As the chief technology officer and assistant dean of the Stanford Graduate School of Education, Paul Kim spends more time than most pondering how artificial intelligence (AI) can impact education. He believes most educators don’t think about it enough, and those who do worry too much about it. “We’re at a very early stage of understanding what AI can possibly do for us, especially in the education teaching system,” he says. “I think the possibilities are huge.”
One thing he doesn’t see as a possibility? AI making human teachers obsolete. “Some people say it will replace teachers, but I disagree,” he says. “I think AI will augment what educators do.”
One edtech company that has embraced an AI-plus-human-teacher approach is Squirrel AI Learning, a Chinese company that has opened over 2000 after-school learning centers in more than 400 cities across China in the past three years. Squirrel—so named because, like the furry, nut-gathering rodent, it is fast, agile, plans ahead and “cracks tough nuts,” according to founder and CEO Derek Haoyang Li—uses an AI adaptive learning system and human instructors to give students a personalized education on subjects ranging from Chinese language to physics.
With a primary to high school student base of nearly two million, the company has generated huge data sets that allow it to find deep patterns on concepts—be they algebraic equations or noun-verb agreement—that are difficult for certain types of students at certain levels. “We have decomposed knowledge points to the super nano level,” says Li. “We have divided mathematics, Chinese, English and physics learning into a large number of abilities, and we hope to help students not only to improve grades but also cultivate their inquiry, critical thinking and other skills through our innovative Modes of Thinking, Capacity, and Methods (MCM) model.”
From an initial evaluation, the Squirrel AI system customizes an individualized learning plan that focuses on a student’s weaknesses and may include a variety of learning materials, such as instructional videos. While the student tackles subjects online, the AI generates ongoing analysis of the student’s state of knowledge. Here’s the twist: human coaches are on site to interpret the AI feedback and offer the student extra support. That individually paced path to improvement is something traditional education can’t deliver.
It’s something many other education companies using AI can’t deliver either. “Some companies using AI in education don’t have any human component at all,” says Kim. “Kids go through their exercises on the web with no human interaction.” Kim says that human tutors can look at the students’ performance and guide them at a much more personal level. “I think that’s great, especially if we think that one-on-one teacher to student relationship is the best relationship.”
In Li’s view, that personal learning path is essential; he believes that group-based instruction is a waste of time. “The greatest pain in traditional education is that every child is required to study at a fixed time and at a fixed rate,” he says. “But that’s like running a marathon and making everyone run at the same speed. Through technology, we can find each child's level of learning and speed—kind of like one-on-one running. Give each child enough time, they can create a miracle.”
Li’s approach essentially takes one of the oldest methods of education—one-on-one tutoring—and delivers it at scale. “Studies have shown that people who’ve had one-on-one tutoring score higher on a final exam than 95% of the students who haven’t had one-on-one tutoring,” says Tom Mitchell, the Interim Dean of Computer Science at Carnegie Mellon University and Squirrel AI’s chief AI officer. “So we know that individual tutoring by a person is extremely effective. Of course, it’s also extremely expensive.” He adds, “The potential impact of AI is to get back to that one-on-one tutoring and get even better improvement, because AI can learn from more experience than a human teacher will ever get.”
To illustrate the importance of the huge data sets Squirrel AI is able to generate, Mitchell points to AI advances in medical diagnostics. Thanks to machine learning, computers can now detect cancerous lesions on your skin as well or better than trained physicians. Why? Because the computers have examined and compared a pool of sample lesions far greater than what a human could experience in a lifetime. Likewise, with two million students, says Mitchell, “our algorithm will see vastly more student successes and failures than a human tutor can see in a 100-year career.”
In addition to doing deep diagnosis of learning difficulties, AI, with its endless well of patience, can potentially take over many of the things teachers find tedious and time-consuming—like administrative duties, grading and certain evaluation tasks—leaving the humans more time for individual coaching.
“Giving individualized attention is not easy for teachers,” says Kim. “That’s where AI comes into play. It collects and generates insights so that the teachers can look at the reports and understand where the students have been and where they are struggling and what pathway is better for them. So in the end, the ultimate decision maker is the human, not the computer. I think that is the key difference between Squirrel AI and other companies that may not have the human component.” He adds, “At the end of the day, it's the human who gives wisdom to our students, not AI.”
Li has seen the positive effects of his system close to home. His twin sons, second-graders who have been learning eighth-grade physics through Squirrel AI for about five months, recently surprised him as he was rolling his luggage out the door for a trip. “One son said, ‘Dad, do you know why your luggage suddenly became very heavy?’” recalls Li. “I said, ‘I don’t know.’ He said, ‘Because when you moved from marble to the carpet, the friction increased.’ He was already using physics knowledge in real life! That made me very happy.”
