “coastal geomorphologist.” Translated, that means she studies landscape change over decades.
“At its most basic, it’s really how water and in particular waves move around sediment, usually sand,” she said. “I study this using a variety of different tools. A lot of times, I’m using computer models where the underlying physics of how water moves sand is built into the model and you run it over long enough time scales that you can look at what we call morphodynamic change, which is the change in the size and shape of the landscape. I also use remote sensing and satellite imagery to study historic land changes.”
Decades of satellite imagery and data on the 400-plus coral atolls around the world are available, showing the changing size of the islands along an atoll. Traditionally, a geomorphologist must manually classify each measurement from a land-based sensor or a satellite. And manual classification poses tremendous problems, simply because there is so much data.
“What we’re trying to do with AI is figure out a more efficient and more accurate way to automatically classify those satellite images,” Ortiz said. “The reason that we want to do this automatically is because a given atoll can have 100 islands, and up until now, people have done things where they hand-digitize the data points. I’ve done it. It’s just really boring and takes a really long time.”
Her goal is to develop AI models that can more accurately predict how climate change will affect people who live and work on various atolls. Will they be able to keep producing the products that drive their economies? Will they even have islands to live on?
“These are unique ecosystems,” Ortiz said. “It’s kind of funny because you can have one island, like a four-square-mile island that can have 18,000 to 20,000 people living on it. Then the island next door has one person. Population density can vary a lot. Some of them, like the Maldives, the Seychelles, and French Polynesia, are places that can bring in a lot of money from tourism. Then there are fisheries and growing coconut palms for oil, sometimes aquaculture with regards to growing pearls. And the big question is, will the people who live on these landscapes continue doing that?”
Romero refers to herself as an “ecological psychologist.”
She contrasts her work with neuropsychologists, who study what brains do when their owners are asked questions, and cognitive psychologists, who study what happens in the mind as people undertake various activities.
“Our goal is to understand how people act in context,” she said. “As ecological psychologists, we understand that the brain is part of a full body, which is embedded in a specific context. It’s moving or behaving within a specific situation. We try to take what we call a systems perspective. For example, you and I are talking, and it’s just important to understand what I’m saying and what you’re saying not in isolation. We look at the interaction between the two.”
Romero’s goal is to apply that approach to the interactions between psychologists and their patients. When a clinician examines a child who might have autism spectrum disorder, they must assess not only what the child says but also how they conduct themselves over the course of the entire conversation, how they play games, and how they solve problems.
“The clinician has to be able to do the assessment, have conversations with children, take notes, and notice different behaviors all at the same time,” Romero said. She has spent the last year working inside the Davis Institute to develop AI tools that can analyze video recordings of sessions, unearthing patterns of behavior and speech. Psychologists could stay deeply immersed in the conversation as it happens and rely on AI to help analyze the session after it’s finished.
Her ultimate goal is to go beyond simply diagnosing autism to assisting psychologists when they recommend treatment protocols.
“We’re hoping the AI tools can help us eventually create a system where the clinician can say, ‘Yes, your child has autism,’ and we think these specific services would help them,” Romero said. “So the parent has a place to start and doesn’t have to go to three or four extra people, who then have to do their own assessment and then start whatever intervention it is.”
The Davis Institute’s approach and methodology, where faculty researchers collaborate closely with each other and with the institute’s technologists, offer the potential for a new brand of AI development—one that is, as Colby officials have dreamed of from the beginning, truly human-centered.
And arguably, such an approach could only work in the small liberal arts environment that Colby embodies. Ortiz came to Colby after a long stint at North Carolina State University, where she was one of 50 faculty in environmental engineering. She noted that Colby has more than 200 faculty total. At NC State, Ortiz collaborated only with faculty members in her own department. At Colby, she collaborates with a fellow researcher whose work is nothing like hers and highly qualified technologists who can apply AI tools to speed up her learning.
“I was on one of the initial faculty committees when the Davis Institute was created,” Ortiz said. “I was very excited about it, but I really wasn’t sure how it would play out.” After a year spent working inside the institute, she called her experience “really fantastic and somewhat surprising. Amanda, the director, has really instituted this wonderful culture of collaboration and collegiality, which is unusual in academia sometimes.”
Shared values hold that culture together, Stent said: a deep belief that artificial intelligence should focus on making life better for all people. There’s a high probability that the institute’s work will, at some point, attract large investments from the private sector, but Stent said that should never become the primary focus.
“We will continually pay attention to what our values are and whether the opportunities that are presented to us and the activities we are engaging in help forward our values or not,” Stent said. “We want to be proactive about our values—and not just let them go when the money comes.”
Colby AI Entrepreneurs Win Maine Startup Challenge
With their ScholarSheets idea, Desmond Frimpong, Miz Insigne, and Peggy Jones envision a simpler way for international students to apply for scholarships
Getting a Head Start on AI
The Davis Institute for Artificial Intelligence and Halloran Lab for Entrepreneurship host SureStart, a summer program that teaches real-world skills to students interested in AI
Colby Debates a Blueprint for an AI Bill of Rights
The campus hears from a coauthor of the White House’s statement of principles for artificial intelligence
Elevating the Role of Undergraduates
Adaobi Nebuwa ’24, a computer science neophyte when she got to Colby, now plays a pivotal role in one CS professor’s lab
A Vital Element of AI? Empathy
How one Colby graduate is putting humanity into AI technology