Colby Faculty Using Artificial Intelligence Tools in Variety of Cross-Disciplinary Research

Colby faculty from across campus are incorporating artificial intelligence and machine learning in a variety of ways throughout their research. News of the recent $30-million gift to establish the first cross-disciplinary institute for artificial intelligence (AI) at a liberal arts college has created even more excitement around this research area, and faculty are looking forward to the many opportunities it will foster and inspire.

Made possible by the tremendous generosity of the Davis family and trustee of its charitable foundation Andrew Davis ’85, LL.D. ’15, the Davis Institute for Artificial Intelligence will provide new pathways for talented students and faculty to research, create, and apply AI and machine learning (ML) across disciplines while setting a precedent for how liberal arts colleges can shape the future of AI.

To continue the conversation around AI and ML and introduce faculty, staff, and students to the topic, Colby faculty recently shared related research during two Zoom presentations sponsored by the Davis Institute. Below are descriptions of their research as well as videos of their presentations.

Aaron Hanlon and Megan Cook, English
AI and the Study of Literature

When it comes to AI and the study of literature, Associate Professors of English Aaron Hanlon and Megan Cook consider three major categories: building AI, which is work on developing AI itself; using AI, AI as a set of tools for the study of literature; and guiding AI, studying literature to understand social implications of AI. 

In terms of building AI, studying literature can play a key role in developments and natural language processing. In using AI as a tool, literature scholars frequently analyze large corpus or body of text in search of patterns that reveal something about language or history, for example. What literary scholars call “The archive” has expanded rapidly with AI capabilities and can now be analyzed quickly. This allows researchers to discover bias in the archive and design falsification experiments, such as the Crowd Detector which is training an algorithm to predict instances of crowds using textual clues. The researchers hope this work will be useful beyond literary studies, such as analyzing the concept of fake news.

Collaboration with other fields will remain key to AI work more broadly in literature and the humanities, according to the researchers. For example, biologists and manuscript scholars have developed a technique that allows for noninvasive extraction of protein from parchment, or the animal skin used as a writing surface in most medieval manuscripts. Comparing the protein “fingerprint” of the parchment to known samples identifies the species of animal whose skin was used to make the book, which gives historians new information about how and when a book was made. 

There is a long history of AI in literature dating back at least a millennium. “Imaginative writing, literature, and language have always played a productive role in the development of AI, and we’re excited to see this continue at the Davis Institute,” Cook said.

Oliver Layton, Computer Science
Robust biologically inspired self-motion and object motion estimation during drone flights

Assistant Professor of Computer Science Oliver Layton researches computational neuroscience, which looks at how the brain processes information and holds the key to biological intelligence. This field looks at simulating how the brain works on a computer and developing smarter technology by studying brain design.

Layton is developing AI and autonomous systems that incorporate brain-like elements that perceive and act in real time in dynamic environments, just like a human brain. His goal is to achieve human-like, self-motion perception and navigation in an artificial system while operating in realistic environments. 

Layton also is working with several students on related projects, including Eli Decker ’21 and Mufaddal Ali ’21 who are developing and running human self-motion estimation experiments in virtual reality; Sinan Yumurtaci ’23, who is conducting a comparative analysis of neural mechanisms in different areas of the brain to determine what optimizes self-motion performance; and Natalie Maus ’21, who is undertaking a deep learning of motion and optic-flow patterns in dynamic environments.

Alejandra Cortiz, Geology
Atolls and AI: Survival or Drowning

Alejandra Cortiz, an assistant professor in the Geology Department, is using AI to determine what drives change on the world’s atolls, or low-lying tropical islands containing coral, and how those changes can be measured. Her research looks into what processes drive coastal geomorphic change and how the effects of sea-level rise affect coastal evolution.

Cortiz and her collaborators are using a model to predict if the offshore wave height controls how wide a reef flat or motu is and if that is then seen on a global scale. To determine this, they are using satellite images to identify objects, calculate metrics, and compare data and are training a deep neural network in AI that can then learn what is land, a reef, or vegetation, etc. 

The long-term plan is to feed images directly into the AI model, which will automatically classify atoll landscapes and motu widths. As the research team collects this new data, they will easily see changes by comparing it to past information.  

Ekaterina Seregina, Economics
Research in AI/ML: Economics Perspective

Ekaterina Seregina, who will join Colby this fall as an assistant professor of economics and finance, is primarily focused on researching asset management and financial economics. While Seregina has used ML in her recent research, she cautions that almost none of the AI/ML algorithms in pure form work for the majority of economic problems and the current AI tools must be modified for use in economics.

Seregina is developing an ML-advised financial advisor that determines where investors should allocate funds. These robo-advisors are digital platforms that provide financial advice to investors in an automated way. They offer many benefits as opposed to human financial advisors, such as providing services at a lower cost and serving individuals with lower levels of wealth. They also are constantly monitored and can be improved over time. 

Nick Record, Bigelow Laboratory for Ocean Sciences
AI and Big Data at Bigelow

Nick Record, a senior research scientist at Bigelow Laboratory for Ocean Sciences, focuses on computational biology, working on everything from “viruses to whales and all kinds of things in between.” Through the Center for Ocean Forecasting, Record uses algorithms and data to predict factors such as where and when whales are likely to be found and to forecast harmful algal blooms. Several cross-disciplinary collaborations related to AI/ML that Record is interested in include forecasts and early warning systems as climate adaptation tools and issues of data justice and algorithmic accountability.

Bigelow, which is a close teaching and research partner with Colby, is using AI and big data in a variety of areas including single-cell genomics, identifying genomes and linking them with environmental dynamics; ocean optics/remote sensing to identify algae; and environmental DNA, which can be linked to forecasting. Through their Data Discovery Initiative, Bigelow also is using big data and AI to integrate across disciplines in complex ocean systems science.

Hannah Wolfe, Computer Science
Interactive Media/Artificial Intelligence

Hannah Wolfe is an assistant professor of computer science with a background in arts and software engineering. 

Her interactive art installation Cacophonic Choir aims to bring attention to the firsthand stories of sexual assault survivors by addressing the ways in which their experiences are distorted by digital and mass media and how these distortions might affect the survivors. In the exhibit, survivors’ stories become less random and more coherent as visitors approach each speaker. To achieve this level of distortion and interaction, Wolfe and her collaborators employed several digital media techniques including ML, physical computing, digital audio signal processing, and digital design and fabrication. This past year Wolfe created a virtual version of the exhibit with Colby computer science student Rayna Hata ’23.

Denise Bruesewitz, Environmental Studies
Understanding the Biogeochemistry of Aquatic Systems: A Story of Building a Bigger Toolbox

Denise Bruesewitz, associate professor and director of environmental studies, focuses her teaching and learning on connecting freshwater and people through areas such as carbon storage and climate change. Several of her large research questions are difficult to answer using traditional water samples that are only collected at one place and time.

Bruesewitz is working with a large group of collaborators from several institutions to build out AI tools, such as computational methods and autonomous robotics systems, to create more efficient modeling and predictions of harmful algal blooms. The long-term goal of the interdisciplinary project is to think about the drivers of where, when, why, and how algal blooms are formed and to combine tools of eDNA, robotics, and big data with traditional samples to better understand these systems.

They hope to eventually share their approach with groups such as lake associations or water districts that could utilize these types of big data sets.

Josh Martin, Biology
Artificial Intelligence on the Move: Bio-Inspired Design and New Tools for Big Data

AI in robots struggles to come close to matching biological intelligence in terms of bodily movement, and animals like the praying mantis make it look easy, according to Assistant Professor of Biology Josh Martin. A goal of Martin’s research is to understand how insects move and model that intelligence to control robots.

Martin and collaborators built artificial neural networks inspired by the structure and function of the insect’s real neurons and used them to control MantisBot, a bio-inspired robot with AI. The MantisBot is beginning to recreate coordinated movements of the insect, as well as intelligence that allows it to attract and catch prey. 

Daisy Dan ’20, a biology and computer science major, developed an AI tool to determine how such varied legs of the 2,400 species of praying mantis can all serve successful hunters. Dan used images of over 700 species of praying mantis and trained an AI algorithm to identify the same points on every species it encounters. The AI automatically generated data that allowed the researchers to calculate a measure of mechanical efficiency.