I study the convergence of two complex systems: the human brain and machine intelligence. For the first time, we have built artificial systems that approach human-level performance, yet both distinct forms of intelligence remain largely mysterious. My research seeks to open these “black boxes” by building a two-way bridge where AI and neuroscience mutually inform and advance one another. My work spans three main areas:

Interpreting and Designing Human-Aligned AI

First, I work to make machine intelligence more transparent and cognitively plausible. I analyze the internal structures of language models to understand how they represent information, reason, and generalize. Beyond interpretation, I use these insights to actively design AI architectures that align more closely with human perception and cognition. By bridging the gap between cognition mechanisms and artificial design, I aim to create systems that perform like humans while thinking in ways that are robust and interpretable.

AI as a Window and Tool for Neuroscience

Second, I leverage AI to accelerate our understanding of the brain. I treat advanced AI models as computational hypotheses and align their internal representations with neural data. This allows me to test which features best explain human cognition across domains ranging from language to mathematics. Furthermore, I am expanding this synergy by employing LLM Agents to assist in the scientific process itself. These agents help automate complex reasoning and hypothesis generation in neuroscience research.

From Brain Foundation Models to Clinical BCIs

Finally, I pursue a translational pipeline that connects basic science to patient care. My work operates on two levels. Upstream, I develop brain foundation models. These are large-scale systems trained to learn universal representations of neural activity. Downstream, I adapt these powerful foundation models for specific clinical applications. By combining these models with invasive electrophysiology, my collaborators and I decode speech and language to power next-generation Brain-Computer Interfaces (BCIs). Our ultimate goal is to restore communication for people with severe impairments.