Stanford backs five projects linking AI to infant learning, therapy and brain biology

Infant-viewpoint video, psychotherapy conversations and personal experiences of smell will become inputs to five Stanford research projects linking AI with neuroscience. Stanford’s Wu Tsai Neurosciences Institute and Institute for Human-Centered Artificial Intelligence have awarded a second round of joint grants, backing efforts to understand the brain, guide treatment research and make AI learn more like people.

The November 11 announcement brings together studies of depression, infant vision, neuronal cell types, psychotherapy and smell. The projects give AI different jobs: analyzing conversations, connecting brain measurements and modeling biological differences. Other work runs in the opposite direction, asking whether the way people learn and experience the world can inform AI design.

Learning from a baby’s view—and a person’s sense of smell

Cameron Ellis, Michael Frank and Dan Yamins will combine video taken from a baby’s point of view with functional magnetic resonance imaging, or fMRI, brain scans. Their planned AI system will train on both kinds of data from an infant. The goal is to study how babies learn while exploring whether AI can learn more efficiently.

Stanford frames the project around a learning gap: infants learn quickly despite encountering a small, biased fraction of the data used to train AI systems. The proposed model would bring their visual experience and brain data into the same training process, with the aim of making it learn more like an infant.

Todd Coleman and Emily Fox will investigate another part of human experience: smell. They plan to use electromagnetic brain recordings and large language models to connect neural activity with individuals’ subjective experiences of odors. Their premise is that smell offers a way to study intelligence shaped by personal experience and emotional context, rather than pattern recognition alone.

Connecting treatment conversations and brain measurements

Corey Keller and Manish Saggar will study how psychotherapy dialogue relates to brain activity in real time. They will record electrical activity from inside the brains of patients who are also undergoing epilepsy monitoring while those patients participate in therapy sessions. Large language models will characterize the dialogue; mathematical methods called topology will help map changing brain states.

The researchers’ long-term goal is a neural atlas of psychotherapy: a map linking elements of therapeutic conversations with brain dynamics that could inform more effective treatment. The immediate investigation concerns those relationships, rather than an AI system delivering therapy.

Cammie Rolle and Ellen Kuhl will focus on depression subtypes identified through brain imaging. They plan to use AI to connect fMRI results with electroencephalography, or EEG, another type of brain scan. They will then evaluate whether EEG alone, or a combination of methods, can refine those subtypes and help guide better treatments.

Explaining why brain cells become different

The fifth project turns to a developmental puzzle: how a human brain produces many kinds of cells even though those cells are essentially genetically identical. Marius Wernig and Anshul Kundaje will draw on genetics and AI to build a foundation model of neuronal cell types. Their target is the molecular basis of differentiation—the process by which cells develop distinct identities.