Guiding Education’s AI Revolution

“We are thinking about AI in so many ways,” Dean Strunk said. “There is no shortage of ‘new, shiny’ innovations in the world of AI. But at Penn GSE, innovation isn’t about doing what is new. It is about what is needed and what works in practice, about strengthening teaching, expanding opportunity, and sustaining a healthy democracy.”

Besides the analytics degree, the school has leaned into AI and education through multiple initiatives that include hiring new faculty, collaborating with school districts on professional development, and conducting novel AI-focused research. Professors are creating AI tools that annotate readings or design lesson plans without dated views of education and are developing hands-on ways to expose students to AI’s pros and cons. One researcher is reimagining high-tech AI to advance more equitable access; another is analyzing in real time how teens are using—or not using—AI technology to write. There’s even JeepyTA, a virtual teaching assistant developed by the Penn Center for Learning Analytics and trained on course content, examples of responses, and prompts that can serve as digital, first-line teaching support.

Time is of the essence. While AI has been around since the 1940s, when the programmable digital computer was invented, it was ChatGPT’s debut in November 2022 that grabbed the attention and imagination of the public. Long before ChatGPT’s arrival, however, GSE researchers were already leading the way in AI in education. With each iteration, generative AI has improved its ability to give human-like responses, and as a result, interest in leveraging the technology to improve teaching and learning has exploded. At the same time, the risks generative AI poses are top of mind, whether that’s the environmental impact, the lack of regulation, the use of copyrighted content, the hallucinations and misinformation, the outsourcing of learning, and more.

“Penn GSE is in the frontier of thinking about AI,” said Seiji Isotani, one of two new hires this year, an associate professor who is faculty director of the Learning Analytics and Artificial Intelligence program, and a pioneer in gamified intelligent tutoring systems. “Here at Penn, we’re not just applying AI to education, we’re co-constructing it with educators and learners to address the systemic challenges that affect teaching and learning worldwide.”

In his own research, Isotani redesigns AI tech to bring its benefits to classrooms with limited access to computers or the internet, guided by a human-centered AI design approach that aligns technological innovation with the real needs of teachers and learners. The goal? To effectively integrate AI technologies in classrooms and ensure more equitable educational outcomes.

In one of his projects, Brazilian teachers photographed students’ essays with low-cost cell phones, widely available devices even in the most under-resourced areas, and uploaded the images (when the internet was available) to an AI platform co-created with educators. The AI provided feedback and pedagogical suggestions, equipping teachers with insights into students’ strengths and weaknesses and offering strategies to tailor classroom activities to promote deeper, more meaningful learning experiences.

His colleague, associate professor Shiyan Jiang, GSE’s other new hire, leads in AI literacy—a skill, she argued, in low supply and yet essential as the technology permeates every aspect of life. With two National Science Foundation (NSF) grants totaling $3 million, she developed StoryQ, a web-based machine learning platform that finds patterns in text.

In one study, high school journalism students learned about machine modeling by using StoryQ to analyze Yelp reviews about ice cream shops and predict whether reviews were positive or negative by classifying certain words, Jiang said. That led to discussions about fundamental AI concepts and how the model made its decisions. A new project underway will target elementary students.

Six years ago, AI literacy barely existed, Jiang said, adding that she struggled to find a journal to publish her findings. Now, it’s the opposite. “The hype is, ‘AI can do this, AI can do that,’” she said. “I open the black box on AI, help stakeholders understand the promises and limitations, and am really transparent about how AI makes decisions.”

Educators also need AI literacy, of course. Earlier this year, GSE piloted the Pioneering AI in School Systems (PASS) program with the Philadelphia School District. The first-of-its-kind K–12 professional development project, designed with the district and supported by the Marrazzo Family Foundation, guides educators on AI usage policy and trains them on innovative ways to bring the tool into the classroom, while also addressing ethical considerations.

Often, GSE found that teachers who earned certificates in its Virtual Online Teaching (VOLT) and Experiences in Applied Computational Thinking (EXACT) programs returned to their classrooms enthusiastic to try out new ideas only to meet resistance from higher-ups, said Betty Chandy, GED’05, GRD’13, director for AI and technology-enhanced learning initiatives at Catalyst and the pilot’s co-leader with Golden. PASS aims to solve that problem and serve as a template for districts around the country. Rather than focus only on teachers, as AI workshops typically do, PASS takes a systems-level approach by also holding targeted sessions for district- and building-level leaders, she said.

“When teachers go back into their schools, there’s a vertical alignment,” Chandy said. “The district’s aware, the principal’s aware, and teachers have the space to be cutting-edge.” With a just-awarded $1 million from Google.org, PASS will expand to five more districts, including a rural one, she said.

For example, Chandy said educators can redesign assessments. For a paper, she suggests, let students use ChatGPT as one more tool in their toolkit and then probe them on the prompts they tried and the results elicited, which in turn could lead to discussions on the accuracy of the LLM’s outputs and the ethics of scraping other people’s content. “This is a way to build critical literacy around AI,” Chandy said. Teachers also could produce worksheets personalized to learners’ interests (like a recent Eagles’ game or cupcakes) within seconds with LLMs. “There are ways teachers can bring authenticity to the classroom by connecting to their students’ lived experience. Historically, that has been very hard. I believe ChatGPT allows teachers to bridge that gap.”

Fran Newberg, the Philadelphia School District’s deputy chief in the Office of Educational Technology, said Penn GSE brought a broader perspective. “Marry that with practitioners,” she said, “and it creates a nice balance of research, pedagogy, and practice.”

Amy Stornaiuolo, GSE professor of literacy education and faculty director of the Philadelphia Writing Project, has long studied digital writing. In her groundbreaking study Writing with AI, which concluded in August, she looked at how experienced teen writers are using machine learning to create stories and develop AI literacy, analyzing moment-to-moment screen captures of their writing—a new methodology. Stornaiuolo found the youth participants used AI as a “thought partner” to co-write, brainstorm, or get feedback. Crucially, she added, they prompted ChatGPT within tight boundaries, asking it, for example, to improve clarity but without rewriting their stories.

“AI amplifies and changes the way people are writing,” Stornaiuolo said. “Writers want to do the thinking work. They want AI to have a very specific role.” But novice writers without AI experience, she found, tended to copy and paste the output into their own essays. “They didn’t have any of those boundaries to keep AI in its place.”

“After working with this team, I honed in on my interest—not just public policy but educational policy,” she said. “The experience has been absolutely amazing.”

In the Digital Discourse Project, a James S. McDonnell Foundation–funded, five-year, longitudinal study that wraps up this year, Stornaiuolo is exploring how high school teachers facilitate online literature discussions. One major finding, she said, is that school leaders are often focused on AI’s assistive dimensions, such as correcting grammar or facilitating cheating. But, Stornaiuolo said, her study participants “are much more focused on AI’s creative dimensions.” Two teacher-researchers, for instance, created and programmed the “AI Thinking Partners” bot, which allows students to ask questions of a fictional author while annotating a reading assignment and even program their own author or character bots.

How AI will shape teaching and learning is the number one question the country faces, according to Stornaiuolo. “Penn GSE is focused on offering actionable and meaningful guidance around AI,” she said, “and not just responding to it.”

One concrete project already underway is the development of “babyGPTs.” Yasmin Kafai, the Lori and Michael Milken President’s Distinguished Professor, puts students in the designer seat. From 2003 to 2014, she worked with former colleagues from the MIT Media Lab to create and research Scratch, the popular, kid-friendly platform for sharing games, stories, and animations. Now, Kafai’s seeking to do the same for generative AI. In one project, students can take publicly available datasets—Shakespeare’s work, recipes, or screenplays—and train models to generate information from a prompt, she said. Voilà! babyGPTs.

“You also learn by making it,” said Kafai, who ran a three-day, NSF-funded workshop in the spring that brought together researchers, educators, and industry experts on ways to empower teachers and students as AI creators. “By producing it, they understand the potential as well as the limitations of that tool. We have this tendency, when we see new technology, of being consumers and not understanding these are designed artifacts.”

In one study, 60 teachers in China and Taiwan are testing Craftpad—technology Chen is developing—to help them design lesson plans. His research has found that generative AI models often suggest outdated pedagogical designs that may stifle student voice and creativity. Craftpad, however, is a flexible workspace that centers teachers as designers—keep the human in the loop, as the mantra goes—rather than as mere end users, Chen said.

“We’re not training the model,” he said. “We’ve designed Craftpad to reinject, reemphasize teacher expertise. We want to have room for teachers to express their professional taste and creativity. That’s the philosophy behind it.”

Chen’s latest pilot project, InkSpire, uses AI to support better-designed scaffolds for college science texts that encourage interaction and higher-order thinking. “The instructor will be in the driver’s seat to tweak and deploy reading assignments and what the student sees in the interface,” he said. “On the student side, they can read more effectively and deeply.”

Ultimately, Chen said, the goal must be AI that “augments your learning, not replaces your learning. If we don’t do this mindfully, then we lose our individual, personal authenticity, we lose social context with each other.”

At Catalyst, Golden focuses on bringing people and ideas together to move education forward in novel ways. Clearly, the advent of generative AI has proved a perfect fit.

“AI is just another technology we can use as a tool,” he said. “It’s a new way of doing the ‘how’ in the conversation as we pursue the ‘what’ and ‘why.’”

“There is potential for AI to have transformational impact on education,” he said. “We’re trying to be very thoughtful about something that could increase efficiency and personalize education in new ways that we’ve only dreamed of. At the same time, we’re making sure the downsides are addressed as well.”

Back on Haiying Li’s LLM class on Zoom, student and teaching assistant Andy Pan is agog at the speed and quality of the SAT practice app.

“This is fascinating,” said the corporate strategy consultant based in San Francisco who has worked for Amazon, Disney, and others on instructional design. “No wonder there’s massive displacement in so many domains.”

Pan, a venture capitalist, said he’s pursuing the GSE degree to learn more about the machine learning analysis he often employs as he considers investments.

The program, Pan said, has “really opened my eyes. How do we make sure data integrity is in check, balanced, not just analysis for analysis’ sake. Beneath the surface, there are numerous technical challenges as well as critical judgment calls that must be made by the designer and the resource allocator. These choices can significantly affect the outcome. You’re really an insider, rather than an observer.”

For Penn GSE’s dean, that’s the goal—to ensure that AI isn’t just a technology to be adopted but a tool to be shaped and augmented by teachers and students.

“Penn GSE has long been a place that bridges theory and practice to tackle the thorniest challenges in education,” said Dean Strunk. “AI is already reshaping how people learn and work and fundamentally changing how teachers teach and students participate in their learning. Our responsibility is to ensure that it serves learners and the educators who guide them in a way that is equitable, ethical, and evidence-based.”

Alumni AI Trailblazers

“The way we teach, we’re preparing our kids for the industrial era,” said the founder of the Philadelphia-based Teacher in a Suit, which offers AI literacy training to educators, and the AI innovation lead at Foundation Academy Charter School in Trenton, New Jersey. “Our education system still teaches students to do what AI already does best—read, summarize, and reproduce information. The next generation of learners needs to do what humans must still do best: connect ideas, apply judgment, and turn information into meaning.”

Since launching his business in 2023, Roberts has built, with the help of his Penn GSE Education Entrepreneurship degree, a reputation as an authority in AI applications in education. He made the spring cover of EdTech magazine and has given keynotes at education conferences.

“Every teacher needs to be a learning designer,” Roberts said, urging educators to reimagine assignments that take advantage of AI’s power. “Why can’t every single student have a bestseller-quality novel about their life experience, or leave high school with a profitable business?”

Other GSE alums are also shaping the AI space, looking beyond the hype and creating new and effective ways to improve teaching and learning.

“As large language models become ubiquitous, we must critically evaluate how to use them to enhance learning in our products and guide our partners in navigating their impact on the workforce,” said Austin Crouse, GED’25, an account manager at the skills intelligence platform SkyHive by Cornerstone in Birmingham, Alabama. GSE’s Learning Analytics program, especially courses on “Core Methods in Educational Data Mining” and “Feature Engineering,” gave him the foundational knowledge he needed, Crouse said. “It equipped me with the skills to evaluate new models for their true impact.”

Meanwhile, Candace Thille, GRD’13, an associate professor of education at Stanford University and former director of learning sciences at Amazon, has spent decades working at the intersection of machine learning and the science of human learning, in particular, building the infrastructure to most effectively support learners to develop their desired skills and knowledge.

“To truly personalize learning, one must consider features about the learner, features about the thing being learned, and features about the context. That’s way too many features for me, as a human, to manage,” she said. “That’s not too many features for algorithms to consider, and that algorithm can support my decision-making.”

Thille is currently using generative AI to build a learning-support agent that is also a research platform. The platform can simultaneously support learners with evidence-based learning-support strategies and help instructors test the effectiveness of those strategies.

“Generative AI,” she said, “has the potential to truly support the democratization of education, but only with human guidance grounded in the science of human learning.”

Alumni AI Trailblazers

“The way we teach, we’re preparing our kids for the industrial era,” said the founder of the Philadelphia-based Teacher in a Suit, which offers AI literacy training to educators, and the AI innovation lead at Foundation Academy Charter School in Trenton, New Jersey. “Our education system still teaches students to do what AI already does best—read, summarize, and reproduce information. The next generation of learners needs to do what humans must still do best: connect ideas, apply judgment, and turn information into meaning.”

Since launching his business in 2023, Roberts has built, with the help of his Penn GSE Education Entrepreneurship degree, a reputation as an authority in AI applications in education. He made the spring cover of EdTech magazine and has given keynotes at education conferences.

Other GSE alums are also shaping the AI space, looking beyond the hype and creating new and effective ways to improve teaching and learning.

“As large language models become ubiquitous, we must critically evaluate how to use them to enhance learning in our products and guide our partners in navigating their impact on the workforce,” said Austin Crouse, GED’25, an account manager at the skills intelligence platform SkyHive by Cornerstone in Birmingham, Alabama. GSE’s Learning Analytics program, especially courses on “Core Methods in Educational Data Mining” and “Feature Engineering,” gave him the foundational knowledge he needed, Crouse said. “It equipped me with the skills to evaluate new models for their true impact.”

“To truly personalize learning, one must consider features about the learner, features about the thing being learned, and features about the context. That’s way too many features for me, as a human, to manage,” she said. “That’s not too many features for algorithms to consider, and that algorithm can support my decision-making.”

Thille is currently using generative AI to build a learning-support agent that is also a research platform. The platform can simultaneously support learners with evidence-based learning-support strategies and help instructors test the effectiveness of those strategies.

“Generative AI,” she said, “has the potential to truly support the democratization of education, but only with human guidance grounded in the science of human learning.”