Brave New Words: How AI Will Revolutionize Education (and Why That's a Good Thing)
The words “with me” tickled Salman Khan’s imagination when he was naming his new invention, an AI tutor for academic studies. He chose the name Khanmigo because it sounds like “conmigo,” Spanish for “with me.”
If Khan is correct, starting at an early age, students will have a personal AI tutor with them at all times. The tutor will possess endless patience and vast knowledge and cost little to nothing. It will stay with students through college and aid their every academic endeavor.
Until now, such top-notch tutoring was available only to the wealthy.
Khan founded the non-profit Khan Academy, and in Brave New Words he offers a compelling, if not idealistic, vision of how AI tutoring will revolutionize education. It’s a valuable primer for educators, administrators, and policy makers.
Learning gaps are a major obstacle to subject mastery, and because classrooms move at a fixed pace, teachers lack time to help all students with gaps before advancing to next levels. The gaps accumulate. Kahn believes Khanmigo can help eradicate this critical problem for education on a global scale.
Khan’s optimism may be colored by the excitement of having had early access to ChatGPT, which in the summer of 2022, OpenAI’s Sam Altman and Greg Brockman invited him to inspect. It was four months before ChatGPT was released, and the sneak preview was a bid to collaborate: OpenAI was seeking trusted partners “to showcase socially positive and real-world examples.”
Khan instantly recognized Chat GPT as “a technology that seemed to tie together every thread of my journey.” His mission is to give all students access to education regardless of income (Khan Academy offers free courses on math, science, history, humanities, computing, and more), and he could envision ChatGPT providing something missing from his platform and classrooms: on-demand, personalized tutoring.
He describes his awe when his daughter collaborated with ChatGPT, building characters and dialogue for a fictional story. For him it conjured a potential world where AI tutors help humans exercise their “curiosity muscles.”
Khan’s development team quickly spun up Khanmigo, which today is available to anyone for $4.25 a month. Thirty thousand teachers and students in the U.S. are piloting it, and early results show students are gaining confidence in learning.
Khanmigo is designed to "nudge students forward with leading questions without giving away the answer," writes Khan. Students can ask Khanmigo things they’d be afraid to ask in class or the same question again—no shame. That level of individual support can help teachers ensure their students stay motivated.
Will human teachers still be needed? Khan’s assurances about their continued relevance in the face of AI tutoring, while comforting, lack substantive evidence in the face of rapidly advancing AI capabilities. Those capabilities form the scaffolding of Khan’s vision, which sprawls way beyond conventional tutoring.
Khan imagines a hyper-personalized "super tutor," an entirely new entity that will likely thoroughly disrupt education. He touches upon many potential effects, on everything from testing to college admissions to our definitions of cheating. While the prospects for improving education merit excitement, the effects of this powerful technology on teacher employment are impossible to predict.
Khan touches upon potential harms of an ever-present AI tutor, but again, maybe too breezily. With hyper-personalization via digital technology, privacy is a concern. Khanmigo's long-term memory, critical to addressing knowledge gaps and tailoring responses, can monitor students' work habits and provide detailed reports to teachers and parents.
Khan’s discussions of these capabilities remain upbeat, while for the reader they might evoke images of the helicopter parent meets Big Brother meets the CCP’s social credit system.
Analyzing decades of data about their charges, AI tutors might someday write college recommendations, says Khan. This could make recommendations more accurate. But one is left wondering if students’ mistakes, like a felony, might follow them around for life.
Khan treats privacy and security perhaps too lightly, saying “companies like OpenAI seem to be putting good guardrails in place to avoid giving away any sensitive information about an individual” and suggesting that “nefarious users” are the main concern. (Some see the technology and its makers as equally worrisome.) With AI’s exponential growth, books on the topic may become outdated quickly, and that seems to be the case in giving the benefit of the doubt to OpenAI, which in May disbanded its AI risk “super alignment” team and lost key executives Jan Leike and Ilya Sutskever in part over safety concerns.
Still, Khan's ideas are inspiring, challenging us to take advantage of personalized, AI-assisted learning and make it universally accessible. He is an informed optimist on AI’s feasibility, which is clearly here to stay: “Students who learn to harness AI ethically and productively will not just accelerate their learning, but also maintain a competitive edge throughout their careers.” For entrepreneurial types, he says, AI will bring back the experience of pre-industrial craftsmanship, where with AI tools, a small group of people can run companies with “armies of generative AI.”
While Khan does not directly discuss the potential of a rogue AI tutor, he does warn against an Orwellian future which he feels we can avoid by becoming AI-literate, erecting proper guardrails and embracing an abundance mindset for education, which seems like sound advice.
After all, he says, the AI tutor is here today: "This is not a drill."