Fan Xing, Co-founder and CTO of Yixue Group's Squirrel AI attends the O'Reilly Artificial Intelligence Conference

BEIJING, June 26, 2019 /PRNewswire/ -- From June 18 to 21, O'Reilly Artificial Intelligence Conference was held in Beijing with the theme of "Breaking the Barriers between Theory and Reality". The conference gathered professionals from Google, Facebook, eBay, Squirrel AI, Uber, Microsoft, Alibaba, Amazon, SAS, Unity, SalesForce, IBM, Tencent, MIT, Berkeley, Stanford and Oxford Universities and conducted an in-depth exploration in the trends of AI and the reforms it has brought to the field.

Fan Xing, a co-founder and CTO of Squirrel AI Learning by Yixue Group delivered a speech with the title of AI Technology Empowering Traditional Education. He said the current reality of education was that the distribution of quality education resources is uneven and we are still short of excellent teachers. Meanwhile, he said with optimism that AI is actively changing the education of China, and taking Squirrel AI Learning as an example, shared the technical breakthroughs and applications in its development.

Fan believes that at present AI has been able to perform 70% of the conventional teaching tasks which were originally performed by teachers. Yet in terms of educating students, it is difficult for AI to replace the position of teachers. In fact, the separation between teaching (jiao) and educating (yu) has existed since the ancient time. But nowadays we pay more attention to "teaching", but neglect "educating", as what all students are acquiring is knowledge. If AI is applied for the 70% of teaching tasks, then teachers can spend more time and energy on the task of educating and truly enable the children to achieve an all-round development.

Yixue Group Squirrel AI aims to address and solve two pain points of education sector, i.e., the uneven allocation of education opportunities and the lack of individualized education through AI adaptive algorithms. Previously the former pain point could only be solved through MOOC, and the latter could only be addressed through one-to-one tutoring, but both failed to meet expectations to solve problems systematically with technologies. Then what can be done by AI to solve these problems?

Fan Xing explained that the artificial intelligence of Squirrel AI is in fact like a system and similar to mobile navigation. The marginal cost for replication is very low and its use is subject to no limitation. With an access to internet and a computer, kids in Beijing or Tibet can enjoy the same AI services.

Every kid has different demands for teaching. It's difficult for the class today to satisfy all their demands. That's why after-class tutoring is so popular, since after-class tutors can offer one-to-one teaching according to the conditions of every kid. What has been denounced for a long time is that though we always wish to change the reality of "studying mechanically", our changes have been quite limited in the past decades. Why? Nowadays our education focuses on the children's mastery of knowledge, but to a certain extent has neglected the training of their abilities, thoughts and methods, which has given rise to problems such as high scores and low abilities.

Why can AI perform 70% of education functions? In fact it depends on the gene of AI, which is data. Fan asked, for example, when a kid answers a question, what data can we collect? We can know when and where the kid answered the question, what the question was, what the result was, and the reason for his mistake, and we can even collect his facial expression when he answered the question, how his concentration was and whether he was nervous. With AI, we can collect so much data. Supported with the data, the application of algorithms has became self-evident.

Using AI, we can win back the time wasted by the kids in the classroom and train their learning according to their ideas and methods. Then their condition will be improved. Then what functions can AI provide? AI can provide every kid with a tutor who can individualize the teaching plan and goals according to the conditions of every kid so that the content kids are learning is indeed necessary, which has overturned the conventional education mode.

Unprecedented revolution brought by AI technology to education

In recent years, AI empowerment keeps activating the education market of China. The education sector has always been a field attracting fast flows of cash and capital, in which the twelve-year education of primary and secondary schools is particularly attractive to investment and has refreshed records in financing amount. For example, AI unicorn companies represented by Squirrel AI of Yixue Group had raised funds of nearly 1 billion yuan in accumulation by July 2018.

Fan Xing explained that in professional fields Squirrel AI is known as adaptive education, namely artificial intelligence adaptation. Artificial intelligence education has developed very rapidly in the world today, but its development is slower in China. The AI learning has had a long history of development abroad. Hundreds of companies are engaged in AI education and hundreds of millions of users are using such kind of systems.

In the US, AI education is a popular track. Zuckerberg has made a big bet on it, as he has taken AI education as the focus of his business in the future. So did Bill Gates. The Bill Gates Cameron Foundation has allocated a quota for education. The AI education project, BYJU'S of India has raised fund from the foundation. They wish to solve the unfairness in education thoroughly with AI education.

Viewed internationally, this area has developed quickly. Yixue Group's Squirrel AI is also advancing in this area at a fast speed. In August 2018, depending on its technical advantages, a Squirrel AI teaching robot defeated the teachers from 100 cities of China and made a good performance in the man-machine competitions for several times. How did it accomplish this? Fan introduced the learning engine structure of Squirrel AI, which can be divided into three levels:

At the topmost is learning goals, which mainly comprises several aspects. First are the causes for the frequent mistakes of kids. Second, the learning map of the kids generally determines what to learn first and what to learn later. Third is the content map and the entire knowledge system. At the middle are learning records and individualized learning recommendation for every kid. The learning detection not only relies on static things, but even more relies on historical learning data and may include learning data of other kids. At the bottom is learning foundation, including AI engine and AI architecture, such as evaluation engine, recommendation engine.

Squirrel AI has also taken the initiative globally to separate learning thoughts, abilities and methods. In the past all these things could not be improved and the way to improve them was to "read more books" and gain these abilities by a lot of sub-consciousness training, but there were no targets, values and directions. Like swimming, a kid can also learn to swim after swimming in the river every day. Yet if the kid is trained by a professional coach, the effect will be different. The core difference between the two lies in efficiency. Abilities, thoughts and methods can be trained and improve efficiency. The above-mentioned problems of "studying mechanically" and "high scores and low abilities" can also be changed through training.

Fan also introduced two major engines of Squirrel AI Learning system, i.e., state evaluation engine and recommendation engine. The former engine is used to detect, judge and evaluate, and the latter engine is to provide an optimal path based on the outcome of current detection and judgement and the conditions of the object of detection. Working on following this path, it can accurately detect the knowledge loopholes of every kid.

In fact more than learning, we also need to achieve dynamic goals. Everyone needs to reach different levels of the same goal. Some kids encounter problems in concepts, some kids in understanding, or some kids in application. In the learning process, kids will also change. Therefore, the kids need dynamic goals that can be adjusted in a real-time manner. This is difficult to be achieved depending on teachers. Only machines can fulfill it more easily.

The information only obtained from the behaviors of kids is incomplete. But we can gain more information through dialogues to help the system to make judgement. The dialogue identification engine of Squirrel AI Learning can collect more information as the system is working, including the result of every question answered by the kid, and even every dragging of the mouse. All emotions of the kid can be judged in this way.

What is the role played by our teachers? Teachers will keep monitoring the conditions of the kids. Through the panel, the system will send some warnings to the teachers which can help their decision-making.

Finally, Fan said that AI will bring an unprecedented revolution to education. AI+ education will break the limitation of location and time and integrate the experience and wisdom of hundreds of special-class teachers to provide truly individualized education to every kid. It will customize learning plans based on the learning speed of every kid and cultivate their learning thoughts, abilities and methods. For example, the homework for students can be cut by 80%. The phenomenon of a life-determining exam will vanish gradually. Education will become more transparent and fair as technology is improved. More kids will embrace a brighter future.

SOURCE Squirrel AI Learning