This August, thousands of Chinese middle school students participated in a two-day “AI teacher vs. Human teacher” competition. Students were separated into two groups: one received math tutoring from experienced teachers, the other from an intelligent learning system. The result was favorable to the machines, as the students from the AI group delivered a better score in the final test (5.4 points vs 0.7 points).
The company behind the AI teacher is Squirrel AI Learning from Yixue Group. The Shanghai-based AI & E-learning startup aims to replace some human teacher roles with automatic and intelligent machines. Squirrel AI raised CN¥1 billion (US$150 million) in its most recent funding round.
Squirrel AI Learning is not an android that stands in front of a whiteboard delivering lectures, but rather an E-learning system that students can digitally access and interact with. Introduced this June, Squirrel AI uses customized resources and learning activities to identify and address the unique needs of each learner based on their profiles, learning level, strengths and weaknesses — an educational method is known as AI Adaptive Learning.
Born in the hopeful 1970s, adaptive learning had for decades failed to deliver on its promise — the failure generally attributed to limited computational resources and algorithms. A turnaround emerged between 2005 and 2010 as US startups such as Knewton and Dreambox started integrating AI algorithms into their teaching systems. After Arizona State University teamed up with Knewton their pass rates rose by 17 percent and course withdrawals dropped by 56 percent. Knewton and Dreambox have since grown into billion dollar companies.
“Adaptive learning is concerned about helping a student learn according to their diverse needs and learning behavior and not what traditional learning preaches. It’s more like providing a personalized trainer to every individual at a mass scale, so that students can learn at their own pace,” says Ketan Kapoor, CEO and co-founder of skill-assessment tech company Mettl.
The success of AI-powered adaptive learning in the US was an eye-opener for Derek Haoyang Li. In 2014, Li founded YiXue Group and the Squirrel AI Learning system, whose CN¥270 million seed round set a record in the K-12 education industry. Last year the company released its first E-learning class, comprising an AI system for testing students and providing video course recommendations, and a human teacher for additional instructions and Q&A. By February 2018 the company had more than 100,000 paid subscribers with a 90 percent student retention rate, and had opened 1000 AI-powered learning centers across China.
Devoted to China’s K-12 education for 15 years, Li has long been critical of China’s teacher quality gap. At the recent TechCrunch Disrupt conference in San Francisco Li told Synced “In China, there is only one senior teacher out of every thousand teachers. So how can you hope to enable all students to receive their educations from top teachers?”
A report from big four accounting firm Deloitte suggests China’s education industry is entering a golden age. The market size of Chinese kindergarten through 12th grade education has doubled over the past five years and the pace is expected to continue for the next five years.
Chinese media has already nicknamed Squirrel AI “China’s Knewton,” but Li expects even more.
Squirrel AI Learning’s knowledge graph and user portrait are standout features. The system ingests and parses all middle school textbooks into a cross-disciplinary knowledge graph with 30,000 academic concepts. It uses test score data such as type of wrong answers, and even time spent on each question to build profiles for each each student based on their strengths, weaknesses, interests, and knowledge level.
The system crafts course recommendations from its database of 400,000 video courses and 10 million questions. The Yixue research team told Synced they use classification trees and fuzzy logic for optimal learning content recommendation; genetic algorithm and evolution theory for recommending the best learning path; and reinforcement learning and deep learning for determining the right teaching mode.
One challenge facing Squirrel AI Learning, however, is convincing parents the system actually works. Not every parent buys the magic of AI, and credibility is not built in one day. Besides, there are concerns about whether the massive data collected from an adaptive learning system might be misinterpreted or even misused. Angela Estrella, a professional development associate and instructional coach at Stanford University’s Center to Support Excellence in Teaching, asked in an EdSurge interview: “Does having access to all this data empower parents, or does it play into our fears of all the great expectations we have for our kids? Not every click needs to count, and sometimes you need a clean slate.”
Most education experts doubt whether adaptive learning can truly replace human teachers. Douglas Fisher, a professor in the Department of Educational Leadership at San Diego State University’s Adaptive Learning Programs, cautions that “adaptive learning is not intended for new instruction, but rather to provide distributed practice to strengthen students’ fluency and stamina.”
Li however firmly believes that AI can more efficiently teach students, and that the role of teachers in the classroom will be significantly changed in future.
“AI will generally replace teachers in the work of imparting knowledge. Meanwhile, teachers will spend more time helping students to deal with complexity, diversity, and changes to prepare them for the future.”
Journalist: Tony Peng | Editor: Michael Sarazen