AAAS Explores What Artificial Intelligence Teaches Us About Ourselves

New Approaches to Machine Learning Offer Insights into Human Experience
Kunda directs Vanderbilt’s Laboratory for Artificial Intelligence and Visual Analogical Systems, and is a deputy director of the Vanderbilt Center for Autism and Innovation. | Stephen Waldron/AAAS

An artificial intelligence innovator and a professor of moral theology and ethics cast a unique light on profound questions concerning how we as humans experience the world, think and relate to technology at a public lecture sponsored by the American Association for the Advancement of Science Dialogue on Science, Ethics, and Religion (DoSER).

Speaker Maithilee Kunda, assistant professor of computer science and computer engineering at Vanderbilt University, works in artificial intelligence (AI), exploring how visual thinking contributes to learning and intelligence, and focusing on applications for individuals on the autism spectrum. Demonstrating an example of visual thinking during her presentation at AAAS, she asked audience members to close their eyes and count the number of windows in their homes, a task most indicated they accomplished by mentally “walking through” the rooms of their houses.

Her example illustrated a kind of intelligence that has been neglected in A1, which is currently based more on linguistic or verbal intelligence. Her example also served as a demonstration for speaker Paul Scherz of how human beings are much more than just complex computers, at least partly because we have physical bodies, or are “embodied.”

“As in Dr. Kunda’s example, when counting the windows in your house, many of you were walking through the house,” said Scherz, assistant professor of moral theology and ethics at the Catholic University of America. “We engage space, even in memory, as people with bodies that move through the world.

“Our embodiment shapes how we engage the world and leads to our concern about the world, and is part of our fundamental care for ourselves and others.”

Kunda and Scherz spoke to an overflow crowd at the annual DoSER holiday lecture, held December 4. The presentation was titled “Of Minds and Machines: What Artificial Intelligence Tells Us About Ourselves.” Like all of the events and programs organized by DoSER, the lecture was designed to build communication between the scientific and religious communities, said DoSER Director Jennifer Wiseman, adding that DoSER’s mission fits within the commitment of AAAS to “relate scientific knowledge and technological development to the purposes, interests and concerns of society at large.” Wiseman called AI one of technology’s most dynamic topics, “a very interesting social experiment impacting our view of what it means to be human.”

Kunda chose the direction of her research career after reading Thinking in Pictures by Temple Grandin, an animal science professor who is autistic and has the ability to problem-solve—and even to visualize huge pieces of machinery before they are built—using her exceptional visual thinking skills. Fascinated with the idea of building better AI systems by incorporating visual intelligence, Kunda and her graduate students set about identifying the building blocks of visual thinking, experimenting with object rotation, scaling and combining. They also began studying how the ability to think visually is learned, employing wearable cameras to track the attention of study subjects as they figured out how to construct colorful patterns out of the component pieces of those patterns, for example.

Kunda pointed out that improved visual intelligence could help engineers design more complex machines, and could also help surgeons—who depend on visual thinking to be aware of, and avoid injury to, organs that are not in view as they operate—to develop more complex surgeries, as just two examples. The challenge to applying such thinking to AI, she said, is knowing how to develop a computational explanation of the process of visual thinking.

“Can we actually understand the computations that are happening when someone is problem-solving this way?” Kunda said. “Can we tease apart the building blocks?”

Using such visual thinking “operators” as rotating, combining and subtracting images, Kunda and her lab were able to program an AI system to match average human scores on a test of reasoning ability. The test, Raven’s Progressive Matrices, usually entails both visual and symbolic reasoning, although brain imaging of study subjects with autism showed they were able to score well using only the areas of their brains associated with visual processes.

Scherz, in his presentation, warned against “disembodied” AI theory that stemmed from Cartesian philosophy, in which the human body is just a mechanism, and human experience can be demeaned. Scherz cautioned against the threats posed by AI to social justice through such phenomena as the replacement of imperfect humans in the workplace with maximally efficient and profitable machines. He cited Catholic social teaching that advocates every individual’s right to work, not only to earn a living but also for personal growth and fulfillment.

Scherz said, however, that he sees Kunda’s work as pushing against such trends as it explores the “complexity and diversity” of human thought, even in such cases as that of Grandin, “who lacks many of the skills seemingly necessary for today’s workplace because of her autism, yet she brings to bear a different way of thinking that more than makes up for that lack.”

“I think that as long as one uses these tools in a humble exploratory manner, as Dr. Kunda does, seeking to understand how human thought works,” said Scherz, “it can be very helpful, by expanding our understanding of human thought, our understanding of ourselves.”

[Associated image: Stephen Waldron/AAAS]