ASU alum wins grant to advance AI using humanities-based theory

When Anna Guerrero was a PhD student in Arizona State University's biology and society doctoral program in the School of Life Sciences, she analyzed nearly 400 years’ worth of image data collected by cell biologists and microbiologists in order to understand the evolution of knowledge about cells.

The collection included drawings, diagrams, graphs, photographs and images made with microscopes, all of which have had important roles in shaping how people have come to understand otherwise invisible entities, like bacteria, over time.

Four-hundred years is a long time, but the number of images was small enough for one PhD student to analyze alone.

“There’s no way I could do the same kind of work (modern scientific theory) by hand with the amount of image data that exists now,” explains Guerrero, who just began a research fellowship with the University of Michigan’s School of Information.

“I need computational tools like machine learning algorithms to find similar patterns among larger numbers of images.”

Many scientists are turning to machine learning and AI to learn about the enormous amounts of data available to them. While Guerrero understands the benefits of doing so, she also worries about what information scientists and other researchers might miss when relying on AI. That’s why, in her current fellowship, she hopes to create frameworks for researchers who use computational tools to gather and analyze image data to have a more nuanced understanding of where that data comes from and what it means.

To do that, she and her collaborator, Abigail Jacobs, a professor at the University of Michigan, will investigate how AI systems are interpreting image data using a theory from the humanities. They will ask AI chatbots to interpret scientific image data and evaluate the outputs. Guerrero and Jacobs hypothesize that the information chatbots can “read” from an image is limited, and they will use ideas from fields like art history, philosophy and semiotics to make the bounds of those limitations explicit.

“Being able to probe what these algorithms point to in an image, or what is being identified as important for certain sets of pictures is sort of the task,” Guerrero explains, “And for the most part, that’s black-boxed. And so we’re trying to understand what’s happening in the box.”

Through that work, Guerrero hopes to create tutorials to help other researchers think critically about the image data they use.

“The main thing we would like everyone who thinks about images to take away is to just ask, 'What’s in an image?'”

Ultimately, Guerrero’s larger motivation is to build the general public’s information literacy in an era when it’s harder than ever to know whether what you see online is real.

“I think people are pretty unprepared to live in a world where images they see don’t mean what they think they mean,” she shares. “So it’s pressing to help people build that literacy ability. And I think frameworks that come from the humanities are the best way to sort of think about what pictures mean — when, who made them, and where they come from.”

In addition to starting her fellowship funded by a grant from Schmidt Sciences’ Humanities and AI Virtual Institute, Guerrero also won a Whitman Fellowship from the Marine Biological Laboratory in Woods Hole, Massachusetts, to design an exhibit about innovation in microscopy technology.

As a PhD student, Guerrero created two exhibits for the laboratory; the first exhibit displayed images that advanced theories of what a cell was between the late 1600s to present day, and the second exhibit shared information about some of the laboratory scientists who made those images.

The third exhibit challenges the narrative that technology advances in a linear fashion.

“The central message is about how scientists are making choices," Guerrero says, "and that’s why technology advances in certain directions and not others. And that’s important for scientists to see.”