Early-Career Spotlight: Patrick Emami Works Toward ‘Humans in the Loop’ With AI

Aug. 12, 2026 | By Anna Squires | Contact media relations

Share

Patrick Emami is interested in things that make life feel rich and full: making music, indulging in scientific curiosity, and … artificial intelligence?

Many people wonder if artificial intelligence (AI) could undermine or replace human creativity and ingenuity in scientific fields. But Emami, a computational science researcher who focuses on machine learning, reinforcement learning, and agentic system design, believes that a “cointelligence” between humans and AI could make scientific research even more rewarding.

Cointelligence, Emami said, stems from a vision of “collaborative intelligence” pioneered in the 1960s by computer scientist J.C.R. Licklider. Licklider envisioned a deeply integrated partnership where humans set goals and perform evaluations while computers handle routine processing, allowing them to solve complex problems together.

Now, Emami is developing an AI assistant that acts like a research collaborator through the Theseus project.

“The goal is not to give AI a task to complete autonomously and then ask a person to pick up the pieces afterwards,” Emami said. “The best results are probably going to come from people who co-work alongside AI systems.”

Deep Learning

Emami first became interested in the interplay between people and computers when he got into music—specifically the electric guitar, back in high school.

While he originally wanted to pursue a career in audio engineering, he switched to a bachelor’s degree in computer programming at the University of Florida. That gave Emami the opportunity to explore reinforcement learning strategies for small mobile robots, intelligent traffic intersections, and autonomous vehicles.

His first project involved teaching a small robot to recognize human gestures, then execute different workflows based on what it saw.

“I fell in love with it,” Emami said. After earning a Ph.D. in computer science, that love brought him to the National Laboratory of the Rockies (NLR), where in 2021 he took a graduate internship to work on reinforcement learning for intelligent transportation systems under Qichao Wang and Juliette Ugirumurera. He focused on designing a reinforcement learning agent that could safely and autonomously change traffic lights in response to variable traffic demand.

“At the time, the application of deep learning techniques in the energy systems space was fairly underexplored, so it seemed like a good time to dive into it,” Emami said.

Humans ‘in the Loop’ With AI

Now a full-time researcher in the AI, Learning, and Intelligent Systems group at NLR, Emami researches advanced topics surrounding large language models (LLMs), including multimodal LLMs for scientific data, human-agent collaboration, uncertainty quantification, planning and reasoning, and benchmark design. He also develops conversational agentic systems, like the GATES Assistant, that integrate different generative AI models in accessible, user-friendly ways.

Emami spends much of his time thinking about “LLMs that are being repurposed to take action in the world.” At the moment, he is immersed in developing the ScienceHands platform, part of the NLR-led Theseus project funded by the U.S. Department of Energy Office of Science’s Advanced Scientific Computing Research program.

“We’re bringing to life a human-centric vision of how scientists could work within human-agent research teams on complex scientific projects,” Emami said.

As modern LLMs become more capable, Emami explained, they have gained the ability to write code and converse with each other in natural language. This is leading human scientists to ask, “Could multiple LLMs team up with human scientists to solve bigger problems faster by working in parallel?” One way to think of this question is that every LLM has its own sandbox filled with “tools” and bounded by rules and guidelines. If LLMs—and humans—can merge sandboxes, they could tackle bigger problems with a greater variety of tools.

Emami’s vision is for LLMs and humans to work “in the loop” together—which may not require the dramatic paradigm shift being foretold today.

“One question I’m thinking about is, ‘How can we create the most nonintrusive ways of integrating AI into how people work?’” he said. “Let’s say you reimagine the use of AI in your daily work, but you don’t ask, ‘What could I give this assistant to make me more productive and work faster?’ What if instead, you asked, ‘What could I ask this AI assistant to teach me today so that, in a year from now, I’m more of an expert in my field?’”

That, he said, might lead to people having even more opportunities to learn, grow, indulge their curiosity, and become better at the work they do.

“If somebody uses my AI tools and says, ‘Hey, I learned more about this physical phenomenon by using this tool,’ then I'm happy,” Emami said. “Happier than if they told me AI solved a problem for them but they don’t understand how they got there.”

Learn more about NLR’s artificial intelligence research, including focus areas in machine learning and reinforcement learning.