Audrey Woods, MIT CSAIL Alliances | November 19, 2025
Before becoming a master’s student at MIT CSAIL, Aiden Foucault Etheridge spent two years prototyping ice cream flavors at Toscanini’s Ice Cream in Cambridge, MA. While it didn’t directly link to artificial intelligence or computer science research, it taught him how to take ideas from testing to final product while incorporating data, customer requirements, and validation sets.
Now, Etheridge is working with MIT CSAIL Researcher Amar Gupta and CSAIL Alliances Affiliate Itaú Unibanco to leverage AI in fraud detection and predicting customer behavior, joining a global effort to apply emerging technologies in the FinTech space.
MIT, ICE CREAM, AND BRAZILIAN BANKS
Etheridge has always loved math and came to MIT because he felt the community was “the right level of nerdy for me.” Originally, he planned to become a mechanical engineer, but “once you’re at MIT, it’s hard not to be around computer science.” He took a few theoretical computer science classes and enjoyed the problem-solving aspect of them, which led him to consider AI research and graduate school.
But his path took a delicious two-year detour at Toscanini’s. He describes how working in ice cream flavors was “shockingly similar to the world of AI research, where you have an idea of something you want to create, and then you have to go about prototyping it, testing it, seeing how people like your flavor and what they might add to it. Then you roll out the final product.” Etheridge describes the experience as fun and educational and invites readers to try the two flavors he created which are still in rotation—Maine Blueberry Mint and Drunken Cherry.
Returning to MIT as a Master’s in Engineering student, Etheridge has joined a larger, ongoing collaboration between CSAIL’s Dr. Gupta and the Brazilian financial services company Itaú Unibanco. Broadly, this project is exploring the various ways AI programs can help banks rapidly detect anomalies such as fraud. Etheridge’s group—one of four student groups in this collaboration—is focused specifically on detecting credit card fraud.
One of the challenges in using AI to detect credit card fraud is a machine learning phenomenon called overfitting. Etheridge says, “it is incredible working with Itaú because they track every transaction, so they have this huge breadth of data that I get to play around with and test and train models on.” However, one drawback of such a large sample size is that there are relatively few examples of fraud compared to the number of normal data points. “If we try to just train a model to predict whether something is fraud or not, the vast majority of transactions won’t be fraud.” This means that the model might learn to predict not based on some underlying causal mechanism but on strange statistical anomalies that occur within the data. For example, a model might notice that a few fraudulent transactions happen to occur at 3:17AM and begin flagging all transactions at 3:17AM as fraud even if the time of day isn’t actually meaningful.
To mitigate this, Etheridge and his team are exploring ideas from LLM design even though they aren’t using LLMs themselves. Take ChatGPT, which is trained via self-supervised learning to predict the next word in a string of words. “This principle can be applied to very good effect in the financial transaction world, where customers are always making transactions.” Using self-supervised learning, Etheridge’s model is being trained to predict, with a high degree of accuracy, the next transaction a given customer might make based on their previous purchasing history. In doing so, their model can flag—and hopefully stop—fraudulent behavior as it is occurring.
Applying such a model in the market is no simple matter. Human beings aren’t always predictable—someone who goes to Sephora every Friday might decide to go to a Burger King across town or dollar store instead—and it’s difficult to calibrate the model for such uncertainty. Etheridge explains, “while there’s a huge amount of credit card fraud going on, customers don’t want to be inconvenienced by a blocked transaction or even a text message checking to see if a purchase was them. So you have to try very hard to only flag something as fraud when you’re 99.9% sure.”
Despite challenges, Etheridge sees this area as “a very hot topic of research in the financial industry” and is excited to be working on something that many other banks and financial service providers are trying to solve. In the industry, he says there’s general enthusiasm around the idea of financial foundation models which could deeply understand customers and their transactions for other opportunities like targeted advertising, travel planning, and contact-free checkout.
In his spare time, Etheridge is also advising a telehealth company called OHealth to help them develop AI models which would make healthcare in developing countries more accessible and effective. “These areas have a lack of trained professionals or an inability to connect those professionals with the people who really need their care. We are designing AI systems that can assess a person’s condition and get them connected with the right professional in a time-efficient manner.” This project has been both fun and fulfilling and Etheridge hopes to continue doing it “for as long as they want me to.”
FUTURE PLANS & FINTECH
After he finishes his master’s degree, Etheridge plans to get a job in industry, hopefully doing some kind of research on theoretical models. He says he genuinely loves the joy of discovery and is “excited by the prospect of gaining insight into the underlying principles behind intelligent behavior and learning.” He sees FinTech as a great space for this work “because there are so many unexplored problems and puzzles to solve on a day-to-day basis.”
For now, he appreciates working with Itaú as it allows him to apply abstract concepts of intelligence in “a very down-to-Earth benchmarked task.” There are obstacles and the solutions he and his colleagues come up with sometimes don’t work in practice as well as they did in theory. But for Etheridge, the challenge is part of the draw. “Things always get grittier than you expect. That’s what makes it interesting.”
Spotlight bonus videos! YouTube Shorts:
