Next semester, the course AI and HR will be mostly device-free, with one notable exception: A small white robot with lopsided camera-lens eyes and a custom-built artificial intelligence. 

The course instructor, Scott Seidenberger ’16, postdoctoral associate in human resource studies in the ILR School, is designing the robot and chatbot to illustrate how AI may manifest in the workplace, while his device-free approach to lectures and exams keeps students focused on learning. 

Seidenberger and other instructors across Cornell are working to strike a balance in the classroom – using AI as a teaching tool but also making choices to prevent its use and preserve the learning experience. 

“We know a lot more now than we did two years ago, and we’re learning more all the time,” said Robert Vanderlan ’88, executive director of the Center for Teaching Innovation, which is organizing the fellowship program. “This sort of experimentation is beginning to distill some approaches that can be widely adopted.” 

The classroom robot 

Seidenberger is building his virtual AI agent – named Hopper in honor of computing pioneer Grace Hopper – from scratch, so students can “get under the hood” and see the logic on which it is built, and they can make updates and improvements to it. 

“Companies are deploying AI agents to interact more and more with humans every day and deal with more sensitive human tasks and data,” said Seidenberger, an AI Experimenter Faculty Fellow. “It’s really important for our HR students to be able to learn how AI works at a practical level so they can go into industry and be more hands-on participants in guiding, guardrailing and developing AI, and not just leave it to software engineers to make very human decisions that are very part and parcel of the HR craft.” 

Seidenberger programmed Hopper to be able to answer basic administrative questions about the class itself. It is nowhere near as good as a real person, he said, but giving it a purpose that is relevant to the class will give students real experience using the technology. 

Putting Hopper in a robot body takes the exercise a step farther and shows students another possibility for how AI might be deployed in a workplace.

“We’re all used to the realm of chatbots, but we’re already seeing more capable voice agents or avatar-based agents, where you can have natural plain language conversations in real time,” Seidenberger said. “The next logical step is: What if I take that brain, that capability and I put it in a physical robot? How does that affect the relationship that we as people have with that entity?” 

While Hopper may represent the future of technology, Seidenberger will require students to put away laptops, phones and tablets during lectures to decrease distractions. 

“I have them in class two days a week for 75 minutes for 15 weeks,” he said. “I plan on having an analog classroom so we can all stay focused on honing those cognitive skills together in person.” 

AI but verify 

Suélia Fleury Rosa, an AI Experimenters Faculty Fellow and associate teaching professor of biomedical engineering in the Cornell Duffield College of Engineering, is designing activities for her Digital Control Strategies for Biomedical Applications course where students first complete a part without AI and then use AI to solve another part but evaluate and verify its answers. 

The students in her course will go on to develop and improve biomedical devices like surgical robots and prosthetics, which must synthesize real-time data from the body accurately and quickly. AI is an important tool in the field, so students need to learn how to use it and to recognize when an AI-generated answer may be incomplete, misleading or technically incorrect. 

“A response that sounds plausible is not enough,” Rosa said. “Students need to demonstrate that it is mathematically, computationally and physiologically reasonable.” 

Compared with when she started teaching 30 years ago, Rosa spends far more time preparing for classes now, testing prompts to see how the AI will answer and anticipating problems her students may run into. 

“It’s an amazing tool,” she said. “But it’s a challenge, and it’s changed everything.” 

Conversations with historical figures 

 In Architectural History I, a survey course spanning antiquity to the Enlightenment, students can “chat” with the dead, or at least AI versions of them, thanks to an innovation created by Pamela Karimi, associate professor and associate dean in the College of Architecture, Art and Planning (AAP) and an AI Experimenters Faculty Fellow. 

 Her first-year students study John Evelyn, an architectural visionary and early environmental thinker. In 1661, long before the Industrial Revolution, Evelyn wrote “Fumifugium,” proposing to reshape London to clear its smoky air. 

More than just hearing from Evelyn, Karimi wanted to deepen students’ understanding of his world. 

With the help of Andre Hafner, director of AAP IT Solutions, she developed AI chatbots to play the parts of King Charles II and England’s Parliament, to whom Evelyn addressed “Fumifugium.” His proposals were not put into effect at that time. However, the students imagined the discussions that could have taken place in Parliament and debated with the chatbots as if they were Evelyn about creating green spaces and regulating the smoke-producing trades of the time, like lime burning. In doing so, they better understood the political and environmental challenges Evelyn sought to address. 

Karimi kept the focus of the tool narrow and grounded in trusted primary and secondary sources to create historically informed simulations that students could question and challenge. 

“For architecture, the possibilities are particularly interesting,” Karimi said. “Imagine Leonardo da Vinci or Filippo Brunelleschi being brought into a midterm studio review, not as an oracle telling students whether their projects are good or bad, but as another critical voice in the room; one that places students’ projects in conversation with centuries of architectural thought.” 

Even if the AI actor gets it a bit wrong, there’s educational value in discussing the accuracy of the simulation. 

“Used this way, AI can become a tool for close reading – especially of primary sources – as well as for deeper historical understanding, debate and critical thinking rather than an answer machine,” Karimi said. 

AI and writing 

Writing expert Kate Navickas decided to grapple head-on with AI’s impact on writing, starting in fall 2024 when she launched her first AI and Writing First-Year Writing Seminar. 

“This is a public conversation where I see smart people having very divergent perspectives,” said Navickas, associate teaching professor in the John S. Knight Institute for Writing in the Disciplines (College of Arts and Sciences) and director of Cornell Writing Centers. “For me, that makes a perfect first-year writing course because there are a lot of ways students can join the conversation. It’s safe to be pro-AI. It’s safe to be anti-AI.” 

In the course, students set goals for their writing and work on them in one-on-one conferences with Navickas. They also read about how AI tools work and the biases inherent in how they are built, and they experiment with it in their own writing, reflecting on its uses and limits. 

At the end of the seminar, she expects students to have more confidence in their writing and more agency in the decisions they make around if and how to use generative AI in their writing process. 

Results of an in-class experiment stuck with Katie Lu ’29, a computer science and mathematics major (A&S) who took the course in fall 2025. Everyone in her class gave an AI the same writing prompt and nine out of 10 responses started with the word “both.” 

“The homogenization effect is very severe,” she said. “I realized that if I use AI for my creative work, then I won’t be creative in any sense.” 

Signs of maturity 

As instructors adjust to the new teaching landscape, CTI supports them with individual consultations, training, webinars and department workshops. Last year, more than 1,000 instructors – about half of the faculty – accessed CTI’s experts and programs, many for help related to AI. 

“This is a chance to think about what sort of learning experiences we want for students,” Vanderlan said. “I think a lot of our instructors are doing that in really creative ways and under a lot of pressure, because it does take time to rethink.” 

For instance, after an initial panic about no longer being able to assign essays as homework, because students can “write” them with AI in seconds, some instructors have returned to them but pair them with short in-person conferences, which can reveal a lack of understanding if a student used AI inappropriately. 

“You learn a lot actually talking about your work with a professor or with a TA who can ask good questions and who can challenge your thinking,” Vanderlan said. “That’s actually a rich learning experience if it’s done thoughtfully.” 

Students are also concerned about the loss of learning. Lu said she appreciates reminders not to take shortcuts. 

“It’s really easy for us to not struggle at all and just finish the assignment very smoothly and save time, then go do something entertaining,” she said. “When the professor keeps reminding us or showing us evidence of how the struggle in learning will actually help us improve, I am more likely to be self-disciplined and try to do the hard work first.”