As artificial intelligence (AI) continues to reshape industries across the globe, higher education finds itself at a pivotal crossroads—especially in the fields of science, technology, engineering, and mathematics (STEM). One of the leaders driving this transformation is the AI4STEM Education Center at the University of Georgia (UGA), where researchers and educators are leveraging the power of AI to enhance teaching, learning, and assessment in science education.

At the heart of this effort is a bold vision: to develop tools that can help teachers engage students in authentic scientific practices while reducing their workload. “Teachers are under pressure to align their instruction and assessment with next-generation science standards,” says Xiaoming Zhai, PhD, associate professor in the Department of Mathematics, Science, and Social Studies Education and affiliated assistant professor at UGA’s Institute for Artificial Intelligence. “But if they don’t have time-saving tools—like automated feedback or AI-supported scoring—they’re less likely to implement high-quality assessments, and students lose out.”

From Grading to Guidance: AI as Teaching Assistant

Much of Zhai’s early work at the center focused on automating the assessment process. He and his team built machine learning systems capable of evaluating students’ scientific performance—such as written explanations or data modeling—with a level of nuance that previously required expert human scorers. These tools allowed teachers to receive immediate, actionable feedback about how students were engaging with core scientific practices.

But automation was only the beginning.

Today, Zhai is pushing the boundaries with the development of “GenAgents”—AI-powered multi-agent systems designed to interact directly with students and educators in real time. These agents are being trained to support students in complex scientific tasks such as argumentation, data analysis, and explanation-building.

Imagine two students debating how to interpret a set of experimental results. One believes a chemical reaction caused a change in temperature; the other attributes the shift to a measurement error. A GenAgent could step in—not just to correct them, but to play the role of moderator, challenger, or facilitator, depending on what’s needed.

“For high-performing students, the agent might push them to defend their reasoning more deeply,” Zhai says. “For those struggling to stay on task, it might help redirect the conversation back to the scientific content.”

Addressing Equity and Bias

As powerful as AI can be, its integration into education raises critical questions about fairness and transparency. Zhai’s team is proactively studying these concerns, especially in the realm of assessment.

“Bias is something we take seriously,” he says. “We want to ensure AI-generated scores are as fair—if not fairer—than human scoring.” To that end, his team compares AI scores with those produced by panels of trained human experts, using these “gold standard” benchmarks to evaluate and refine their models.

They also explore how AI systems perform across diverse student populations, including gender and English language learners. One line of research, for instance, examines whether AI scoring amplifies or reduces disparities commonly seen in human grading.

“We’re not just looking at whether one group scores higher,” Zhai explains. “We’re analyzing whether AI is exacerbating existing differences or helping to close those gaps.”

Building the Next Generation of Interdisciplinary Experts

Central to UGA’s vision is its commitment to interdisciplinary collaboration. The AI4STEM Education Center brings together experts in STEM education, learning sciences, computer science, and AI to work toward a common goal: building tools that are not only technically advanced but also pedagogically sound.

A typical project team might include faculty developing science content, engineers designing learning platforms, and AI specialists creating the underlying architecture of the GenAgents. One current initiative includes more than 40 contributors, each playing a vital role in research, design, development, and testing.

But Zhai’s ambitions go even further.

“We’re not just doing research,” he says. “We’re training the next generation of scholars—doctoral and graduate students who will leave with expertise in both AI and education. That’s just as exciting to me as the technologies we’re building.”

Reaching Underrepresented Learners

Another major initiative at the center is the siSTEMas project, funded by the National Institutes of Health. This program uses game-based learning to engage middle school students—particularly those from underrepresented backgrounds—in STEM.

What makes this project unique is its multilingual approach. The team is developing both English and Spanish versions of their science learning games to ensure accessibility for bilingual students. “We often see performance differences tied more to language proficiency than science understanding,” Zhai says. “By reducing language barriers, we aim to make high-quality learning environments available to all.”

Barriers to Broader Adoption

Despite promising advances, several challenges remain in scaling AI integration in STEM classrooms.

First, public perception of AI, especially concerns amplified by media coverage, can influence how willing teachers and parents are to adopt these technologies. “Many people hear more about the risks of AI than the benefits,” Zhai says. “That creates hesitation.”

Second, there’s a pressing need for professional development. Teachers must not only learn how to use AI tools but also develop the judgment to integrate them ethically and effectively. Zhai’s team is currently building a framework to support this process, helping educators evolve from AI observers to confident co-creators of AI-enhanced lessons.

Finally, interdisciplinary collaboration, while essential, is inherently complex. It requires not only administrative support and infrastructure but also a cultural shift in how academic disciplines value each other’s contributions.

The Road Ahead

As the AI4STEM Education Center continues its work, its mission remains grounded in a deeply human goal: empowering educators and students alike to thrive in an increasingly complex world.

“I’m a teacher at heart,” Zhai says. “And to me, preparing the next generation—not just with content knowledge, but with the tools and confidence to think critically and creatively—that’s the most important thing we can do.”

With projects like GenAgent, siSTEMas, and AI-powered assessments on the horizon, the future of STEM education looks more interactive, inclusive and intelligent than ever before.