AI Economy
How should education adapt to AI? Six perspectives.
Six experts weigh in on how education can best adapt to the AI era.
Our second cohort of AI Economy Institute Senior Fellows just published a book of research-driven essays Learning Curves: Rethinking Education from Classrooms to Careers in an AI-Driven World. This book explores what shifts higher education institutions can and should make to help ensure that students and teachers alike are well-equipped to make the most of the opportunities afforded by AI and to thrive in the new AI economy. In this second edition of our Perspectives series, six fellows from our second cohort weigh in on the same series of questions, demonstrating where there’s consensus on the path forward and which areas are ripe for further research and exploration.
Sarah Rodriguez
Virginia Tech Foundation
Sarah Rodriguez is an Associate Professor of Engineering Education at Virginia Tech. Her research focuses on the engineering and computing identity development of historically marginalized populations in higher education. Dr. Rodriguez is currently involved in large-scale interdisciplinary projects on institutional environments and STEM identity, sponsored by the National Science Foundation (NSF) and the Kapor Center.
Wendy DuBow
University of Colorado Boulder
Dr. Wendy DuBow is Director of Strategies for Education Research & Evaluation at the Center for Technology Workforce Innovation (formerly NCWIT) and affiliate faculty member in Women and Gender Studies at the University of Colorado – Boulder. Dr. DuBow conducts mixed methods of social science research, with a focus on systemic levers that can broaden participation in technology for historically marginalized and excluded populations.
Robert Seamans
New York University
Robert Seamans is a Professor at NYU’s Stern School of Business, Director of the Stern Center for the Future of Management, and a nonresident Senior Fellow at the Brookings Institution. His research focuses on the economic impact of AI, robotics, and advanced technologies. His work has been published in top academic journals and cited by outlets like The Atlantic, The Economist, and The New York Times. In 2015, he served as the senior economist for technology and innovation on President Obama’s Council of Economic Advisers.
Bharat Chandar
Stanford University
Bharat Chandar is a postdoctoral researcher at the Stanford Digital Economy Lab, part of the Institute for Human-Centered AI. Dr. Chandar’s research focuses on AI’s impact on the labor market and productivity using a combination of big data and company partnerships. Bharat is a co-author on the recent “Canaries in the Coalmine” paper from Stanford.
Skye Zhao
University of Manchester, England
Xin Zhao (Skye) is a Lecturer in Generative AI for Education at the Manchester Institute of Education and a partner in UNESCO’s AI competency frameworks. Dr. Zhao also serves on the UN expert panel for Generative AI. Her research focuses on ethical, inclusive uses of AI in education.
Michael Impink
HEC Paris, France
Michael Impink is an Assistant Professor of Strategy at HEC Paris and a research affiliate at Hi! Paris (AI for Society and Business) and Boston University TPRI. His research focuses on how digitization impacts firm structure and performance. Prior to the PhD, Michael was a senior manager at Microsoft based in Seattle and Singapore and a fellow at Harvard University’s Weatherhead Center for International Affairs.
01
AI fluency
What does it mean to be fluent in AI?
Sarah Rodriguez
Virginia Tech Foundation
AI fluency is the ability to thoughtfully and effectively use AI in everyday work, learning, and decision-making while also understanding its strengths, limitations, and broader impacts. It is not simply knowing how to use a tool. It involves critical judgment, ethical awareness, and the ability to evaluate AI-generated outputs.
Sarah Rodriguez
Virginia Tech Foundation
Michael Impink
HEC Paris, France
AI fluency is not just the ability to use AI tools; it’s the ability to know when to use them and how their output may be biased. Users must know what these systems do reliably and where they fail. A fluent user can break a task into parts AI can help with, give clear instructions, and check the output against their own expertise.
Michael Impink
HEC Paris, France
Bharat Chandar
Stanford University
We define AI fluency at work by the AI capability gap—how far someone’s AI usage is from the most advanced peer AI users in their organization. For students, fluency should likewise be defined by knowledge of and ability to use frontier capabilities, even if those capabilities are selectively accessed to further learning objectives.
Bharat Chandar
Stanford University
Skye Zhao
University of Manchester, England
AI fluency is the capacity to understand and apply AI, or choose not to apply it, in ways that are responsible, transparent, inclusive, safe, and sustainable across different professional and everyday contexts.
Skye Zhao
University of Manchester, England
02
The Classroom Question
What should teachers keep in mind when incorporating AI into the classroom?
Bharat Chandar
Stanford University
Teachers should ensure students are equipped to deeply understand a domain via first principles thinking.
At the same time, they should ensure their students know how to get the most value from AI. In some cases, these objectives are in tension because AI can lead to cognitive offloading that undermines deep understanding. In practice, students should be exposed to a mix of environments with both constrained and full AI access.
Bharat Chandar
Stanford University
Skye Zhao
University of Manchester, England
It is important to reflect on the intended learning objectives and consider how AI can, or cannot, support the learning process and help students achieve the intended learning outcomes, rather than focusing primarily on how to detect AI-related cheating.
Skye Zhao
University of Manchester, England
Wendy DuBow
University of Colorado Boulder
Most students’ post-college roles will require them to be savvy users of AI. College students’ foundational knowledge should be balanced with opportunities to use AI in teacher-guided, learning-enhancing assignments. Second, I would encourage educators to develop assignments and classroom culture that reinforce the benefits and rewards of struggling before being given the solutions.
Wendy DuBow
University of Colorado Boulder
Robert Seamans
New York University
We are still in the early days of AI in education. The technology and best practices around its use are still evolving. Teachers should experiment with its use and seek feedback from students about what is working or not.
Robert Seamans
New York University
03
The Learning Process
How do we ensure that AI doesn’t replace the learning process?
Bharat Chandar
Stanford University
New systems can be designed to ensure students receive the benefits of personalized, on-demand AI instruction without the downsides of cognitive offloading. As these tools become more common, schools and universities should shift toward their comparative advantage: providing forums to develop social and leadership skills and interpersonal learning.
Bharat Chandar
Stanford University
Wendy DuBow
University of Colorado Boulder
As all thoughtful educators know, their homework assignments, exam formats, and grading rubrics all incentivize certain types of learning. It is no longer the case that assessing a student’s final product is a reflection of their learning. Assignments and assessments need to take into account the new capable tools students have at their fingertips and very intentionally encourage deep thinking and active learning
Wendy DuBow
University of Colorado Boulder
Sarah Rodriguez
Virginia Tech Foundation
The most important question is not what AI can do, but what learners still need to do for themselves. AI can automate routine tasks, provide feedback, and support idea generation, but deep learning requires reflection, persistence, collaboration, and human connection. Educators should intentionally design experiences that emphasize discussion, teamwork, problem-solving, and application.
Sarah Rodriguez
Virginia Tech Foundation
Michael Impink
HEC Paris, France
The key is to separate the product from the process.
Learning happens through effort, including all the wrong turns and mistakes along the way. AI used as a study partner or tutor could actively engage students in learning, while AI simply used to churn out assignments discourages deep engagement with the material. When students must explain and defend their thinking, outsourcing it to AI stops being a shortcut.
Michael Impink
HEC Paris, France
04
Closing The Divide
What’s the best way to close the AI diffusion gap?
Skye Zhao
University of Manchester, England
The most important tool at our disposal is capacity building, especially in higher education, where opportunities to develop AI competency can be embedded for both learners and educators.
Skye Zhao
University of Manchester, England
Wendy DuBow
University of Colorado Boulder
Currently in the US, the diffusion gap is largely a function of economics. Some people have access to technologies that others do not. One way to equalize access and fluency is to provide free genAI subscriptions to students in K-12 schools. But—and this is a big but—the introduction of AI must be accompanied by sufficient training for educators and students on how to use AI in ways that do not shortcut their learning or compromise their values.
Wendy DuBow
University of Colorado Boulder
Robert Seamans
New York University
Demonstration of use cases is important for diffusion. People need to see the variety of things that AI is capable of in order to imagine how it will be useful to them.
Robert Seamans
New York University
Sarah Rodriguez
Virginia Tech Foundation
Human potential and community colleges!
Our most powerful tool is investing in people. Closing the AI diffusion gap requires building capacity among faculty, staff, students, and community partners so they can confidently engage with emerging technologies. Community colleges are especially important because they sit at the intersection of education, workforce development, and regional economic needs.
Sarah Rodriguez
Virginia Tech Foundation
05
Navigating Career Changes
How can the institutions that support workers through career changes expand access to AI?
Bharat Chandar
Stanford University
Institutions closest to labor market transitions should provide tutorials for getting set up with AI. After that, they should provide examples for how people can use AI to learn how to use AI better. This is the most scalable and cost-effective way for quickly expanding access and reducing capabilities gaps.
Bharat Chandar
Stanford University
Michael Impink
HEC Paris, France
Community colleges, vocational schools, public employment services, and unions already exist to aid workers in times of change. This is one of those times. Learning how AI can complement their skills and using AI in real-life tasks can give workers knowledge of the technology and an understanding of what changes to expect.
Michael Impink
HEC Paris, France
Sarah Rodriguez
Virginia Tech Foundation
Expanding access requires meeting people where they are and recognizing that learners bring a wealth of different forms of knowledge. Flexible credentials, multiple entry points, industry partnerships, and stackable learning opportunities ensure that students and workers can engage with AI in ways that align with their aspirations, values, and workforce needs.
Sarah Rodriguez
Virginia Tech Foundation
Wendy DuBow
University of Colorado Boulder
This adoption must be accompanied by education and support for both educators and students to ensure students are well-served by AI tools. I can envision a trend where post-secondary institutions will market themselves as AI-free, making themselves very attractive to certain students.
Wendy DuBow
University of Colorado Boulder
06
The Human Skills
What are the most important skills for early-career workers to focus on developing?
Bharat Chandar
Stanford University
Early-career workers need to thread the needle between developing deep expertise while becoming AI native. A good rule of thumb is that deep understanding occurs when you can convince yourself you can explain all the relevant details. If something doesn’t make sense in your own head, you do not understand it.
Bharat Chandar
Stanford University
Sarah Rodriguez
Virginia Tech Foundation
While AI tools continue to evolve, durable skills will remain essential. Early-career professionals should focus on critical thinking, communication, collaboration, adaptability, and the ability to learn continuously. Workers should also recognize the strengths they bring from their communities, cultures, and lived experiences.
Sarah Rodriguez
Virginia Tech Foundation
Robert Seamans
New York University
Curiosity, experimentation, and entrepreneurial thinking are all important. Managers should encourage early-career workers to lean into AI and to experiment with ways to use it for their work.
Robert Seamans
New York University
Michael Impink
HEC Paris, France
Judgment, decision-making, and discernment are the soft skills that matter most. As AI takes over routine tasks, the value shifts to deciding what problem to solve, evaluating whether an output is correct, and knowing what to do with it. New workers should seek out deliberate ways to build these skills, including mentorship and challenging assignments.
Michael Impink
HEC Paris, France
07
The Global Conversation
How can we make sure workers around the world have a voice in AI?
Sarah Rodriguez
Virginia Tech Foundation
Centering worker voices globally begins with recognizing that workforce needs are shaped by local contexts, cultures, and communities. Effective AI and workforce strategies cannot be developed solely from the top down; employers, educators, policymakers, workers, and learners must be partners in the process.
Sarah Rodriguez
Virginia Tech Foundation
Skye Zhao
University of Manchester, England
We need to build platforms that allow global voices to be heard and ensure that perspectives from diverse linguistic and cultural backgrounds are represented, particularly at a time when AI guidance, datasets, and training materials tend to reflect dominant languages and cultural contexts.
Skye Zhao
University of Manchester, England
Wendy DuBow
University of Colorado Boulder
Relevance, representativeness, and technological sovereignty are the three keys to ensuring that worker voices are centered: Relevance ensures that AI models are tailored to the local context and that local talent becomes expert developers, not just users, of third-party technologies. Representativeness ensures that AI models reflect native language sources and cultural norms. Perhaps most importantly, technological sovereignty ensures that ownership remains regional and not yet another form of colonization.
Wendy DuBow
University of Colorado Boulder
Michael Impink
HEC Paris, France
We need to bring workers into decisions about how AI is designed and deployed, not just consult them afterward. Much of the labor behind creating AI happens in the Global South, yet these regions are slower to adopt AI for productive uses that could benefit their economies. Engaging workers in emerging economies requires building products that are in their own languages and work in their unique contexts.
Michael Impink
HEC Paris, France
AIEI editorial team byline
Ask six experts the same questions and you won’t get the same answers. That’s the point. But patterns emerge. The through-line isn’t consensus. It’s clarity about where the levers are. AI is spreading faster and the gap is growing wider, and that’s a choice, not a forecast. By offering insights from AI Economy Institute Senior Fellows, this series aims to shed light on how the diffusion of AI could broaden access to opportunity around the world.