Reading Artificial Minds: Hudson Golino Brings Human Insight to Machine Intelligence

Artificial intelligence is often described as a black box, one that’s conspicuously inscrutable. Hudson Golino is determined to open that box.

An associate professor of quantitative methods in the Department of Psychology, Golino works at the intersection of psychology, artificial intelligence and network science. His research asks simple questions with far-reaching consequences: How do humans and machines represent, organize and transform information, and how can we make those processes transparent and measurable?

“A central theme of my work is interpretability…building AI systems that don’t just perform well but whose internal structure and behavior can be understood, evaluated and aligned with human cognition,” Golino said.

Ultimately, Golino studies how meaning takes shape inside a system — tracing how ideas form, connect and evolve as a product of machine architectures and human input. That perspective has placed him among a growing group of scholars linking psychological theory to modern AI, especially large language models.

An Interdisciplinary Home in Arts & Sciences

Golino believes that the College of Arts & Sciences’ efforts to promote interdisciplinary collaboration positively impacts those doing boundary-crossing research. The college’s emphasis on intellectual flexibility has allowed him to connect quantitative psychology with computer science, data science and cognitive science, while staying grounded in core psychological theory.

“This flexibility has been essential for developing research programs that really span traditional disciplinary boundaries, especially when you’re working with emerging AI technologies,” Golino said.

His perspective is one the leadership of the College underscores.

“In Arts & Sciences, we believe the most important breakthroughs happen where disciplines meet and values guide discovery,” said Christa Acampora, Buckner W. Clay Professor of Philosophy and dean of Arts & Sciences. “Dr. Golino’s work shows how AI can be both intellectually ambitious and deeply responsible, advancing technology while keeping human understanding at its center.”

That environment has also allowed Golino to focus on building tools that others can use — not just publishing findings but creating infrastructure that lowers the barrier to advanced research.

Support from the college, and private foundations (like the Jefferson Trust) has helped him develop a strong open-science ecosystem in his lab. His team creates software that makes sophisticated methods usable in everyday research and teaching contexts, extending the reach of AI and psychometrics well beyond a single discipline.

“Our lab builds free, open-source software so researchers, students and practitioners worldwide can use advanced methods without prohibitive technical barriers,” Golino said.

Research at the Frontier of AI and Measurement

Among Golino’s biggest projects is AI-GENIE, Automatic Item Generation and Validation via Network-Integrated Evaluation. The system uses AI and data-driven psychology methods to automatically create and test survey and assessment questions.  

“What once took months can now be done in hours, while preserving rigorous psychometric standards grounded in network science and information theory,” Golino said.

The tool can be a game-changer for scholars who don’t have the resources to build effective measurement tools from scratch, explained Lara Lee Russell-Lasalandra, a third-year doctoral student in quantitative psychology who came to UVA to work specifically with Golino. AI-GENIE brings structure to what can seem like an untamed intellectual frontier.

“But it doesn’t have to be the Wild West,” Russell-Lasalandra said. “We can establish frameworks and quality checks and best practices with a scientific mind and scientific rigor.”

Another of Golino’s major contributions, Construct Emergence Tracing, or CET, tackles one of AI’s most persistent challenges: explainability. CET maps how abstract concepts emerge across the layers of transformer models, revealing when and where meaning becomes organized inside an AI system. By combining network analysis with information-theoretic metrics, the method offers rare insight into how “understanding” develops inside a machine and how that process parallels human cognition.

Together, these efforts form the foundation of what Golino calls generative psychometrics, a new framework for studying measurement, cognition and intelligence in both humans and machines.

Teaching AI with Purpose

Golino’s mission also involves bringing his research directly into the classroom. His new course, Large Language Models with Applications in R (PSY 5500), is open to advanced undergraduates and graduate students. It blends hands-on technical training with critical analysis of AI’s capabilities.

“Students need to learn not only how to use large language models, but how to interrogate their limitations, biases and internal structure,” Golino said.

Since 2022, Golino has incorporated large language models as “programming co-pilots,” helping students gain confidence in coding, statistics and data science while learning when not to trust automation.

Students learn to design automated pipelines for text analysis, item generation and validation entirely within R, an opensource programming language and software environment designed specifically for statistics, data analysis, and data visualization, using tools they can take into research, graduate study or industry.

According to Russell-Lasalandra, the approach is essential to preparing students for the world they will encounter after graduation.

“[Hudson] gives students the tools, the foundation, and the opportunity to use them. That’s very close to how research really works,” she said. “You have a safety net if you need guidance, but you also have the opportunity to explore creative ways to find solutions on your own.”

Looking Ahead

Golino’s current work explores how abstraction, depth and even “wisdom” in language might be quantified using network structure and algorithmic complexity, and he’s developing adaptive, AI-generated assessments that evolve dynamically while remaining psychometrically sound.

He’s also working with second-year systems engineering major Iain McDonald Ramirez to develop an open-source project that evaluates the accuracy of agents built with large language models and their ability to generate accurate data from unstructured web content. An opportunity Ramirez hopes will be a springboard into a career that bridges the gap between science and industry.  

As AI becomes increasingly entrenched in the work of science, education and society, Golino’s research reflects the increasingly important role Arts & Sciences plays at UVA: grounding technological innovation in human ethics and values and bringing a deeply interdisciplinary approach to using technology to change lives for the better.

“We can’t compete with private labs like Google or OpenAI,” Golino added, “But we do have something powerful here at the College: We have the people with the time to research and think deeply about these things.”