In early 2023, Robert Bray, associate professor of operations, was preparing to teach his MBA class on data science when he began hearing more about the new generative artificial intelligence platform ChatGPT. Out of curiosity, he typed a tough question about data science into the interface. “This question took MBA students two hours to solve,” he says. ChatGPT did it almost instantly.
As the implications of generative AI’s power dawned on him, he realized that large language models such as ChatGPT threatened to make his course obsolete.
“It felt like a nightmare,” he recalls. But instead of lamenting the new technology, Bray decided to embrace it. Two months before the class was due to start, he totally redesigned the course to teach students how to use generative AI to help process and analyze data. At the same time, he integrated the technology into homework assignments, training AI to interact with students to help them learn rather than just providing the correct answer.
“I was like, let’s make the class this crazy experiment,” says Bray. To his surprise, he found that students who delegated functions to ChatGPT not only enjoyed assignments more but also were able to do higher-level analysis.
During the three and a half years since generative AI upended all aspects of society, business schools have faced the challenge of responding to the transformative technology. Kellogg professors have met that challenge head-on, turning it into an opportunity to turbocharge their teaching. Across all areas of the school, they’ve integrated generative AI into their coursework, using AI to better teach business concepts while also teaching students how to use AI in business.
“From an academic perspective, we should be teaching about new technologies,” says Jan Van Mieghem, deputy dean and A. C. Buehler Professor. “And it’s also what the market requires. We’ve gotten strong signals from students that employers are asking about it.” Since fall 2024, Van Mieghem has spearheaded an effort, along with dozens of faculty members, to systematically incorporate AI into the classroom, including introducing the school’s first new course code in decades, AIML (artificial intelligence and machine learning).
Rather than create a new department for AI, Kellogg has seeded the topic across all seven existing departments. Faculty in five of the seven departments simultaneously teach a class about the fundamentals of AI, each bringing their own perspective. “Students can tailor their experience based on their interests,” Van Mieghem explains. “If they are going into marketing, they can take an introduction to AI with a focus on marketing applications.”
At the same time, the school has provided faculty members with free access to ChatGPT and Anthropic’s Claude and introduced lunch sessions in which faculty can share the innovative ways they’ve been using AI tools for learning.
Daniel Savage uses digital tools, including pen plotters, to create works that feel both mechanical and handmade. Depicted above, a professor navigates teaching with AI.
Using AI to teach
Sébastien Martin, associate professor of operations, had been thinking about AI for years before the advent of generative AI models, designing complex algorithms to optimize business systems. So when ChatGPT and other LLMs came along, it was only natural for him to experiment with them in his courses. “The technology is insanely powerful, but we don’t know how to use it,” he says. “The main bottleneck is our creativity, not AI’s capability.”
The mistake that many teachers make, Martin says, is to treat generative AI as an adversary that gets in the way of learning. He acknowledges that AI has had an impact on traditional homework by allowing students to input questions and get automatic answers. “It’s broken the link between what a student writes and whether they have thought about it or not,” he says. But that just means teachers must think creatively about how to redesign lessons using AI to help students think through problems. “We have to reinvent what we do.”
Martin began by creating AI teaching assistants that guide students through learning experiences with interactive conversations to teach core concepts. A favorite technique is something he calls “teach the AI,” in which he trains the LLM to pretend it doesn’t know anything about a topic. Students must teach it a core concept — a sort of digital version of the Socratic method. Not only does this help students show their knowledge of ideas, he says, but it can also personalize instruction by asking different questions depending on their level of understanding.
“The biggest constraint for a professor is that you need to create an experience that will be consumed by many people who are not at equivalent levels,” says Martin. “This technique offers personalized education as a private tutor would.”
Bray has also incorporated customized AI tutors in his courses, and in the spirit of experimentation, he and Martin pitted their tutors head-to-head against basic ChatGPT. In an experiment involving nearly 18,000 homework questions, students reported that AI tutors were both more enjoyable and more helpful than ChatGPT.
In addition to creating AI teaching assistants, Martin has redesigned several business cases, training AI to role-play various characters inside a company. “Students can attend a board meeting and interact with executives, even though it’s just AI talking,” he says. He has also taught workshops to help other faculty design their own teaching assistants, and he created a Kellogg AI tutor (affectionately known as Kai) that faculty can access through the campus intranet.
I encourage students to use AI for all assignments and then write reflections on what worked and didn’t.
— Matt Groh, Assistant Professor of Management and Organizations
Jim Lecinski, clinical professor of marketing, has also designed cases using generative AI, creating role-plays for students in which they interact with AI personas representing a chief marketing officer, director of design and other positions within a company. He also uses AI himself in preparing to teach new cases to students. “Teaching a brand-new case is perhaps the single most harrowing high-wire act for business school professors,” he says. Unlike in a lecture, where a teacher controls the flow of conversation, students can take a case discussion in many different directions.
To better anticipate the discussion, Lecinski creates AI personas for himself and his students and uploads a case, watching AI play out the discussion. He then runs the simulation again with just the student personas, interacting with them himself as AI coaches and critiques him. This gradually increases his confidence and effectiveness in presenting the material. “By the time I walk into the classroom,” he says, “I’ve got this.”
Lecinski has also invited AI to chime in during class discussions using Hume’s Empathic Voice Interface, a voice-activated AI tool. “I’ll say to students, ‘Let’s see what AI thinks about segmentation or positioning,’ and it comes through the speakers in the room,” he says. The technique offers a third-party point of view that can sometimes lower the stakes and defuse tension during discussions, he says, since the professor and students aren’t confronting each other directly when they disagree.
As the power of generative AI has grown, so have the opportunities to use it as a teaching tool. Matt Groh, assistant professor of management and organizations, has long focused his research on the interactions between human and artificial intelligence through his Human-AI Collaboration Lab, and now he’s using AI in the classroom as well. “I encourage students to use AI for all assignments and then write reflections on what worked and didn’t,” he says.
To teach the concept of gradient descent, the fundamental algorithm that enables machine learning, Groh designed an interactive video game for students to play. “They learn one of the core concepts of AI, and at the same time they can see how easy it was for me to build this game in an hour,” he says.
He also used AI to design pop quizzes and then had students critique the questions and answers the LLM provided. The exercise showed students both the power and the limitations of AI, especially when answers were more subjective — showing how human intelligence can still be superior when it comes to nuance and creativity.
Groh isn’t the only Kellogg faculty member to create a learning game. Brian Uzzi, the Richard L. Thomas Professor of Leadership and Organizational Change, was one of the first business professors at Kellogg to teach a course on AI when he introduced Human and Machine Intelligence in 2016. From the very beginning, he says, students were always asking who was smarter — AI or humans?
Recently, Uzzzi decided to put that question to the test with an in-class simulation exercise called Beat the Bot (BTB), co-developed with Dawei “David” Wang ’19 MS, ’22 PhD, a sociologist and innovation researcher at Hong Kong University. Over 90 minutes, teams of students race against AI to perform a series of dynamic tasks requiring decision-making and creativity. Along the way, they can see the relative strengths and weaknesses of AI in action — performing creative tasks better and more quickly than the average person, but not as effectively as the best performers. Ultimately, they collaborate with the AI to solve problems together.
“BTB shows students that in using a bot they may be lowering their own creative potential by accepting average answers,” Uzzi says. On the other hand, he says, using the bot to help think through a problem can focus human intelligence and creativity. “It provides a template within which you can add your own unique ideas, experiences and emotions to create a novel solution, product or service that stands out from everyone else.”
My point is to break through the imposter syndrome and show them we can all be AI experts.
— Sébastien Martin, Associate Professor of Operations
Teaching to use AI
Using LLMs to enhance teaching is just one part of Kellogg’s AI efforts. Across disciplines, faculty members are also preparing students for an AI-powered world by teaching them how to use the technology in the workplace. Leigh Thompson, J. Jay Gerber Professor of Dispute Resolution & Organizations, teaches a tiered approach in her Negotiations in the Virtual World class.
“We can’t teach the same negotiations course we taught five years ago, or even three years ago, because AI negotiations are here whether we want to admit it or not,” says Thompson. “The real challenge is, how do we as humans stay in the driver’s seat?”
The most basic use of AI for negotiations, which Thompson calls level 1, is as a wordsmith and communications coach — for example, asking ChatGPT how to respond to a job offer over email in a way that conveys gratitude while still asking for a higher salary. Level 2, Thompson says, is using an LLM for research; for example, researching comparable salaries to negotiate a better offer. Level 3 uses generative AI as a simulator to role-play scenarios, such as preparing for an in-person job negotiation. Finally, level 4 students build their own “negotiation bot” to do their negotiating for them. And that is exactly what Thompson challenges students to do.
Even then, however, she teaches students not to take their hands off the wheel. “Don’t go skiing when your AI bot is negotiating,” Thompson says. AI negotiation bots have been known to lie to get a better deal, opening their creators to legal liability; on the other hand, they can be manipulated by a savvy human into overriding their programming and giving away too much. “You have to anticipate that humans are going to change their behavior once they know they’re negotiating with an AI bot,” she says. “The gloves are off, so you have to plan for that.”
In Thompson’s course, students confront both ethical and practical issues around transparency and honesty as they create their own custom AI bot to negotiate on a specific issue, whether it’s a sale on Facebook Marketplace or a family inheritance. She says the experience prepares them to feel comfortable leveraging AI for workplace negotiations at whatever level is required.
Martin’s AI Foundations for Managers course also culminates in the creation of an AI agent that students customize around an area of interest within a five-week timeframe. Unlike platforms such as ChatGPT or Claude, an AI agent is designed to autonomously perform a specific task. “The project has to matter, and it has to relate to their own deep expertise,” says Martin, who has collected student projects into an online showcase.
Among them are agents that send investors daily stock summaries, prompt patients to adhere to medicine schedules, turn household conversations into shared to-do lists and help restaurants plan and track menu prices. The most important aspect of the project, says Martin, is giving students confidence that they can create their own AI tools, no matter the eventual use. “My point is to break through the imposter syndrome and show them we can all be AI experts,” he says.
In AI for Marketers, Lecinski’s students use AI for every step of the marketing process. “First, we use AI for market analysis and customer insights; then for segmentation, targeting and positioning; and then to produce new product concepts and ideas,” he says. After that, students learn to use generative AI to produce ads, vibe code a website to launch a new product and build a post-campaign analysis — all for a virtual chief marketing officer.
“Along with these AI hard skills, they also learn interpersonal skills, like how to talk to an executive about AI,” says Lecinski. “A few weeks after they leave us, they are going to be in those meetings. Now they can walk in saying, ‘I’ve been there; I know how this is going to go down because of what I learned at Kellogg.’”
For his AI Foundations for Managers course this year, Groh illustrated the practicality of AI by assigning homework based on OpenAI’s GDPval, a benchmark for evaluating AI model capabilities on real-world economically valuable tasks across 44 occupations. “I asked students to solve these problems with Claude and ChatGPT, with the dual goals of thinking about how relevant they actually are to the real world and how comprehensively the AI solves them,” Groh says. “Most students agreed these problems would have taken days, if not weeks, for a professional to complete without AI. Sometimes AI could solve them in a few minutes, but often it took an hour or so of back-and-forth iterations with the AI and learning by the students to solve these problems satisfactorily.”
In another assignment, students used Claude Code, a more advanced interface that allows users to code by issuing plain-language instructions, to crawl websites to collect data on product manager jobs at AI companies. “They were able to turn that data into a flashy website without ever having coded in their lives,” Groh says. For their final project in the course, students built an AI adoption playbook, using the power of AI to conduct deep research and make tailored playbooks for a variety of real-world companies. As a check on the research, Groh asked students to interview executives at the companies to see how well their reports aligned with the problems those executives were facing.
As for Bray, he’s continued to stay abreast of the latest technology for his data operations course, also transitioning to Claude Code so that students don’t have to learn to code. He says the key is teaching students just enough about coding that they understand what’s happening on the back end. “Previously, knowing 5% of a computer language was like building 5% of a bridge — it bought you nothing,” he says. “Now, knowing 5% or 10% of a computer language is vastly more valuable, helping students code much more effectively.” When generative AI produces errors, Bray says, it’s usually because the user miscommunicated exactly what they wanted the algorithm to do. “It’s like writing a cookbook recipe,” he says. “You have to chunk the analysis into small enough pieces that you can describe them with perfect fidelity so there is no ambiguity. Once you do, the AI agent will be 99.9% accurate.”
Equally important, Bray says, is checking every snippet of code after the fact to make sure the AI interpreted it correctly. Users who can learn that much have a huge advantage over even expert programmers. “Previously, the best data scientists were engineers who had technical knowledge,” he says. “Now that Claude Code has dried up that technical moat, I truly believe the very best data scientists in an organization are the MBAs who have the business sense to come up with the best insights or questions to pursue.”
It’s no longer, ‘What do you know about AI?’ It’s ‘What have you built?
— Jan A. Van Mieghem, A. C. Buehler Professor and Professor of Operations
Looking to the future
As far as Kellogg’s faculty members have come in integrating AI into the classroom, the school is only getting started in its efforts to ramp up the technology, says Van Mieghem. This fall, Kellogg will create an AI support group to help faculty create innovative ways to use AI. At the same time, a team of faculty is examining all 800 business cases in Kellogg’s library to identify which have the most potential to be converted into AI cases that can be used both within the school and beyond.
To supplement the foundations courses, Kellogg is launching three new AI courses this fall: AI Consulting Lab, which will teach students how to use generative AI in consulting work; Building AI Companies, which will help students create new startups; and AI Lab, which focuses on building AI agents within the context of existing companies. “I’m a huge believer in experiments,” Van Mieghem says about the new courses, which will continue to develop and change along with AI technology. “If an experiment doesn’t go well, we’ll learn from it, and if it does, we’ll adopt and institutionalize it.”
Just in the past few years, he says, workplace expectations have risen. Instead of hiring employees who can engineer prompts for an existing LLM, companies are looking for those who can use generative AI to design their own custom bots and agents. “It’s no longer, ‘What do you know about AI?’” Van Mieghem says. “It’s ‘What have you built?’”
Kellogg students, in every discipline, are becoming increasingly able to answer that question, as they learn how to use AI in every aspect of their business school education — and beyond.