AI In Education: Forming Citizens When AI Can Do The Work - Forbes

OpenAI’s new ChatGPT Work is designed to do more than merely answer a question or draft a paragraph. Give it a goal, and it can gather information from files, apps, and the web; it can break a project into steps and produce a coordinated set of outputs. In envisioning AI in education, OpenAI imagines faculty and staff using it to revise courses, assemble accreditation materials, and manage projects.

While the appeal is efficiency, the larger implications are harder to ignore. As machines become better at carrying out assigned work, schools must ask what human beings should be educated to do.

That question was at the center of my recent conversation with Leah Belsky, OpenAI's vice president of education. Belsky notes roughly 20% of conversations on ChatGPT involve education, learning, or information, and about 40% of its users are under 24. Young people are already using the technology at scale, often without guidance about when it supports learning and when it replaces it. This lack of clarity around the use of AI as a tool and the ultimate goal of education has the potential to unwind what should be a great leap forward in learning.

AI In Education: Extending Thought Rather Than Avoiding It

Belsky draws a useful distinction between using AI to avoid thought and using it to extend thought. The first removes the effort that learning requires. The second can challenge assumptions, expose gaps in understanding, and open new lines of inquiry.

She recalls an analogy from the online course “Learning How to Learn”: The brain is like a rubber band. It has to stretch to change. A student who struggles to retrieve an idea, connect two concepts, or work through a difficult problem may be doing the work that makes learning durable.

Generative AI can make that struggle disappear. A student asks a homework question and receives a fluent explanation. The answer makes sense, and the student feels a sense of progress, but fluency is not mastery.

“Even though you're not experiencing that friction,” Belsky says, “it feels like learning.”

That may be one of the technology's most subtle risks. Students can reach the end of an assignment, having answered all the questions or produced all the required work, without knowing whether their minds traveled the necessary distance for learning to have actually occurred.

OpenAI's Study Mode, introduced last year, is meant to counter that tendency. Instead of immediately supplying an answer, it asks follow-up questions: What does the student understand? Where did the reasoning break down? What approach might work next?

Though its design follows recommendations from teachers and learning scientists, no product setting can solve the problem on its own. A student can always leave the guided experience and ask for the answer directly, perhaps in a second chat session running outside the programmatic observer. Schools will have to teach students something more basic: how to recognize productive difficulty. More specifically, how to know when convenience becomes evasion and why learning cannot be measured by the speed with which an assignment disappears. This is not a new challenge. Anyone who has ever given students a reading assignment with questions to be answered will know that many students just read the questions and then skim for the answers without doing the actual reading. The point that students must understand is that the learning does not come from having produced a certain work product, but from having engaged in a specific work process leading to that product.

AI In Education As A Question Partner

Belsky identifies a canonical best use of AI as a question partner for students. One example comes from professors at Harvard who have “flipped the case.” Instead of preparing for a business-school discussion by reading a case and waiting for class, students can interrogate a custom GPT loaded with the relevant materials. They arrive having tested claims and explored alternatives.

At Duke's Fuqua School of Business, a professor has used AI to analyze student group conversations. The system can surface participation patterns and gaps in understanding, making the quality of a conversation visible. For students frustrated by the experience of group work and eager for recognition of their contributions, this is a game-changer.

Belsky compares these experiments to her experience at Yale Law School, where sustained questioning by a professor was a defining part of legal education. That kind of attention has always been scarce. AI may make some version of it available more often and to more students, allowing everyone to experience the hot seat with every class, even if only via the AI.

The insight here is that the valuable resource is not information, which is already abundant. It is intellectual engagement: being pressed to clarify a claim, defend an inference, confront a contradiction, or reconsider an assumption.

AI In Education: Promoting Agency Before Skills

Better questioning only takes us part of the way toward better education. The deeper issue is what education is trying to produce.

Belsky says she is often asked which skills schools should teach for the age of AI. She has come to think the question begins too far downstream.

“Maybe the right question is not what skills should we teach,” she says, “but what kind of human do we want our education system to produce?”

Her answer is a person who can identify a problem worth solving and who has the confidence, motivation, and agency to act on it.

An example she gives is a “problem to prototype” course that her son attended. Students began with a problem, then determined what they needed to learn, whom they needed to work with, and which tools they needed to build a response.

That sequence reverses the usual logic of school. Traditional education often asks students to acquire knowledge first and promises that they may someday find a use for it. An agency-centered education begins with purpose and lets that purpose create a demand for knowledge. While such an approach was a recipe for frustration in a world where knowledge acquisition was hard, in the AI era, it drives motivation.

Other examples Belsky cites include students using AI to redesign online games for blind players or improve the distribution of unused cafeteria food to food banks. Schools have often confined meaningful creation to entrepreneurship clubs, hackathons, and extracurricular programs. AI allows building and problem-solving to become a more ordinary part of education.

AI In Education: Citizenship Gives Agency Its Direction

In discussing the ultimate value and purpose of education, our conversation turned to citizenship. Public education has never been justified solely as job training. In a republic, schools are also expected to form citizens. I suggested to Belsky that forming citizens is, in an important sense, forming agents: people able to examine claims, choose among competing ends, work with others, and act in the world.

An engaged citizen needs to do more than passively receive information. A citizen needs to ask where it comes from, what evidence supports it, and what the consequences are of acting on it. In a world of generated answers, provenance becomes part of agency. The response “ChatGPT said so” cannot be the end of an argument.

Belsky embraces the connection, describing AI-enabled creation as part of “22nd-century entrepreneurship” and “22nd-century citizenship.”

The phrase joins two purposes of education that are often treated as rivals. The economic case asks schools to prepare students for a changing labor market. The civic case asks them to prepare students for self-government. Agency links the two.

A person who can identify an important problem, gather evidence, persuade collaborators, and accept responsibility for a decision is equipped to create a company, contribute inside an organization, or participate meaningfully in a democracy. In each case, the person must be more than an executor of instructions.

Students can think of themselves as tools, or as people who use tools. If their value lies mainly in carrying out a task defined by someone else, then a faster and cheaper tool poses an obvious threat. The remedy is not to withhold the better tool but to prepare students to operate at a different level.

What problem are we solving? Why does it matter? Which evidence should we trust? Who bears the cost? What should happen next?

Those are questions of judgment. They position the student as an author of purposes rather than an instrument of someone else's purpose.

Agency, however, is not automatically virtuous. A capable person can pursue a foolish or destructive goal. A student can build something impressive without understanding the people it affects. Education must develop not only the power to act, but the judgment to decide what is worth doing.

That is why citizenship is more than a rhetorical flourish. It gives agency a public and ethical direction. Students should be asked not only whether they can build something, but also why it should be built, whose needs it would serve, and who would be accountable if it failed.

ChatGPT Work makes these questions immediate. The machine can increasingly help gather evidence, organize a project, and produce the deliverables. It can supply more of the means.

Education’s task is to form people capable of choosing the ends, and citizens willing to answer for those choices. To be a net positive, AI in education must deliver on this promise.