Artificial intelligence has moved from being a distant frontier to a policy battleground. President Trump’s AI Action Plan centers on harnessing “the full power of American innovation” through three pillars: accelerating AI development, building domestic infrastructure, and asserting American leadership in international AI diplomacy. The contrast with former President Biden’s vision is stark. Where Trump prioritizes innovation, Biden’s framework emphasized regulation, including “governing the development and use” of AI through institutions designed to conduct “robust, reliable, repeatable, and standardized evaluations.”
While Washington debates oversight versus acceleration, universities are grappling with something more immediate: how to teach AI, critique it, build it, and prepare students for a world where it’s almost always going to be around. But the important question isn’t simply where universities such as Vanderbilt stand on the innovation-versus-regulation spectrum. It’s the extent to which they are uniquely positioned to contribute to a central debate that neither Washington nor Silicon Valley can resolve alone.
Jesse Spencer-Smith, Professor of the Practice at Vanderbilt’s College of Connected Computing and Director of the Data Science Institute, points to Vanderbilt’s early engagement with federal guidance on regulating large AI models.
Early in the Biden administration, federal guidance began pushing to regulate large-parameter AI models more heavily. Bigger models, the thinking went, posed greater risk. Universities, including Vanderbilt researchers, helped shape that guidance, according to Spencer-Smith.
This thinking became outdated almost immediately. Small models quickly proved just as capable in specific tasks, and it became clear that reasoning ability, not size, was an important variable that needed to be emphasized further. The framework was obsolete before it could be implemented. According to Spencer-Smith, this is an illustration of the core challenge in AI governance. Technology often moves faster than institutions designed to oversee and regulate it.
Spencer-Smith notes that the largest models are primarily built by companies such as OpenAI, Google, and Anthropic, which have resources universities generally cannot match. Still, he emphasizes that academic institutions are powering frontier AI development through open-source tools, theoretical advances, and independent evaluation.
Scholars in Vanderbilt’s Political Science Department, for example, were among the first to seriously test whether large language models could substitute for human focus groups. On the surface, they appeared capable. Yet, when they looked closer, they realized that the process of aligning models for safety fundamentally disrupts their ability to simulate human subgroups accurately. That finding had real implications for researchers and policymakers, according to Spencer-Smith. Importantly, it came from a university, and it wouldn’t have come from a company whose product it calls into question.
The federal tension between innovation and regulation plays out in the classroom as well. Spencer-Smith structures his graduate AI course so students can use AI freely on assignments and exams, except when answering core exam questions. AI is very good at producing explanations that feel thorough, but students can mistake that fluency for true understanding. Students can finish an AI-assisted assignment convinced they’ve grasped something they’ve really only skimmed, according to Spencer-Smith.
Dr. Ole Molvig, Assistant Professor of History and of Science, Technology, and Society at Vanderbilt, raises a different concern, however. He fears that AI is becoming increasingly partisan. People are less likely to form opinions based on their own experience with the technology and more likely to adopt whatever position aligns with their politics.
Both professors agree universities have a responsibility at this moment, not just to engage, but to model a different kind of engagement. Universities have the power to house genuine intellectual diversity across technical and social dimensions. If people don’t talk across those lines, Molvig warns, we see the consequences play out in public life when we think about technology.
Spencer-Smith believes the next leap in AI capability isn’t all about building bigger models. Instead, he believes there will be an increased focus on training models to reason the way domain experts do, in areas such as legal reasoning, medical diagnosis, or historical analysis. He calls current AI reasoning the “McDonald’s of reasoning,” as it is efficient, but shallow. Universities, with centuries of accumulated, specialized expertise, are uniquely positioned to provide the deep, domain-specific knowledge necessary to push AI forward.
Spencer-Smith emphasizes that removing universities from the equation entirely will eventually lead to AI development running out of road. One can push technology only so far before you need new science to go further, and that much of that science happens at universities like Vanderbilt.
For these Vanderbilt professors, then, the critical question isn’t necessarily how to manage universities’ role in AI. Rather, it is whether policymakers are paying enough attention to preserving and supporting the unique contributions universities can make. As they make clear, this does not only occur at the larger institutional level. Instead, it can, and should, start at the individual level, with informed citizens, professors and teachers alike, collaborate and explore across academic disciplines, test claims independently, and remain cognizant of the ways in which AI can simultaneously expand and constrain their abilities.
