Sunday Pick: Can AI really teach your kids? with Leah Belsky | from WorkLife with Molly Graham

Leah Belsky, who leads education at OpenAI, sees AI as a potential tutor and a tool for helping students create, but not as a substitute for the human and institutional parts of education. Her nearly 20 years in education technology include work at Coursera to make university courses and workforce skills available without admission or expensive tuition. Some of that access goal was met, she said, but course completion was low, many learners gained confidence without deeply learning, and employers continued to use degrees as a hiring filter. AI is spreading differently from earlier education technology, which was often adopted by schools and universities first: students are using it directly.

The promise of AI tutoring is to offer patient, personalized explanations, respond to questions, notice confusion, and present ideas in different ways. Belsky invoked Benjamin Bloom’s finding that students with one-on-one tutors progressed much faster than students in classrooms, but cautioned that building effective AI tutors is harder than she expected and will take time. Answering questions and tailoring information are not enough: learning also requires reflection, application, memory, and the effort of working through an idea. An AI-generated homework answer may feel like learning while taking away the practice students need to develop foundational skills.

Belsky has shifted her emphasis from teaching her son to use AI as a tutor to helping him use it to make things. Coding tools helped him start a website for a children’s 3D-printed fidget business, an experience she said demonstrated how AI can give students agency. She wants schools to make room for students to identify problems, build solutions, and prototype. Other examples she cited included students exploring whether existing drugs might help treat mental-health conditions such as bipolar disorder, creating an accessible version of Wordle for a blind friend that led to an educational-games company, and building a system for Penn students to send unused cafeteria food to food banks. These projects, she said, show students using AI to address problems they care about, not only to cheat.

The risks of cheating and effects on cognition are real, Belsky acknowledged, and require responses from both people and technology. OpenAI developed Study Mode with hundreds of teachers and learning scientists to encourage students to reflect and explain what they know rather than simply receive answers. But adults also need to help students tell the difference between learning and receiving personalized information. Belsky argues that classroom AI should be purpose-built and used intentionally, not left open all day. At her son’s school, technology is allowed in specified situations but not for writing assignments; she also encourages him to use chat without avoiding the struggle that can be part of learning.

Schools provide structure, accountability, motivation, and human relationships that AI does not yet offer, Belsky said. She expects AI to become a better tutor, while teachers retain an important role in motivating students and bringing them together. AI can also help teachers: Belsky said teachers using it can save about six to seven hours a week on administrative work, potentially freeing more time for students. She wants students to build and create repeatedly during high school, rather than encounter entrepreneurship, making, and coding only as extracurricular activities.

Belsky emphasized emotional intelligence, leadership, judgment, perception, and the ability to identify problems and build solutions as useful future-ready skills, while noting that no one knows what future jobs will look like. She pointed to Estonia’s education minister’s distinction between lower-order learning, such as recalling facts, and higher-order work, such as reflecting, creating, and analyzing. Neither she nor Molly Graham presented a settled answer to how children and adults should relate to AI. Graham stressed that access to information or an infinitely patient AI companion does not itself produce learning, which can depend on struggle; Belsky said schools and society are still working out what education should accomplish when AI makes intelligence widely accessible.