In the age of AI, small colleges are punching above their weight

In an era of billion-dollar research budgets and AI-driven discovery, small research colleges are proving that scale isn’t everything. Through close mentorship, collaborative problem solving, and a willingness to integrate new tools into the scientific process, institutions like Wellesley and Bowdoin are preparing the next generation of scientists to navigate and shape a rapidly evolving research landscape.

In 2023, the Massachusetts Institute of Technology (MIT) spent over one billion dollars on research and development. Wellesley College, just a half-hour drive from Cambridge, spent less than one-hundredth of MIT’s budget, but that didn’t stop astrophysicist and Wellesley alum Dr. Lamiya Mowla from publishing a first-author paper in Nature the following year. 

Galen Stone Tower, Wellesley College (Credit: The Swellesley Report)

According to the Carnegie Classification system, Wellesley is categorized as a small research college. Instead of graduate students or postdoctoral fellows, undergraduates with full course loads conduct research with guidance from their professors. While small research colleges may lack graduate degree programs and large-scale budgets, their focus on mentorship, collaboration, and critical thinking are shaping how the next generation of scientists will conduct research in an increasingly AI-driven world. 

“I think AI is going to become the new norm in the way that we use Google as a verb,” says Mowla. “As teachers, I think we have to just teach students how to use it safely and in a way that they're still using problem-solving skills.”

How Wellesley students helped analyze the Firefly Sparkle Galaxy 

During JWST’s first week of operation in 2022, Mowla reunited with the Canadian NIRISS Unbiased Cluster Survey (CANUCS) team for the first time since the COVID-19 pandemic. Together, they examined the inaugural deep field snapshot, a picture of the galaxy cluster SMACS 0723. The cluster’s immense mass distorts, or gravitationally lenses, the light of astronomical objects behind it like a magnifying glass. 

Hidden among the warped white streaks of light, Mowla spotted something extraordinary: three distorted images of the same background galaxy, stretched like a cosmic funhouse mirror. “When it comes to more complex structures like this, they’re still identified by the human eye,” she explains. “We haven’t been able to find an automated technique, but we’ll get better at training our machine learning models to do it.” 

Mowla named the galaxy Sparkler for the glimmering compact sources surrounding it. Further photometric analysis of the dozen “sparkles” suggested that at least five are globular clusters, massive groupings of hundreds of thousands of ancient, red stars that may have formed in the early universe.     

Color images of the Sparkler galaxy and its surrounding environs. (Credit: Lamiya Mowla et al 2022 The Astrophysical Journal Letters 937)

“It was three weeks of very little sleep and nothing other than just living and breathing that data set all together,” says Mowla. “At that time, I was still part of the University of Toronto which is a big R1 [very high research activity] institution and there is always that push to be the first one to make a big discovery.”

Two years later, Mowla’s Nature-worthy discovery was a small, low-mass galaxy in a JWST deep-field photograph of the galaxy cluster MACS J1423. The new find, a delicate, luminous wisp with its own chain of star clusters, earned the name Firefly Sparkle. 

Lamiya Mowla, Assistant Professor of Astronomy at Wellesley College (Credit: UofT News)

“Being at Wellesley, I don’t feel the need to do something where I’m worried about being scooped,” says Mowla. “I choose projects that can be done on a longer time scale so I can take my undergraduate students along with me through the research process.”

Using ChatGPT to format Python tutorials, she walks undergraduate researchers through the rigors of astronomical data analysis, covering everything from opening multidimensional data files to basic plotting. She’s even used her own slides, each one designed for a 75-minute lecture, to generate extensive tutorials with self-evaluating questions. 

One undergraduate from Bangladesh, who reached out after seeing Mowla’s name in the news, adapted the code for other high-redshift galaxies. That student’s analysis on photometric error was included in the Firefly Sparkle Nature paper. 

“As an undergrad student, it's quite a feat to be on that paper,” says Mowla. “Now, she's working on her own first-author paper on all of the galaxies that she has analyzed over the past two years.”

Inside Bowdoin’s mentorship model for training future physicists 

At Bowdoin College in Maine, physics professor and affiliate member of the high-energy physics Fermi Large Area Telescope (Fermi LAT) collaboration, Fe McBride knows what it’s like to feel lost at a big university. As an undergraduate at Germany’s Friedrich-Alexander-Universität (FAU) Erlangen, she sat in lecture halls with hundreds of students and no personal connection to professors.

Hubbard Hall, Bowdoin College, Maine (Credit: Bowdoin College)

“Being a student at a research-focused university can be a very lonely experience.” says McBride. “I failed classes and had to retake them. There were not any support structures in place and you were just thrown off the deep end.”

After teaching a 200-person class during her Eberley Research fellowship at Pennsylvania State University, McBride decided that she wanted to work for a smaller school with a focus on teaching and mentorship where “everyone has an open door policy.” In 2022, she joined the faculty at Bowdoin, a 2,000-student research college with roughly half the research budget of Wellesley College.    

Fe Mcbride, Assistant Professor Department of Physics and Astronomy at Bowdoin College. (Credit: Bowdoin College)

“What specifically impressed me about Bowdoin was that the physics and astronomy department is a very inviting place,” says McBride. “Students randomly come in to talk, even if they’re not taking a class with me that semester.” 

Without the pressure of competing for a single tenure-track position or applying for federal grants, McBride can take on a heavier teaching load during the academic year, leaving summers for more intensive undergraduate research projects.   

“My main goals are to provide excellent teaching and a lot of student hours,” she says. “This can be office hours, mentoring students, advising students, and just making sure we guide everyone along the way.”  

From hallucinations to insight: Teaching AI literacy in physics 

As educators, McBride and Mowla have had to navigate students using AI tools in the classroom. The results have been a mixed bag of confident wrong answers and confusing hallucinations.

In one of Mowla’s first-year physics labs, students used a pinhole camera and two blocks of paraffin wax separated by a thin piece of foil to determine the Sun’s diameter and surface temperature. After helping the students take measurements outside, Mowla noticed a group using ChatGPT to find the solar flux, or the total radiated energy, from the Sun. The result was a twisted version of Wien’s law, an equation relating absolute temperature and the peak wavelength of emitted light. 

“I told them to check the units and they saw that the ChatGPT answer was wrong,” she says. It became a good moment for them to realize that an AI model’s answer is not always right.”

McBride found that in-class exercises where students can observe and record AI hallucinations has helped them understand the limitations of large language models and reflect on where in the learning process they need to put in more effort. She has built a chatbot using AI that asks her advanced students conceptual questions without giving away answers, encouraging deeper reasoning. 

“No one would give a first grader a calculator and ask ‘what’s three plus one?’ We teach people how to do addition and once they understand the basics, do we really need them to do everything in their head or on paper?” says McBride. “I hope that eventually AI will become the same thing as a calculator.”

Small schools, big impact

By embedding students in the full arc of the scientific process and by treating AI as both a tool and a teaching moment, small research institutions are shaping scientists who are as adaptable as they are technically skilled. Unlike larger universities, where undergraduates often orbit around research led by postdocs and faculty, here they take on the work directly: framing questions, running experiments, and grappling with both the promise and pitfalls of new tools.

In a research culture dominated by speed, automation, and large-scale infrastructure, their approach offers a reminder: the next era of discovery may depend less on expensive resources, and more on how we train the minds that use it.

Bryné Hadnott is a planetary scientist turned freelance science communicator based in Oakland, CA. Their work has appeared in Physics Today, The Planetary Society, and Stanford University’s Kavli Institute of Particle Astrophysics and Cosmology. Follow them on LinkedIn.