Abigail Gertner’s career has centered on artificial intelligence. The department manager of AI-Enhanced Discovery and Decisions at MITRE talked with Karina Wright about having a front row seat to AI’s evolution.
From Big Idea to Problem-solving: How AI Shaped a Computer Scientist’s Career
Abigail Gertner has watched artificial intelligence (AI) grow from a primarily academic study with few practical applications to today’s vast number of uses.
“When I started out, we were developing rule-based systems and belief propagation algorithms to make inferences on probabilistic graphical models. The computing hardware didn’t exist to support the kind of large-scale model development we can do today.”
The turning point, Gertner says, came with “the availability of GPUs [graphics processing units] for training larger AI models, along with the massive amount of data available on the internet.”
Both led to the rapid progress in machine learning (computers learning models based on data) we’ve seen in recent years, “moving the field forward to solve a lot of the problems AI aimed to solve decades ago.”
Any situation where a human needs information or analysis—or there's some complexity to deal with that AI is better suited to handle—we develop tools to help make those kinds of consequential choices.
“In our AI and Autonomy Innovation Center, we're working on large-scale problems across the full spectrum of our sponsors, including defense, transportation, and health. Each sector has unique challenges, and solutions can bring meaningful impact for the country and, in some cases, the world.”
MITRE sponsors, Gertner explains, are using AI-enabled applications to increase public safety and boost the efficiency of government services, among many other uses.
Gertner emphasizes the importance of AI assurance—the process of ensuring that AI systems perform effectively, with acceptable levels of risk.
Her team’s research runs the gamut, from using language processing tools to help fight disinformation, to policy equity analysis, to applying AI to improve social services delivery.
“We're developing new approaches to problems no one has worked on before,” she says.
And that’s right where Gertner wants to be.
“It’s not just the variety of applications, but also the technical challenge that I enjoy.”
AI for Public Good
As an AI researcher who's also a department manager, Gertner juggles many priorities.
“Last year I was co-principal investigator of a research project focused on using large language models (LLMs) for understanding strategic messaging. For example, we can reverse engineer a synthetic message to get an understanding of the intention behind it. How was it generated? What prompt was used? What model?”
The goal is to potentially combat disinformation that LLMs might help proliferate.
Another current focus is the use of LLMs as decision aids.
“No matter how powerful your LLM is, there's still a lot of higher-level system development needed to ensure it’s used in a safe, secure, ethical way, and that it interacts well with humans.”
On that front, Gertner collaborates with The University of Texas at Austin’s Good Systems initiative to promote responsible, ethical, and socially beneficial use of AI. She heads a project developing tools to facilitate services for people experiencing or at risk of homelessness.
Client-focused tools can help a person understand what services they might be eligible for—and make the application process easier, Gertner explains. For case workers, AI tools can help summarize large volumes of case notes into an easy-to-digest format.
Gertner and her team also work with stakeholders to establish priorities and identify potential risks associated with using AI in the social services domain.
“A recognized weakness of LLMs is their potential to hallucinate, to make up things that aren’t true. It’s important to design systems in such a way that users don’t act on incorrect information.”
Persistence Is Everything
MITRE landed on Gertner’s radar during graduate school, when she met researchers at a user modeling conference. She submitted her resume, but there wasn’t an open position at the time.
She instead did postdoctoral research at the University of Pittsburgh. There, Gertner applied Bayesian networks, a type of model for reasoning under uncertainty, to model students’ knowledge and understanding of physics. The goal was to build an intelligent tutoring system.
Four years after her first interview, Gertner joined MITRE in 2000.
“It turned out the work I did in Pittsburgh was applicable to my early MITRE work—developing training applications as part of our independent R&D program.”
Gertner holds a Ph.D. in computer science, but, she says, “machine learning wasn’t a big thing when I got here. Along the way, I've been able to take classes and learn about rapidly growing areas.”
The question now, as Gertner sees it, is “just how far can we take things with these machine learning model-based approaches?”
When she’s not pursuing better AI or spending time with family, Gertner enjoys what she calls her “odd collection of hobbies.” She sings in a chorus. She lifts weights. She’s also an “obsessed knitter,” which helps her focus.
“Right now, I'm learning to play the mandolin. I love to learn new things.”
Interested in solving problems for a safer world? Join our community of innovators, learners, knowledge-sharers, and risk takers. View our Job Openings. Subscribe to our MITRE 360 Newsletter.