Global development of AI-based education w
Contents
1. Introduction to AI-based education 1 1.1 Development history 1 1.2 Application at home and abroad 2 1.3 Difference from traditional education and value added 5 1.4 Critical technologies 7 2. Analysis of AI-based education systems 10 2.1 Analysis of intelligent adaptive education structure and model 10 2.2 Three main application scenarios for intelligent adaptive education 16 3. AI is transforming education industry 29 3.1 AI drives the role change of ecosystem participants 29 3.2 Intelligence becomes the prevailing trend of education industry 31 4. Trends of investment in AI-based education 36 4.1 China is becoming one of the most active regions for investment around the globe 36 4.2 Investment in the AI-based education segment remains fragmented 40 4.3 AI-based education investment and integration trends for the next step 42 5. Future challenges, prospect and thinking of AI-based education 47 Deloitte China Contacts / Whitepaper Editorial Board 49
Development of AI-based education in China | 1. Introduction to AI-based education
1. Introduction to AI-based education
1.1 Development history of approaches from mathematics, since 2010. Established educational Artificial intelligence (AI) is accelerating engineering and economics.In organizations such as Tomorrow the integration of information 1970's, Jaime Carbonell established Advancing Life and New Oriental, as technology into education, providing intelligent tutoring systems, under well as capital market in China stepped support or even alternative approaches which computers were used to assist into AI-based education around for practioners in the industry. AI-based education. In 1993, the first World 2016. After decades of development, education in the future will be driven Conference on Artificial Intelligence AI has evolved from a concept of by the use of data, intensive technology in Education was held in Edinburgh, reasoning to a technology combining application, integrated innovation, and UK. AI-based education has been natural language processing, speech service optimization. developing over time. In the early 21st recognition and image recognition century, with the establishment of based on deep learning, along with AI-based education is still developing various intelligent adaptive education significant improvement in algorithm and not yet fully mature. While AI is a companies such as Cognitive Tutor, model. Big data provides AI with data hot topic, application of this technology Knewton and Realizeit in the US, AI support and cognitive computing in education industry requires long- technology was increasingly applied in capabilities. Besides, application of term efforts. The initial stage of AI education industry. Intelligent adaptive AI technology in education industry application in education focuses on learning is an AI-enabled education requires cooperation across different planning and tentative exploration. technology that simulates human disciplines including neurology, Since Allen Newell and Herbert Simon teachers’ one-on-one teaching process, cognitive science, psychology, from Carnegie Mellon University deeply personalized for every type mathematics and education. Multi- invented Artificial Intelligence (AI) in of student. An increasing number of disciplinary application will facilitate the the 1950s, it has evolved to address intelligent adaptive learning providers development of AI-based education. problems of probabilistic and numeric including Yixue Education and Gold- nature, leading to the incorporation Wood Learning have emerged in China
1 Development of AI-based education in China | 1. Introduction to AI-based education
Figure 1-1: Different development stages of AI-based education
•• In the early 21st century , with the establishment of various intelligent adaptive •• Intelligent adaptive education companies education market such as Knewton and landscape will be generally •• It is the late development Realizeit, AI technology stable at the mature stage, phase of AI-based was increasingly applied in with multiple players education, when players education industry； taking a significant share in the field will have to •• Intelligent adaptive respectively. The industry •• In 1988, University of transform to new business learning providers chain will be clearly Montreal organized models against slower and emerged in China divided, and AI-based the first International even zero profit growth. around 2010. Established education companies will Conference on Intelligent educational organizations see rapid revenue growth. Tutoring Systems; and capital market in China •• In 1993, the first World stepped into AI-based Improving stage Conference on Artificial education around 2016. Intelligence in Education Mature stage was held in Edinburgh, UK.
Developing stage
The current Initial stage development stage of AI-based education
Source: Deloitte Research
1.2 Application at home and abroad AI-based education is embracing substantial development both at home and aboard, with AI technology applied in education through different patterns. Under the model of "AI + education", companies prioritize the research and development of educational intelligent products and services to fully satisfy users' need, leveraging the advantages of AI technology. Customers of AI-based education want more convenient and efficient intelligent education.
2 Development of AI-based education in China | 1. Introduction to AI-based education
Figure 1-2: Types of AI Application in education
Types of AI Application in Specific scenarios of AI Application Domestic companies Overseas companies education in education
Intelligent adaptive learning Combined with intelligent adaptive •• Yixue Education(Squirrel AI) •• IBM Watson; learning technology to create a virtual •• 17zuoye; •• Knewton; teacher, not only can permeate the whole teaching process, but also •• Liulishuo; •• DreamBox Learning; support personalized teaching. Each •• Knowbox •• Renaissance Learning; student can learn according to their own rhythm, which helps to improve •• 51Talk; •• Cognitive Learning learning efficiency and enthusiasm. •• Toutiao •• Duolingo;
Human-machine Intelligent source processing and •• Est100 •• LightSail Edution; interaction search technology •• iFlytek; •• Grammarly; Dual teacher classroom Image recognition •• Box Fish •• Cerego
Speech assessment Intelligent language processing and •• danatech; •• Wonder speech recognition •• 91jzx; •• Workshop; Intelligent language Based on language processing , •• Dji; •• RoboKind ‘Millo’; processing application being able to build some grammatical frameworks •• Singsound; •• Sphero
Photo based question Computer vision and image •• lzrobot searching recognition
Source: Deloitte Research
3 Development of AI-based education in China | 1. Introduction to AI-based education
Development of intelligent adaptive Figure 1-3: Percentage of Top 30 "AI + education" companies providing education in the US started as early the following services as in 1990's, while China has not seen rapid development of the model until 1 Quality-oriented the past decade. Foreign language education learning is well received in China, with K12 K12 and English tutoring attracting 1 most customers. In contrast, language Higher education learning is far less popular in the US than in China. Compared with the Corporate training small share of language tutoring in AI-based education in the US, higher Language learning education is the focus, equal to K12 tutoring in the percentage of related Reading service providers. Others
Source: iyiou.com, Deloitte Research
Intelligent adaptive learning of such situation, China is likely to space theory. Leading AI-based technology has been used in the catch up from behind, based on large education companies in China make US and Europe for more than ten population and rapid development. As use of a wide range of technologies years, with over 100 million users for technologies, foreign companies too. For example, technologies applied across different ages. Both product such as Knewton use probabilistic by squirrel AI of Yixue Education and technology have evolved to an graphical model, item response theory broadly cover genetic algorithm, neural advanced level in these regions. and learning analytics, etc.; RealizeIt network technology, machine learning, China, however, is still at the initial adopts information theory, bayesian graph theory, bayesian network, stage in the application of intelligent algorithm and machine learning, etc.; logistic regression model, knowledge adaptive learning technology, lagging and ALEKS mainly leverages knowledge space theory, information theory, behind in data accumulated. Despite bayesian theory, knowledge tracing theory, educational data mining and learning analysis technology.
4 Development of AI-based education in China | 1. Introduction to AI-based education
Figure 1-4: Application of AI-based education in domestic and overseas markets
Domestic market Overseas market
Development Paid after-school tutoring is developing well in China, With various background, students may flow to background with positive prospect. diverse service providers.
Technology At the initial stage of applying intelligent adaptive Intelligent adaptive learning technology has been learning technology, China is lagging behind in the used in the US and Europe for more than 10 years, application accumulation of related data with over 100 million users across different ages
Content Intelligent adaptive education companies in the Mostly prepared by schools and publishing domestics market are inclined to developing teaching houses, teaching materials are featured with clear development materials by themselves. As many different exam- intellectual property rights and complete and oriented textbooks are used in China, these companies systematics contents have to analyze potential examination questions aligned to different textbooks, providing detailed contents to prepare students for tests.
Application K12 and English tutoring K12+, higher education and vocational education scenarios
Representative Yixue Education(Squirrel AI), New Oriental, Tomorrow ALEKS, RealizeIt, Knewton, Dreambox Learning, Advancing Life, Liulishuo Duolingo, Renaissance Learning; Cognitive companies Learning
Source: Deloitte Research
1.3 Difference from traditional goal of education from high scores teaching human resources. Besides, education and value added to quality improvement. Adaptive AI is able to facilitate decision making AI application in education has a learning provides personalized based on educational big data, disruptive influence on traditional solutions in accordance with students' enabling precision teaching and education, as AI has changed the goal specific situation and needs, changing improvement of learning speed and and method of education. Teaching the way of education. AI has disrupted flexibility. The technology can also students in accordance with their the whole education process, playing drive information flow within the aptitude is possible with the support a major role in teaching and liberating school and resource sharing across of AI technology, which transfers the different regions, breaking barriers to information and resource sharing.
5 Development of AI-based education in China | 1. Introduction to AI-based education
Figure 1-5: AI transforms traditional education and creates new value
Traditional education AI-based education Value
Education quality Same teaching materials • AI brings about higher and models are used • examination scores and pass rate, in the education of •• Personalized solutions are developed as well as lower drop-out rate; students with different through a student-centered approach, Students • AI drives the optimization of characteristics. Cramming in accordance with students’ specific • education to the development method prevails, satiation and needs. with high scores in of students’ critical thinking and examinations as the innovation capabilities. focus.
Education efficiency •• AI has changed the whole education Homework checking process, taking the place of human •• Teachers are liberated to focus on consumes a lot of time; in teaching and disrupting previous educational research and one-to- repeated work takes models to educate students in one communication with students; Teachers accordance of their aptitude ; the time for creative •• Students can spend less time in education research and •• Big data and machine learning are completing courses and focus on interaction between used for the analysis and evaluation solving specific learning difficulties; teachers and students. of students’ daily performance and •• Schools realize more efficient, examination results. real-time and precision management and risk response, based on big data. Traditional level-based model of school •• AI technology is fully applied in school management leads to management, enabling information Education equity information silos among exchange at the physical and data different departments, level. AI-based data center provides •• Good resources flow to areas with which in turn results support for daily management and risk insufficient educational resources, in low management response; through remote connection School efficiency and limited between teachers and students or educational resources •• AI technology enables remote high-quality teachers simulation obtained within the connection between high-quality enable by AI. classroom. High quality teachers and students, breaking time educational resources and space barriers and facilitating spread unevenly, educational resource sharing among concentrated in eastern different branches of the school. coastal regions.
Source: Deloitte Research
6 Development of AI-based education in China | 1. Introduction to AI-based education
AI has changed the way to apply same AI breaks the barriers to information contents and models in teaching and resource sharing. The different students, through analyzing emergence of smart campus concept, learning models and individual interconnection among school differences among students. Statistics facilities, department integration and suggest that students under AI-based information exchange have provided educational model are able to achieve a data base for management decision better learning results, with shorter making and risk response. Remote time in completing courses, and teaching and online education have higher pass rate and examination enhanced education equity, enabling scores, compared with those under high-quality educational resources to the traditional educational model. flow from eastern coastal regions to Meanwhile, quality-oriented education remote and poor areas. is likely to be widely adopted with the support of AI technology, which Compared with traditional education, enhances all-round development of AI-based education has enhanced students by improving their critical education quality, efficiency and equity, thinking, innovation and other creating value for students, teachers abilities, while elevating exam-oriented and regional education systems. education efficiency. 1.4 Critical technologies Human teachers are liberated by AI Many AI related technologies have assistants from repetitive and tedious a profound impact on education, of exam paper correcting and daily which five technologies derived from management to focus on creative work the integration of basic theories of of educational research and one-to- computer science including algorithm, one communication with students. graph theory and inferential statistics AI-based decision making with the with edge-cutting theories in other support of big data from educational fields are critical for AI-based education. activities helps teachers to better understand the educational satiation.
7 Development of AI-based education in China | 1. Introduction to AI-based education
Figure 1-6: Illustration of critical technologies for AI-based education
Intelli ent adaptive lea nin platfo m
ducation data minin and Classification p edication anal sis algorithm Expert knowledge system representation Knowledge Bayesian Search graph Logic network Tree C eative lea nin space Machine inference Genetic learning Data mining algorithm Natural language processing Neural sequential network patterns mining Text mining eal-time assessment and searching
ame and mobile tool
Source: Deloitte Research
••Genetic algorithm augmented logistic ••Graph theory: To implement preference and make appropriate regression model helps develop intelligent adaptive learning, it is recommendations accordingly. optimal study path for students imperative to understand students' Based on students' learning status to maximize learning efficiency. comprehension detail as an excellent and objectives, intelligent adaptive This model recommends the right teacher does. Given the difficulty in learning will automatically provide next lesson based on students' defining student's understanding of lessons that best suit students' learning objectives and current knowledge through comprehensive needs, at the right level of difficulty learning status, and adjusts lessons quiz testing, it is necessary to test and in the right sequence, so that in real time based on their evolving smallest unit of knowledge, in order they will not lose confidence nor feel understanding of knowledge. After to accurately understand students' unchallenged because of setting receiving students' feedback to knowledge mastery in detail. goals too high or too low. In this way, learning content pushed to them, students at different achievement ••Machine learning technology: It the system will develop student levels may gradually improve their recommends the most suitable profiles covering their learning credit scores bit by bit, i.e. from 40 content for each student based habits, interests and methods, points to 60 and 70 points, or from on individual student's preference, while automatically optimizing its 70 points to 80 and 90 points. learning habits and style. Some pushing logic. Compared with neural students favor learning in a relaxed ••Bayesian network: It predicts network algorithm used for deep and lively way, while others prefer students' learning ability in order to learning, generic algorithm enables to learn in a rigorous manner. AI help determine the timing for the comprehensive search for quick system will record each student's next stage of learning. For example, access to global optimal solution, avoiding trapped in a local optimum.
8 Development of AI-based education in China | 1. Introduction to AI-based education
the system needs to analyze learning data including learning time, test results to decide when it is duration and test performance, appropriate for the learner to start followed by data analysis to develop with linear equation in two variables, learning models for different based on the extent to which he students. Learning analytics predicts comprehends linear equation in one and monitors students' test scores variable. This requires an appropriate to provide related intervention data processing mechanism to measures. This learning model identify the connection between the helps develop personalized learning two concepts as well as students' objectives, with tailored incentives comprehension level. for each student. All students strive to grow beyond themselves, which •• Deep learning and natural language effectively avoids comparison among processing: Natural language peers, therefore improving students' processing is used to automatically sense of self-achievement. Learning generate labeled learning materials; analytics also provides teachers deep learning can be used to analyze with detailed student data, including student profiles and learning materials, how much time they have invested therefore automatically selecting and their comprehension level, with appropriate content for students from information that helps the system a vast pool of learning resources. and teachers improve teaching ••Knowledge space theory and methods. It also helps recommend information entropy theory: courses that are designed in the Information can be measured order of increasing difficulty. from the perspective of metrology. Leveraging information entropy AI-based education has developed theory, we may test key concepts in the US for a long time, with to gain deeper insights to students' a large number of users across learning status, supplemented with different ages, and both product iterative refinement to identify their and technology having evolved to an weaknesses and learning status advanced level. China, however, is still more efficiently and precisely. at the initial stage, lagging behind in ••Educational data mining and data accumulated. Despite of such learning analytics: The application situation, China is likely to catch up of big data in education mainly falls from behind, with the increase of into two categories, educational capital flowing to the area and the data mining and learning analytics. advantage of demographic dividend. Educational data mining measures and analyzes learning process and activities, collecting students'
9 Development of AI-based education in China | 2. Analysis of AI-based education systems
2. Analysis of AI-based education systems
2.1 Analysis of intelligent adaptive which is radically different from the important role in response to the education structure and model traditional education model (also changing requirements for learning. 2.1.1 Application of AI-based called factory model of education). In Intelligent adaptive teaching system education agricultural and industrial societies, is a computer system providing Disruptive application of in education more than ten years education is timely and personalized instruction includes intelligent adaptive education, enough to for people to handle the or feedback for each learner. All innovative virtual learning space, data challenges in work and life. But in AI intelligent adaptive teaching systems analysis and data privacy. Some major society, as the boundary between are designed to improve learning trends of AI application in education education and work is increasingly value and efficiency, based on multiple and key technologies supporting the blurred, lifelong learning will become computing technologies, especially application are listed in the table the norm. AI-enable intelligent computational intelligence. Such below. The report mainly discusses AI- adaptive education will play a more systems are closely related to AI-based based intelligent adaptive education, design and cognitive learning theory.
Figure 2-1: Application of AI-based education
Application of AI-based education Supportive AI-related technologies
Personalized intelligent adaptive education Real-time formative assessment, AI + intelligent adaptive teaching systems, AI tools, self-improvement systems
Transformation to active learning, PBL, CBL, computational AI (machine learning, KR&R, robotics, computer vision, NLP) thinking
Changing the definition of successful schools and students Adaptive learning method and personalized learning approach, AI-based planning
Innovative learning space AI-based classrooms, virtual labs, A/R, V/R, hearing and sensing technologies
Real-time analysis EDM and predictive analysis
Privacy and trust Digital ecosystem
Breaking limits mobile, online learning and virtual personalized assistants
Source: AI-based Adaptive Education Whitepaper
10 Development of AI-based education in China | 2. Analysis of AI-based education
2.1.2 Intelligence levels of learning and make the whole process covering learning progress and intelligent adaptive learning measurable. With intelligent adaptive teaching method. Intelligent adaptive Many companies have not reached learning systems at this level, users will education in China is still developing at the best intelligence level in their have access to multiple approaches a level below intermediate. Although intelligent adaptive learning products, to the correct answer to any question. many companies have not reached which can be classified into five levels As the most desirable AI application the best intelligence level in their in terms of the application status of in education, Level 5 products focus intelligent adaptive learning products, AI in education. Level 0 and Level on shaping students' imagination and increasingly frequent interaction 1, considered as products at the creativity, monitoring their learning throughout the educational process early stage of AI application, transfer status and analyzing emotional based on intelligent adaptive systems learning contents based on simple condition and subjective initiative, will generate more and more data, rule judgement. Level 2 products to improve learning capabilities and which will provide a clearer picture of provide courses of different difficulty. creativity and stimulate subjective the process of teaching and learning Level 3 products, as the best practice initiative. Products at this level will and enable more accurate information of intelligent adaptive learning so far, drive the all-round development of recommendation. Intelligent adaptive involve knowledge map and probability students, instead of just assisting education systems are likely to be model. Level 4 products provide students in understanding of specific used more widely. high-quality contents to teachers knowledge. and students through computer technology-based intelligent adaptive systems, to support both teaching and
11 Development of AI-based education in China | 2. Analysis of AI-based education systems
Figure 2-2: Intelligence levels of intelligent adaptive
Level of intelligent Level name Definition Stage adaptive education
Level 0 Human-based intelligent Systems recommend courses automatically based on human Early adaptive learning evaluation of students' learning abilities.
Level 1 Intelligent adaptive As a decision tree in nature, rule-based intelligent adaptive Early learning based on learning is applicable in evaluating the correctness of students' simple rules behaviors rather than their understanding of specific knowledge, to provide guidance for the recommendation of courses.
Level 2 Intelligent adaptive Item response theory (IRT)-based evaluation systems Intermediate learning based on recommend courses of proper difficulty in a real-time manner, difficulty level and IRT i.e. the difficulty level of courses provided to students is decided by their performance.
Level 3 Intelligent adaptive With regard to the entire knowledge graph, the system Intermediate learning based on leverages education data mining or Bayesin knowledge tracing knowledge network and to infer students' understanding of knowledge through probability model exercises, therefore adjusting study path in real time while recommending optimized learning tasks. The key is to develop knowledge graph that connects all key concepts, as well as task- specific dialog system powered by natural language processing.
Level 4 Intelligent adaptive Intelligent adaptive learning at this level is designed to improve Upper-intermediate learning based on students' knowledge systems, learning abilities, learning habits multiple learning and creativity, considering multiple learning elements (including elements, capability emotional factors, learning interest and motivation, subjective objectives and initiative and attentiveness) and integrating knowledge map, subdivided knowledge subdivision abilities and thinking methods. map
Level 5 Intelligent adaptive AI-enabled intelligent adaptive learning, based on human- Advanced learning based on in-the-loop, simulates students and human-machine cognitive science, NLP communication to develop teaching strategies leveraging interaction, symbolic reinforcement learning and genetic algorithms. Intelligent reasoning and deep adaptive learning provides learning scenarios and learning learning partners by combining symbolic reasoning, cognitive science and deep learning and integrating individual learning with collective learning. While increasing students' knowledge, it also helps students develop knowledge system and improve learning capability, habits and creativity.
NLP: Natural Language Processing Source: Principles behind Intelligent Adaptive Learning, Deloitte Research, Yixue Education-Squirrel AI
12 Development of AI-based education in China | 2. Analysis of AI-based education
Intelligent adaptive learning provides intelligent adaptive learning products, the process of teaching and learning each learner with personalized increasingly frequent interaction and enable more accurate information plans covering learning progress throughout the educational process recommendation. Intelligent adaptive and teaching method. Although based on intelligent adaptive systems education systems are likely to be many companies have not reached will generate more and more data, used more widely. the best intelligence level in their which will provide a clearer picture of
2.1.3 Intelligent adaptive learning system model Standard adaptive learning model is designed from three dimensions including contents, data and algorithm, integrating multiple technologies into the adaptive education system and focusing on diagnosis and assessment of knowledge understanding, adaptive learning recommendation, analysis and knowledge acquirement, based on automatic machine learning and knowledge permutation model.
Figure 2-3: Technological structure model of intelligent adaptive education
Navigation Assessment Presentation Management Interface Real-time sensor interface interface interface service
Diagnosis and Adaptive learning Sensors designed Knowledge Function Analysis assessment recommendation for specific fields acquirement
Automatic machine NPL and Knowledge IRT, BKT, KST, Model Core services PGM model learning and reinforcement semantic inference model PKS learning model
Data LRS MDS Target Student profile DW Model
Knowledge Learning Content Knowledge structure and map map metadata
Source: AI-based Adaptive Education Whitepaper, Deloitte Research
13 Development of AI-based education in China | 2. Analysis of AI-based education systems
The general model of intelligent adaptive learning system first proposed by Peter Brusilovsky from the School of Information Sciences of the University of Pittsburgurgh is divided into two key parts by technological structure: system model (including learner model, teaching model, knowledge model and interface model) and intelligent adaptive engine.
Figure 2-4: Reference model of intelligent adaptive teaching system Model plays a fundamental and core role in the intelligent adaptive learning Interface model system, with adaptive engine providing power for the system. The building of learner model and knowledge field model is critical for intelligent Adaptive engine adaptive teaching model, and requires systematic and multidimensional modeling and establishment of Contents Assessment Knowledge field association rules. Learner model model
Alignment
Teaching model
14 Development of AI-based education in China | 2. Analysis of AI-based education
01. Learner model Multidimensional behavior data of learners are generated during the learning process concurrently. Therefore, it is necessary to integrate and analyze psychological and behavioral data systematically to establish learner model. In the intelligent adaptive learning system, learner model is critical for improving independent learning capabilities. The first level of learner model focuses on students' thinking and capability, which is used to assess their learning abilities. The second level of learner model maps knowledge points, to give a full picture of learners' knowledge mastery.
Figure 2-5: Learner data model Learner data modeling, which reflects the relations between learning results Communication and a series of variables including behavior Interaction type learning materials, learning resources Facial behavior Social network and teaching behaviors, can be used Attention type Facial expression to predict the future learning trend of Intrapersonal Personality and Media students. Intelligent analysis of learner behavior specialty preference Learning style data will provide students with useful Knowledge Learning Personal learning resources and tasks. With acquired Learning emotion network more objective feedbacks provided by Way of interest Learning behavior educational data, teachers will be able thinking Knowledge not Learning Learning to optimize their education planning yet acquired attitude needs and course development, and adjust Competence educational contents and plans Internal External according to students' learning state.
02. Knowledge field model should always be prepared to decide knowledge field model database, Knowledge field model describes "what to do next", which is determined and transforms these information knowledge structure and establishes by the tutoring model that integrates into numbers. In addition, the engine knowledge structure map with researchers' teaching theories. works out the similarity value of detailed learning contents, usually characteristics with association rules, including desirable expert knowledge, 04. Interface model and recommends proper knowledge mistakes regularly made by students, User interface explains learners' objects to students to meet their rules of making mistakes and performance through multiple input personal needs based on the misunderstanding. media (voice, typing and click) and similarity value obtained. provides output (texts, figures, cartoons 03. Teaching model and agencies) through different media. Intelligent adaptive engine usually Teaching model defines the rules In addition to traditional function combines inter-knowledge point about how to obtain access to of human-machine interface, some adaptive mechanism and intra- each knowledge field based on the systems also have other functions knowledge point adaptive mechanism. information provided by student including natural language interaction, The system firstly detects students' model and how to change user model. speech recognition and learners' learning state through proper task Combining knowledge field model and emotion detection. assignment. Once students adopt learner model, teaching model allows suggestions, the system will apply teachers to figure out what instruction 05. Intelligent adaptive engine intelligent adaptive intervention to strategies and actions they should Intelligent adaptive engine collects students performing tasks. take next. In a hybrid active system, information about the characteristics learners are also likely to take actions, of students and knowledge objects ask questions or seek for help. ITS from learner model database and
15 Development of AI-based education in China | 2. Analysis of AI-based education systems
Figure 2-6: Intelligent adaptive content teaching
no led e no led e point tas no led e point tas point tas
no led e no led e point tas n te nal loop point tas
no led e no led e point tas no led e point tas point tas
Inte nal loop
tep tep tep nd
est Action el 2 Action 2 est i i actice tep tep
2.2 Three main application banks are mainly applied to spoken users’ intelligent adaptive tests and scenarios for intelligent adaptive English testing and assignments. assessments for teaching assistance. education Established brands include English Intelligent adaptive education products Liulishuo, Duolingo, etc. The third category is building fall broadly into three categories by intelligent adaptive teaching platforms application scenario: language learning The second category is teaching to integrate testing, teaching, scenario, teaching aid scenario and aid products that are well applied learning, exercise and assessment intelligent adaptive platform. in Chinese and U.S. markets. With into the intelligent adaptive learning tests and exercises at the core, these system. Several educational tech The first category is foreign language products serve as assistants driving stars including Dreambox, byju’s, learning products that score sentences teaching and intelligent education IBM Watson Education, are emerging from pronunciation and grammar and through intelligent tests. Both Chinese in the foreign intelligent adaptive assist teaching intelligently. English brands featured by iFLYTEK Education, learning sector while China is still in learning has long been an important Zuoyebang and Tencent Education and its infancy led by Yixue Education- part of the global education market foreign brands marked by Renaissance Squirrel AI with a large gap with the where well-developed learning Learning and Knewton leverage foreign market in maturity. products and data in test question their platform strengths to complete
16 Development of AI-based education in China | 2. Analysis of AI-based education
Figure 2-7: Comparison of main technical subjects in intelligent adaptive learning scenarios
Language Teaching aid Self-adaptive learning scenario scenario platform
Score sentences from Improve knowledge structure based Develop an integrated closed-loop pronunciation and grammar and on tests and exercises and assist intelligent teaching of test, exercise, Scenario details assist teaching intelligently teaching teaching and learning on self- adaptive platforms
•• China: led by To C, many English •• Developed markets at home and •• Large gap of maturity related APPs provided abroad •• Starting up in China led by To C Gap between •• Foreign markets: To B and To C •• China: tests for support and model and online intelligent model China and foreign exercises to help understand assisting offline users markets knowledge points
•• Foreign markets: intelligent testing to facilitate intelligent education
•• Oral presentation testing •• Oral presentation testing •• Plan learning paths
•• Homework assigning •• Compositions correcting •• Push learning contents
Main applications •• Layered course scheduling •• Test paper preparation and reading •• Detect disadvantages
•• Partner robot •• Photo-based questing searching •• Predict learning speeds
•• Learning attitude judgement
Source: Deloitte Research
2.2.1 Language learning scenario Language learning scenario is mainly for testing English pronunciation. With increasing attention on oral English in the English education, intelligent educational products for testing English pronunciation are developed through technologies including speech recognition and natural language processing to help learn and test the standard level of oral English pronunciation and intonation, fluency and speaking skills. The application technologies include speech recognition and natural language processing.
Figure 2-8: Illustration of language learning scenario
Students' Intelligent Online Aided self-testing analysis learning teaching •• Students can choose •• Intelligent speech •• Provide online •• Provide testing tests of different testing can score learning courses results for teachers categories for and analyse the or online teachers’ to improve self-adaptive oral oral presentation, tutoring based on teachers’ tutoring presentation pronunciation and students’ weak effeciciency grammar subjects Source: Deloitte
17 Development of AI-based education in China | 2. Analysis of AI-based education systems
Advantages of speech testing: testing has an effect on the results focus more on learning. It uses To B standard tests and detailed of intelligent scoring due to innate and To C business models and offers analysis to improve efficiency. uncertainty of spoken English, different intelligent language learning services Intelligent educational products pronunciation habits and intonation, directly for individual consumers and for English speech testing replace external factors such as noise, enterprises and schools, including teachers to help students in oral generating inaccurate analysis results. K12, language schools, universities and English practice, oral English As it cannot judge pronunciation and companies, etc. testing, and score statistics. They intonation to generate analysis and can significantly improve teachers’ feedbacks, its interactivity also needs Domestic language learning work efficiency and drive intelligent to be improved. application English Liulishuo: free adaptive oral English learning by trial class plus rebate for achieving machines assisting teaching. They English Central, a foreign language goals and “learning without can help in several oral English learning brand: customization teachers” to expand the market. categories including pronunciation at To B end and To C business It is an English learning APP in China of phonetic symbols, essay reading model to capture the market mainly designed for C-end consumers. and oral presentation to reduce by providing multiple language Users’ English levels are graded by teachers’ workload and improve versions for students of different scoring with AI through English level working efficiency, enabling teachers native tongues. tests. Varying courses are provided to provide individualized teaching in The foreign English speech testing for users at different levels to correct accordance with intelligent testing application English Central applies their pronunciations supported by results. Stable and objective score its proprietary speech recognition English exercises such as listening results of intelligent tests can be technology to listen users’ and generate analysis reports of their analyzed accurately to generate speeches and score them based on learning status. The business model efficient feedbacks. pronunciation and grammar, providing that users check in every day and services from teaching, watching, receive rebates upon their completion Disadvantages of speech testing: speaking and teaching. In terms of of class hours is gaining popularity and external factors affecting testing teaching, the one-on-one online attracting students with its intelligent results. The results of English speech teaching model allows students to robot-based teaching model, driving its market expansion.
Figure 2-9: Business model of English Liulishuo
Free trial class
Light-weighting learning materials for students to have initial capabilities before learning
Rebtate for completion of class hours Popular model to expand the market
Customized classes
Robot-based teaching, one new teaching model without human teachers
Source: Deloitte Research
18 Development of AI-based education in China | 2. Analysis of AI-based education
Figure 2-10: Comparison of main language learning products
Category English Central English Boxfish Duolingo ETS Liulishuo
Main To B Enterprises Various Provide English business and schools, international test scores for model including K12, and public reference language schools, schools universities, companies
To C One-to-one C-end users at all Provide multiple Provide exam private teaching ages language learning services model through gaming models
Main Early √ √ market childhood segments education
K-12 √ √ √ √
Higher √ √ √ √ education
Going √ √ √ abroad, exams
Subjects Detailed English listening, Oral English English class Individualized English English for exams subjects speaking, reading teaching and writing
Source: Deloitte Research
English Liulishuo, Boxfish, ETS and gaming models with a focus on Intelligent human-machine Duolingo are main products at home customized learning. EST is the largest interaction. Currently, language and abroad, all with English teaching non-profit organization for English learning products only provide one- aid services at their core and they exams by providing specialized exams way communication for users, yet have specialized in different sectors targeting K-12, vocational education being unable to offer free talk services from the perspective of their business and English exams. in educational scenarios through models and market segments. intelligent language and real-time Language testing applications in Future developments of language communication and correction. In China have their focuses. English learning products Precise speech the future, intelligent AI will empower Liulishuo is specialized in oral English recognition. Minimizing the effect of these APPs with intelligent language for users at all ages to improve their personal pronunciations and external analysis, critical thinking and logical oral English through intelligent speech factors on speech recognition and capability to judge the grammar and testing. Boxfish focuses on class testing is the technology trend of speech mistakes of users’ single teaching in close cooperation with language intelligent learning testing, sentence and simulate the judgement B-end schools to promote products which will enable intelligent judgement of comprehension and logic of directly to students. Duolingo applies to be more objective and accurate to paragraphs in real exams. the strategy of free lifetime English deliver targeted tutoring on speech, learning to help users learn through grammar and sentence for users.
19 Development of AI-based education in China | 2. Analysis of AI-based education systems
Focus on marketing and self-adaptive way based on students’ expansion. Domestic intelligent learning status. Parents can supervise language learning brands need to the status of students’ homework develop customized courses to meet online and teachers can customize varying demands from companies, teaching plans for students in different schools and individuals by customizing learning statuses based on student special language pools meeting analysis reports while the system the course and role requirements will sort out students’ mistakes and of schools and companies to align recommend exercises for them with individualized learning needs of intelligently. different groups. Besides, through more partnerships with offline More objective intelligent organizations to adopt the online correcting. In contrast to human intelligent educational models to correcting, intelligent correcting can offline organizations, offline tutoring mark mistakes and analyze reasons in or the application of AI technology time, correcting faster, more detailed to offline education will be one way and objectively. Products featured with towards the future. intelligent correcting and exercises recommendation can help teachers 2.2.2 Teaching aid products analyze students’ assignments and Teaching aid products connect provide personalized learning status students, parents and teachers, reports, reducing the time for teachers analyze reports and recommend and parents to communicate, enabling test questions more intelligently. parents to better understand students’ These products can assign homework learning status and supervise learning. online and correct intelligently and Considering the current situation, generate students’ learning status intelligent correcting is applied to reports by applying image recognition, math and English, especially widely natural language processing and data used for intelligent correcting of mining technology. Take assignments English compositions and math as an example. Teachers can assign subjective questions. Such real-time homework online which will be and objective correcting can deliver corrected by AI to generate students’ accurate analysis of learning status learning status reports and mistakes and ultimately enhance teachers’ collections, provide feedbacks for efficiency and enable students to teachers, parents and students and develop self-adaptive learning. recommend test questions in a
20 Development of AI-based education in China | 2. Analysis of AI-based education
Figure 2-11: Illustration of testing and question bank products
Analysis of learning Teachers' Self-adaptive Intelligent correcting progress assistant learning
Enable Correct and Analyze students provide students' Accurately to develop results mistakes capture APP Delivery to Teachers’ Students’ self-adaptive in time and generate students' parents teaching self-learning learning and and mark reports on learning complete reasons of learning progress recommeded mistakes status exerciese
Source: Deloitte Research
Foreign products can be Reading, Star Math and Star Early New Oriental, the scores of RealSkill customized to provide Literacy, and Star Custom. and interviewers are almost the same, adaptability for self-adaptive reaching 96.91% while the accuracy learning. Among foreign teaching China’s teaching aid product of intelligent correcting by RealSkill aid products, STAR 360 products market has been mature with a reaches 92.64% and its recognition launched by Renaissance Learning are distinctive strategy of To B model rate of handwritten texts reaches used for standardized computer self- by cooperating with organizations. 95%. Examinees can build a learning adaptive testing during K12 education Among domestic products, RealSkill closed loop composed of intelligent to assess and analyze students’ products jointly launched by iFLYTEK scoring, sentence by sentence levels of learning, advantages and New Oriental focusing on correcting, behavior analysis, intensive and disadvantages, and provide intelligent correcting and oral English sample interpretation and learning student data for teachers. Star 360 practice are designed for IELTS and record through online learning to is composed of three computer- TOTEL exams. As shown by tests of acquire new preparation experience powered self-adaptive tests: Star for exams scientifically.
21 Development of AI-based education in China | 2. Analysis of AI-based education systems
Figure 2-12: Illustration of RealSkill learning closed-loop Other companies in China include Yiqixuexi, Knowbox, Xueba100. Intelligent Yiqixuexi, for example, gathers scoring teachers to prepare and review test papers and assign homework while the system applies knowledge Sentence Learning graphs and intelligent adaptive by sentence record algorithms to collect data and push correcting individualized test questions during learning. Xueba100 offers a testing function to identify where students are weak based on exercises they complete by leveraging algorithms Intensive sample Behavior analysis while its practice function can interpretation push automatically exercises with relevant difficulties and knowledge Source: Deloitte Research points to help reinforce students’ understanding of knowledge. The system can provide automatic testing services and push relevant exercises.
Figure 2-13: Comparison of teaching aid products
Knewton Yiqixuexi Renaissance Learning iFLYTEK
Market Educational K-12 education segments levels
Business To B Education publishers, Public schools Customized courses Smart classes plus models hardware developers, cooperation with online teaching platforms, schools learning management systems, APP providers, etc.
To C For K-12 users Personalized learning
Subjects Specific No limitation on subjects All subjects English, math, etc. Customized subjects subjects
Specific Question application searching scenarios Online √ teaching assistance
Homework √
Intelligent √ √ √ √ adaptive testing
Source: Deloitte Research
22 Development of AI-based education in China | 2. Analysis of AI-based education
Domestic teaching aid products contents and developing self-adaptive various media (text, figure, speech, have grown stronger, most of which course products since 2016. By video, VR, AI, AR) will drive testing are designed for K-12 educational analyzing Knewton’s products, we and learning to develop beyond texts groups. Though the learning model can see Knewton as an AI-based but involve different teaching styles, of all subjects, 17Zuoye is oriented self-adaptive learning platform who intelligent testing and test questions towards B-end public schools and embeds partners’ contents into its of varying categories, expanding the marketing its online homework own system by API and breaks up width of intelligent learning and inspir- platform. Considering core parts of learning contents into knowledge ing students’ learning interest. products, Knewton offers learning points, digitalizes course materials paths by its own systems combining into its system and generates relevant Interactive test banks and test- with clients’ course materials, just knowledge graphs. By collecting ing will replace single exercises. building self-adaptive learning students’ behavior data consistently Teaching aid products will not only platforms to assist teachers instead through its AI-based intelligent self- allow students’ one-way inputs, but of preparing contents. For digital self- adaptive learning platform, combined provide intelligent interactive learning, adaptive learning courses, Knewton with student behavior, algorithms enabling students to raise questions embeds partners’ contents into its recommend personalized course in time and get intelligent responses systems by API and its AI intelligent learning paths, set tasks and goals, online in the system which in turn self-adaptive learning platform respond to students’ actions taken in provides tests and learning feedbacks responds to students’ actions in the system in real time. The system anytime just like a real teacher to tutor the system in real time by collecting will automatically push the next action students. students’ behavior data consistently. for students once they complete a The system will automatically push certain action. All subjects will be covered for the next action for students once differentiated competition. Cur- they complete a certain action. With Future trends of teaching rently, most teaching aid products do technologies at present, it can only aid products not cover all subjects in primary and help teaching instead of replacing Moving forward, teaching aid products junior high school teaching. Therefore, teachers. Knewton is targeting will be applied in front-line, one-to- they need to cover more subjects and publishers and educational companies one precision teaching scenarios. As online intelligent education powering and provide them with self-adaptive important teaching aid tools, intelligent multiple subject learning at the same learning engines (cloud platforms), adaptive platforms will become the time will become a trend. Online starting cooperation with schools to future trend for the following reasons: teaching aid and offline activities can offer self-adaptive course products be closely integrated to serve teach- since 2016. Being cooperative with Text, speech, image and video ing products. The growth at To B the B end as its business model, it recognition will replace single end will generate new profit sources digitalizes all course materials and text-based products. In the future, for products. They can strategically provides adaptive learning plans teaching aid products will go beyond cooperate with teaching aid platforms for these products. Knewton has text test banks but cover all learning and acquire better interpretations of also transformed into preparing resources for teaching. For example, knowledge points and feedbacks on students’ analysis reports.
23 Development of AI-based education in China | 2. Analysis of AI-based education systems
2.2.3 Intelligent adaptive teaching adaptive learning systems collect real time through consistent intelligent platform students’ learning data to plan the adaptive function and recommend Intelligent adaptive teaching best learning paths for students based teaching videos related to knowledge platform can apply general testing, on the understanding of students’ points through tests and exercises. exercise, learning and teaching present capabilities and push The system applying personalized intelligently. Students can complete automatically online teaching videos to recommendation, big data analysis the closed-loop learning through the close the loop of learning. In contrast and collection can use algorithm consistent application of intelligent to intelligent test banks, the platform technology to all teaching parts. adaptive algorithm system. Intelligent can record students’ learning data in
Figure 2-14: Intelligent adaptive teaching platform
Students/Person receiving education
Testing Learning Teaching Exercise Testing
education contents education tools education services
Knowledge domain module Teaching module Learner module ••Learning content ••Diagnosis ••Learning behavior ••Personal ••Learning ••regulation ••Supervision recommendation information behavior ••Problem knowledge domain ••Learning Plan ••Learning content ••Learning log ••Test module recommendation
AI education platform
AI technical enabler
Personalized recommendation engine: personalized learning path and learning contents recommendation
Big data analysis system: learning data analysis and processing
Big data collection system: real-time collection of behavior data of each student
Source: Deloitte Research
24 Development of AI-based education in China | 2. Analysis of AI-based education
In domestic intelligent adaptive The intelligent adaptive system it proprietary tutoring organizations to teaching segment, Yixue developed independently can simulate offer question answering services and Education-Squirrel AI works human teachers’ teaching and apply students’ psychological counseling. with tutoring organizations to AI technologies into teaching, learning, It mainly provides services mainly develop an intelligent education testing, assessment and exercise. about Chinese, math, English, physics platform model integrating online Yixue Education-Squirrel AI uses the with the business model of setting intelligence and offline support. online-offline integrated product up offline tutoring and training Yixue Education-Squirrel AI has started service model, i.e. a teaching model organizations and online intelligent early in diving into AI-based education. integrating online intelligent adaptive adaptive course tutoring. Its system It develops the first intelligent adaptive system with human teachers. On assesses students’ understanding learning engine focused on advanced online side, Yixue Education-Squirrel of knowledge graphs dynamically algorithms with complete independent AI leverages the intelligent adaptive and pushes relevant explanation intellectual property right in China. This learning system and class livestream videos, exercises and test questions engine breaks up knowledge points to provide services for students. On while teachers are responsible for into nanoscale parts and detects offline side, it cooperates with other controlling the learning pace, providing students’ understanding of these offline educational companies or guidance and encouragement and knowledge points more accurately. answering questions.
Figure 2-15: Comparison of main intelligent adaptive teaching platform companies
Yixue Education-Squirrel AI ALEKS RealizeIT BYJUs
Main models Online intelligent education Offering contents Customized courses Personalized combining with offline tutoring and self-platform platform-based organizations contents learning + technical support
Partners Offline tutoring organizations Acquired by McGraw-Hill Cooperation with Invested by Tencent, Education schools, educational Sequoia Capital, etc. organizations, publishers
Technical Independently develop intelligent Identify learners’ knowledge Update knowledge Adjust automatically performance adaptive system and simulate real learned and unlearned points and analyze based on students’ teachers’ teaching quickly and accurately based and update students’ learning habits and on knowledge space theory learning status capabilities. Provide through adaptive questions through data videos, exercises and choose knowledge that collection and analysis and tests is badly needed
Business To B+ To C model combining online To B To B To C models intelligent adaptive platform with offline teaching aid organizations
Market K-12 education K-12 education Meeting needs of C-end education at segments B-end clients all levels
Subjects Chinese, math, English Math, physics and chemistry No limitations on All subjects + courses subjects professional exams+ exams for studying abroad
Source: Deloitte Research
25 Development of AI-based education in China | 2. Analysis of AI-based education systems
ALEKS provides platforms and human teachers and animations learning models and less intelligent contents from its own platform. ALEKS to explain intricate science subject dynamic adjustments. With the PPL system includes three stages of concepts and graphics clearly in accumulation of effective student positioning, preparation and learning a simple comprehensible way. It data, advanced AI algorithms and in- while the procedures can reflect can recommend relevant courses depth understanding of practitioners unique understanding of knowledge based on students’ capabilities and of teaching and research, genuine of each student and divide them preferences to provide individualized AI-based intelligent adaptive learning into course-specific preparation and learning paths. Its gaming designs and products will be iterated in the future. learning modules. Unlike traditional game level challenges make students standard tests and written exams, feel they are playing games through Human-machine interaction ALEKS PPL focuses more on students’ such an edutainment model. will be a new teaching model in unique knowledge gaps and enables classrooms. In intelligent platforms, seamless transition from positioning Comparing with Aleks, RealizeIT, BYJU’s, AI has great potential for growth, such and assessment to targeting learning the Chinese brand Yixue Education- as thinking from human perspective modules to inspire students to gain Squirrel AI can meet international to assist teaching. As AI continues to higher scores. standards from the perspective develop, human-machine interaction of partners and subjects. In terms in classrooms will be a new teaching RealizeIt helps educational of partners, benchmarking with model and the interaction will replace organizations build customized Knewton, Yixue Education-Squirrel AI one-way course inputs. Besides, a platforms. RealizeIt is engaged in cooperates educational publishers learning style model can be built by contents and systems with the ability to and online teaching platforms like using models to analyze learning analyze and update students’ learning MOOC to acquire more platform-based styles and recommending courses status in real time and dynamically. It contents to provide more content based on the learning styles and makes money by providing intelligent tutoring services for C-end users and features of different users. This will adaptive educational systems, course customized services for B-end users. be another application of AI into contents and services for schools, In terms of subjects, Yixue Education- intelligent platforms. educational organizations and Squirrel AI is mainly engaged in publishers. As the most popular Chinese, math and English while both Parallel education and teaching learning APP in India, BYJU’s provides testing and test bank platforms at platformsIn 2014, the State Key contents covering home and abroad can offer platform- Laboratory of Management and based tutoring services without Control for Complex Systems of As the most popular learning APP limitations on subjects. Therefore, Yixue the Chinese Academy of Sciences in India, BYJU’s provides contents Education-Squirrel AI can develop all- put forward the concept of parallel covering K12, English for going abroad subject services in the future. education based on parallel and higher learning, enjoying a high intelligence and ACP theories, penetration rate in Indian market. The future of intelligent adaptive attempting to provide personalized By leveraging its platform, BYJU's education platforms precision education services (see integrates online teaching, exercise Figure 1). The upper part introduces and testing to help users visualize Refinement will be the future the basic principles of parallel all concepts via videos, delivering trend of iteration. Generally, intelligence theory, and the lower more understandable knowledge domestic AI-based intelligent adaptive part lists some parallel teaching points and concepts to improve test learning products have several platforms. Practical teaching systems scores. Exquisite videos are BYJU’s defects, including general data and human teaching systems leverage most prominent feature, integrating with fewer dimensions, incomplete data exchange and human-machine communication for interaction and iteration. This will help assess and visualise the teaching process, analyse
26 Development of AI-based education in China | 2. Analysis of AI-based education
and predict the effectiveness of ••Design various learning scenarios and various teaching strategies, therefore education innovation solutions, and helping learners obtain personalized leverage computer experiments to learning experience that best suits predict, analyze and select the best their needs. solution, to effectively guide teaching Technology roadmap for implementing and administration innovation with parallel education and teaching personalized intelligent adaptive platforms: services throughout the whole teaching process. ••Leverage emerging technologies such as big data, cloud computing, Some programs have been included IoT, virtual and augmented reality to in the 21st Century Education Grand gain multi-sourced heterogeneous Challenges of the US. Meanwhile, with data, information and knowledge the support of the Chinese Association throughout the teaching and of Automation, the research team has education administration process, established CAA intelligent education and analyze the cognitive process of committee, bringing together learning and teaching behaviors; professionals at home and abroad, ••Build knowledge graph covering to advance the research on the different subjects, establish integration of AI and education while teacher (instructor) and learner driving the development of AI talents. model, and develop education robot and personalized knowledge As the age of intelligence calls for recommendation mechanism; qualified and well-trained emerging ••Establish human teaching talents, the research team has put system that corresponds to the forward two core education concepts: practical teaching system based iSTREAM and iCDIOS. Based on on parallel education theory, the two concepts, the research therefore developing the parallel team will leverage existing scientific teaching platform. Provide precise research achievements to integrate digital portrait and performance scientific technologies with education, assessment of teaching process develop AI–based education system for administrators, teachers and for learners, and cultivate a new students through interaction and generation of AI talents. parallel execution of practical and human education systems;
27 Development of AI-based education in China | 2. Analysis of AI-based education systems
Figure2-16. New paradigm of parallel education for personalized and accurate service
Practical system Artificial system
Experiment and Learning and tutoring Manage and control evaluation
Learning Practical Learning Artificial resources teaching resources Multi- teaching model system source system data Robot Virtual Learning Practical Education defined by Artificial scenario scenario learner robot software learner Learning strategy Practical Artificial instructor instructor
Source: Chinese Academy of Science
Figure2-17. iSTREAM AI education system
i A iCDI co e Intelli ent e a - Innova- tive no mal- lea nin i ation di ection section p actice no mali a- stem- standa d- Individu- a allel education cha acte istics tion atic i ation ali ation
o ect- ea ne const uc- plo ation ame-based based methodolo cente tivism lea nin lea nin lea nin
Inte net i tual ma t cience and obots AI secto of thin s ealit manufactu in innovation
cience and Cou ses Innovation ducation base uppo tin cience valuation module innovation and ab pe ience platfo m innovation inde s stem education t ainin Cente activities e uipment
Source: Chinese Academy of Science
28 Development of AI-based education in China | 3. AI is transforming education industrysystems
3. AI is transforming education industry
With its rapid development in the past expand to other market segments and problems. In the new sector of 5 to 10 years, AI technology is changing increase user coverage. education, an intelligence trend is the way participants interact with coming as AI constantly changes the industry ecosystem. It has been put 3.1 AI drives the role change of ecosystem of education industry with into commercial application, bringing ecosystem participants data. Intelligence changes the way varying impact to different parties. Intelligent technologies redefine people study and the channel they In this context, education giants education ecosystem. Industries access learning materials, understands are investing in or independently with mature AI technology application diversified customer demands and developing smart education products. and those new sectors where AI provides customized services. Besides, Start-ups are trying to eliminate the application is still new have different education industry also face other pain points of education industry, pain points, but data collection, trends: mobility, product diversification provide more targeted solutions, processing and analysis can help and online-offline integration. effectively resolve special industry
Figure 3-1: Intelligence trend in China’s education industry
Indispensable AI technolo Constant product improvement AI technology has not been and upgrade introduced into Chinese education Diversified education products will help industry until recent years. It develops to satisfy individualize needs of users rapidly, although still in the initial stage
A e of intelli ence
Deepened mobilit Inte ation of online and offline education The trend of mobile device-based learning will further grow. Many Education companies constantly new technologies will emerge expand and integrate product chains. Offline and online education are likely to be integrated more closely, and Source: Deloitte Research both will develop harmoniously with different focuses
29 Development of AI-based education in China | 3. AI is transforming education industrysystems
AI will restructure the ecosystem existing AI-based education products AI-based education companies: of education industry. Based on AI are designed for K12 students, aiming Except for education organizations, technology, education companies to support in-class learning and more tech start-ups are transforming can provide users with AI-based improve scores. As AI technology can as AI-based education companies. education content, tools and relevant mitigate uneven allocation of quality Represented by New Oriental and services. They collect, analyze and education resources to some degree, Tomorrow Advancing Life, most give feedbacks on user data, then together with the support of education traditional education organizations apply them into five processes: informationization policies, ToB are comprehensive education teaching, learning, assessing, testing market, especially public schools will groups. They overcome technology, and practicing, and finally develop become the focus. resource, data and talent barriers customized solutions and effective to provide AI-based education feedbacks. AI will restructure the Supporting organizations: products via investment, independent relationships between participants AI technology helps change the development and external of the education industry ecosystem, responsibilities of governmental cooperation. In 2017, New Oriental improve the learning efficiency of and non-governmental education invested in 12 companies including students and redefine the education supporting organizations. They spare ARIVOC POWERED, and rolled out industry. more efforts to build smart campus an AI-based education product— and upgrade hardware facilities RealSkill, through the joint venture Intelligence trend helps transform and technologies for schools. To established with iFLYTEK—Oriental the role and responsibility of build smart campus, governments iFly, helping students prepare for the ecosystem participants. In China, grant education budgets to help oral and writing tests of TOEFL and traditional education is organized schools develop information-based IELTS, and develop home AI-education in schools by teachers in the way of management and teaching systems, robot with ASUS as content provider. face-to-face class. With the rise of establish official platforms, and These companies will make AI-based intelligence trend, traditional education advance human-machine coordination education products cover wider areas transforms. New methods such as projects (for teachers). For example, and build a better ecosystem. online teaching and intelligent adaptive tot optimize education management education are gaining popularity. decisions and processes, Shanghai Moreover, AI-based education Therefore, participants of China’s AI- establishes a big data platform to companies also include voice based education ecosystem may have deepen education data use based recognition-based online language different roles and responsibilities, on technologies including data education organizations, image meanwhile, new entrants will flock warehouse, data digging and big data recognition-based online K12 in. Participants of China’s AI-based analysis. Moreover, non-governmental education platforms that enables education ecosystem include users, organizations provide support picture-based question searching, supporting organizations and AI-based through governmental policies. Taking adaptive learning platforms (e.g. education companies: Shanghai Wei Xiao as an example, it Squirrel AI) and AI companies (e.g. gathers internet education suppliers iFLYTEK) are introducing intelligent Users: Except students, intelligence- to build an integrated learning products into schools. based education products began to resource platform . 1 attract organization users such as tutoring centers and schools. Most
1Shanghai Education Informationization 2.0 Action Plan (2018-2022), Shanghai Municipal Education Commission 30 Development of AI-based education in China | 3. AI is transforming education industrysystems
Figure3-2: China’s AI-based education ecosystem
AI-based education companies
raditiona education grou s n ine anguage education organi ations n ine education organi ations
Inte igent ada ti e earning atforms AI start u s ducation informationi ation
uppo tin o ani ations
o ernment on go ernmenta organi ations
se s
c oo s tudents o o After sc oo training organi ations Office workers
Source: Deloitte Research
3.2 Intelligence becomes the in western countries. They are mainly provide adaptive learning solutions to prevailing trend of education designed for extensive scenarios in To publishers and education companies industry B market (such as examination bodies, by digitalizing courses, and then begin 3.2.1 Chinese AI-based education schools and enterprises) and cover to provide courses by collaborating companies will shift the focus from multiple subjects of different stages with schools since 2016. It has raised To C to To B (including early childhood education, over USD180 million of funds by 2019. 2 Compared with mature overseas primary education, secondary markets, China doesn’t have AI-based education and vocational education). Chinese companies began to apply education until recent years, but it These companies can be divided into AI technology mainly in To C market holds great promise. As China’s AI- three categories: intelligent adaptive in recent years. It is still in the initial based education develops, Chinese business units of education groups stage but develops rapidly. In 2018, companies will explore To B market (e.g. Pearson provides computerized the revenue of Squirrel AI exceeded like their overseas counterparts. adaptive testing such as GMAT), RMB500 million and that of English intelligent adaptive education-based liulishuo surpassed RMB600 million. Foreign education companies applied online education platforms, such Factors including large population, AI technology much earlier. The as Coursera and Khan Academy, inadequate education resources adaptive technology can date back to and intelligent adaptive platforms and emphasis on education will 1990s. After nearly a decade, AI-based covering five learning processes, greatly accelerate the development education products have been widely such as Knewton and Aleks. As an of adaptive learning systems. Many accepted by users of all age groups intelligent adaptive platform, Knewton education companies are applying AI technologies. For example, New
2 JMDedu, https://www.jiemodui.com/N/103901 31 Development of AI-based education in China | 3. AI is transforming education industrysystems
Oriental and Tomorrow Advancing Life adaptive education companies most widely applied technology. As invest in or independently develop and AI-based intelligent adaptive competition in To C market intensifies adaptive education products. Besides, education companies. With the and government publishes policies to there are three kinds of companies: unique characteristics of connecting build information-based schools, To B intelligent adaptive platforms such all learning processes, intelligent market will become the new battlefield. as Squirrel AI, online intelligent adaptive learning becomes the
Figure 3-3: Comparison of AI-based education companies
Overseas Domestic
•• To B focused •• To C-focused Business model •• Clients including examination bodies, schools and •• Most clients are after-school tutoring companies organizations
•• The US and Europe have developed more •• At initial stage Technological sophisticated AI-based education with significant •• Core goal: to replace teachers level progress •• To assist teaching, but can’t replace teachers
•• Online intelligent adaptive education platforms •• Online intelligent adaptive education platforms (i.e. Coursera, Khan Academy) (i.e. zuoyebang, English liulishuo and 17zuoye ) •• Intelligent adaptive business units of education •• Intelligent adaptive business units of education groups (i.e. Pearson) groups (i.e. New Oriental and Tomorrow Representatives Advancing Life) •• Intelligent adaptive learning companies (i.e. Knewton, Aleks) •• Intelligent adaptive learning companies (i.e. Yixue- Squirrel AI, ) •• AI companies (i.e. iFLYTEK)
•• To have greater assistance to teaching •• Factors including large population, inadequate education resources and emphasis on education • AI reshapes learning experience, new education Prospect • will greatly accelerate the development of system is coming into being adaptive learning systems. Chinese players are likely to surpass their overseas counterparts
Source: Deloitte Research
32 Development of AI-based education in China | 3. AI is transforming education industrysystems
3.2.2 AI-based education Therefore, AI-based education has development and external companies to be the mainstream become one of the major concerns cooperation. Tencent establishes of education unicorns among industry players. Except an education branch to tighten the AI technology is disrupting education education groups, many tech bonds with education companies and industry. Customization is the biggest companies are developing AI-based organizations, and provide intelligent difference between "AI + education" education products. For example, education products to individuals, and traditional education. With New Oriental overcomes technology, schools, education organizations as technology support, customized resource, data and talent barriers well as education authorities. teaching can improve learning to provide AI-based education capability of students more effectively. products via investment, independent
Figure 3-4: Major education companies
Company Category AI products
Tomorrow Advancing Life K12 training Intelligent adaptive
Yixue Education- Squirrel AI K12 training Intelligent adaptive
VIPKID Online language learning Oral language practicing
Koolearn Online education Comprehensive
English liulihsuo Online language learning Oral language practicing
Yuanfudao K12 training Photo-based question searching
Zhangmen.com K12 training Intelligent adaptive
Zuoyebang K12 training Photo-based question searching
Onion Math K12 training Intelligent adaptive
iFLTEK Education Vocational education Intelligent adaptive
Tencent Education K12 training Intelligent adaptive
Toutiao AI learning K12 training Intelligent adaptive
Fclassroom K12 training Intelligent adaptive
iTutorGroup Online language learning -
Changingedu.com K12 home tutoring O2O -
17Edtech K12 training -
Source: Startuplab, itjuzi.com, public materials, Deloitte
33 Development of AI-based education in China | 3. AI is transforming education industrysystems
3.2.3AI-based intelligent adaptive education products gaining popularity In education industry, AI technology is penetrating into different learning processes with noticeable speed. Those education products that use data to restructure education industry are unprecedentedly revolutionary. Different from traditional education approaches, intelligent education products collect data generated in five learning processes: teaching, learning, practicing, assessing and testing, apply image and voice recognition technologies to analyse problems and develop customized solutions and effective feedbacks through deep learning, adaptation and calculation.
Figure3-5: The application of AI products in five learning processes
ive lea nin Application of AI a o technolo ies Difficulties p ocesses
o ect reference data in ad ance add Inte igent ada ti e o data fre uency ma es it eachin feed ac rocess earning system difficult to quantify Ana yse origina data and feed ac data to uman mac ine AI tec no ogy is ess smart Technical difficulty o timi e on ine teac ing systems interaction system t an teac ers
ee y ana yse earners earning atterns Inte igent ada ti e o data fre uency ma es it ro ide targeted suggestions for earners on earning system difficult to quantify ea nin adjusting learning patterns based on scientific oice recognition om icated earning met odo ogies atterns ead to ana ysis difficulty
• A y recognition tec no ogies to recogni e • Inte igent ada ti e o mac ine recognition testing resu ts su mitted y earners accuracy Assessin earning system ee en earning and e a uate resu ts umerous and com icated according to reset standards testing data •
redict ased on earners e a iour data Inte igent ada ti e arge data o ume se testing data to ma e earning ans earning• system im e and e icit feed ac estin • Image recognition data roduct• • Low analysis difficulty oice recognition roduct
•
se ig data to design a gorit m and Inte igent ada ti e arge data o ume acticin ana yse earner e a iours earning system e most direct feed ac esign customi ed uestion ortfo ios to data strengt en ea oints • The lowest analysis difficulty Ana yse uestion data to generate targeted e a uation re ort
Source: Deloitte Research
34 Development of AI-based education in China | 3. AI is transforming education industrysystems
Existing AI-based education products academic journal says that intelligent deep learning function of AI, they can be divided into three categories: adaptive systems can help improve can customize homework, testing intelligent adaptive learning products, chemistry scores for middle and lower and courses and conduct scientific photo-based searching products and middle level students. assessment. AI technology has been voice assessment products. Among introduced into schools and education them, intelligent adaptive learning Companies engaged in intelligent companies to track learning status, product is the most widely applied adaptive education are constantly record learning data, assess capability, category for it can connect all increasing globally. The number management learning progress and learning processes. of emerging intelligent adaptive connect parents with schools. By companies are growing, so do the using open big data, it can challenge Intelligent adaptive learning systems valuation of and funds raised by traditional teaching systems, make can provide real-time and customized them. Traditional internet education teaching more targeted, quantify learning solutions based on companies are shifting to or investing and visualize learning progress, and students’ learning status, including in intelligent adaptive education improve teaching and learning quality. knowledge reserve diagnosis, including Khan Academy, Coursera, Currently, inadequate user coverage competence assessment and content duolingo, 17zuoye, English Liulishuo, limits the application of learning recommendation. In the processes Tencent, and Toutiao etc. Education information management system of teaching and learning, learners giants like New Oriental and Tomorrow in K12 sector. However, as Chinese vary from each other in learning Advancing Life are also investing universities have high mobile terminal status and competence. With AI in or acquiring intelligent adaptive coverage and inefficient interaction technology, intelligent adaptive education companies. between teachers and students, course systems can use big data this system can be widely applied and algorithm to develop a set of Intelligent adaptive learning to reshape the interaction among effective and standardized courses technologies and products in home universities, teachers and students according to important and difficult and abroad have their unique based on AI technology. teaching points, offering the most characteristics. In the US and Europe, suitable courses for learners of intelligent adaptive companies Data opening is very important. different levels. Computing power mainly serve To B users, represented With open big data, education tech improvement, massive data and by Knewton, ALEKS, RealizeIt and companies can use AI technology to the application of Bayesian network DreamBox. Chinese intelligent adaptive make analysis and give feedbacks, algorithm have greatly accelerated companies are in still in the initial stage which could help companies and the development of adaptive learning and mainly serve To C users. However, schools improve teaching plans systems since 2010. Several research they develop faster in China and can and education quality. Moreover, has been proven that AI technology surpass overseas counterparts. companies can provide technical can significantly improve scores. support for offline education entities In 2018, the thesis research during The integration of AI technology and with their technical strength and data a renewed international academic learning information management reserve. Now, education companies conference found that the intelligent system is an integral part of intelligent and IT companies mainly provide open adaptive system—ALEKS improves adaptive education products. By data in vocational and K12 the pass rate of math by 15%. Another applying cloud computing and education sectors. paper published by a top international
35 Development of AI-based education in China | 4. Trends of investment in AI-based education
4. Trends of investment in AI- based education
Scientific and technological advances 4.1 China is becoming one of the in AI technology including Ireland, are accelerating development in the most active regions for investment Canada, Australia, Israel and India. In education sector. Since internet is around the globe terms of frequency of investment and being used in education, learning New AI companies continue to funds raised, the amount of investment boundaries have begun to vanish. The emerge around the world. By 2019, obtained by US companies exceeded rise of AI is significantly enhancing the number of AI adaptive learning USD2.3 billion. Although China is a users' learning efficiency. Currently, companies, being the mainstream of late-comer in AI-based education, its the wave of "AI + education" is AI-based education, exceeded 100, of unique advantages in application and becoming more intense. Players in the which 52 were from the United States, implementation make it one of the technology-based education segment accounting for half of the total number most popular regions for investment in including VIPKID, Onion Math and of the world. In addition to China, AI-based education globally. Squirrel AI are quickly staking out their AI companies are largely located in turf. The AI-enabled education sector countries or regions with advantages is bringing about a revolution in the traditional education system.
Figure 4-1: Global AI-based education investment between 2016 and Q1 2019 (incomplete statistics)
otal investment amount umbe of investment case China United States Other countries
Source: Crunchbase, Deloitte Research
36 Development of AI-based education in China | 4. Trends of investment in AI-based education
In terms of phase of development, In terms of the form of financing, Chinese players started entering the foreign AI-based education companies foreign AI-based education companies field in 2012 and the government are early-comers. In 2015, M&As of may adopt various forms of financing began to roll out various policies. In foreign AI-based education companies including the conventional venture 2015, AI-based education financing already began when China was only capital investment, as well as private saw rapid growth year-on-year, in the initial phase of development equity investment, debt financing, and heralding the arrival of the "AI + of AI-based education. Between crowd-funding. The flexibility in the education" era in China. In 2016 and 2013 and 2019, the early adaptive choice of financing channels offers 2017, AI-based education accounted learning company ALEKS, learning more development opportunities for for 4% and 7% of the total number of assessment company LearnBop, foreign AI-based companies. education deals in China respectively. online test platform Grokit, Danish In 2018, the figure rose significantly to adaptive learning company Grockit, Compared to developed countries, 19% with a total of 97 deals. and K12 education company Waggle China is a late-comer in AI-based Practice were successively acquired education. With technology lagging by or merged with large international behind the United States and other education groups including Kaplan, developed countries in the past, K12 and McGraw Hill.
Figure 4-2: Percentage of financing deals of China's AI-based education sector over the total number of financing deals of the past years
4 4
Source: itjuzi.com, Deloitte Research
37 Development of AI-based education in China | 4. Trends of investment in AI-based education
The development of China's AI- technological strength and highly respectively. In addition, AI-based based education can be examined developed education industry in China. education companies also emerged in from three aspects. First, from a From 2016 to Q1 2019, 88 deals were the Yangtze River Delta region, Fujian, geographical perspective, AI-based completed in Beijing, making it the and central and western regions. education companies located in municipality with the most frequent Although the financing in these areas Beijing, Shanghai and Guangzhou, the financing deals of AI-based education is less frequent, it reflects that the AI- developed tier-one municipalities/ companies, followed by Shanghai based education sector is reaching out cities, have obtained the most and Guangdong, with 32 and 8 deals to tier-two and tier-three cities. investment leveraging the leading
Figure 4-3: AI investment in China in 2018 (by geography)
ie cities 4
Anhui ei in ian su u ian ichuan ian i unan haan i enan han hai he ian handon iaonin uan don Chon in
Source: itjuzi.com, Deloitte Research
Secondly, in terms of segment, AI has already penetrated a number of segments including K12, language, education informationization, and quality education. In 2018, K12, education informationization, language, and quality education including STEAM became the most popular segments for investment in AI-based education, with 24, 17, 14 and 13 deals respectively. In addition, some early childhood education and vocational education companies adopting AI have also raised certain amount of funds.
38 Development of AI-based education in China | 4. Trends of investment in AI-based education
Figure 4-4: Financing in "AI + education" in China in 2018 (by segment)
4
4
ducation an ua e ualit education a l childhood ocational info mationi ation education education
Source: itjuzi.com, Deloitte Research
Thirdly, in terms of rounds of financing, A and A+) will drop rapidly, but the ••The number of education companies China's AI-based education sector amount of financing will increase entering the expansion stage exhibits the following features: accordingly. It is worth noting that (round B and subsequent the key to the transition stage is rounds) further decreases. The ••First, AI-based education start-ups companies need to establish a education sector is faced with more obtaining early-round (before growth mechanism. Therefore, the intense competition in this mature round A) financing the most state of development of a company stageFrom 2016 to Q1 2019, the frequently. For example, AI-based during the transitional round is number of companies obtaining education companies obtained early- the key to whether the company financing in the AI growth stage round financing for 79 times from can survive or become the next was 50, surpassing the number of 2016 to Q1 2019. unicorn.For example, the frequency companies obtaining financing in the ••As competition in the industry of AI-based education companies transition stage. grows, the frequency of AI-based obtaining round A financing dropped education and the education sector by half (41 times) of those obtaining as a whole obtaining financing in the early-round financing from 2016 to transitional round stage (rounds Q1 2019.
39 Development of AI-based education in China | 4. Trends of investment in AI-based education
Figure 4-5: Financing stages of AI-based education vs the education From 2016 to Q1 2019, the frequency sector between 2016 to 2019 Q1 of AI-based education companies and the entire education sector obtaining AI-based education ducation secto round A financing dropped to 41 times and 516 times respectively. 4 ounds in ea l sta e Furthermore, the number of education companies entering the expansion stage continued to decrease. From 4 ounds in t ansitional sta e 2016 to Q1 2019, only 452 companies in the education sector obtained ounds in o th sta e 4 round B (or above) financing. 50 of them obtained financing in the AI
Source: Public materials, Deloitte Research growth stage, which surpassed the *Note: Rounds in early stage (seed, angel, Pre-A); rounds in transition stage (A, A+); rounds in growth number of companies obtaining stage (round B and subsequent rounds) financing in the transition stage. The AI-based education sector in China is faced with more intense competition in the mature stage.
4.2 Investment in the AI-based Among the four segments, education obtained USD250 million of round E education segment remains informationization, English language financing and USD300 million of round fragmented training, and K12 have all obtained F financing respectively in 2018. The Currently, the combination of AI post-round C financing. While the adaptive learning company Squirrel and education is predominant in education informationization segment AI has raised more than RMB1 billion the following four segments: K12, is backed by policy, its development over the past three years. Therefore, education informationization, quality prospect is limited by the saturating in addition to the development of education, and foreign language market. AI-based education companies business of these companies, certain training. Their use of AI can be in the quality education segment are industry barrier has also been set up classified into the following main mostly in the round A stage, but their in terms of financing amount. areas: academic affairs administration financing amount basically exceeds including intelligent class scheduling, RMB100 million. In the English AI-based K12 education attendance, and test; online education language training segment, English investment has entered mature with intelligent emotion recognition Liulishuo was successfully listed in stage: The application of AI in K12 system enabled by deep learning facial the United States, Hujiang Education education is one of the most popular recognition technology; photo-based has passed the listing hearing of the application scenarios at present, question answering platform for HKEx, and VIPKID, an online English and is the AI segment obtaining K12 homework supported by image teaching company for young people, the most investment. In terms of recognition technology; and education obtained USD500 million of round D+ rounds of investment, post-round start-ups entering the market directly financing. Among the K12 homework A investment accounted for the with adaptive learning platforms. platforms, 17Edtech and Yuanfudao highest proportion, meaning that this
40 Development of AI-based education in China | 4. Trends of investment in AI-based education
portion of companies have already was 51, of which 18 were early is still a more popular segment for established growth mechanism. investment deals, 10 were round A AI investment compared to other Therefore, competition among AI investment deals and 23 were post- segments. Companies including companies in the K12 segment is round A deals. Geographically, Beijing internet giants, renowned VC/PE and increasingly fierce. From 2016 to Q1 remains the region with the largest education giants are establishing their 2019, the total number of investment number of K12 education companies presence in the K12 segment. deals in the K12 education segment obtaining investment. Currently, K12
Figure 4-6: Distribution of AI-based K12 education Figure 4-7: Investment in AI-based education between 2016 and Q1 2019 by round of financing informationization (by number of deals)
eed An e 4
re-A A 4
Source: itjuzi.com, Deloitte Research Source: itjuzi.com, Deloitte Research
Investment in AI-based years were mainly made in early stage of speaking and reading English education informationization is such as the seed round and the angel learning has become one of the most huge: The launch of the Education round. In the AI related education popular application scenario of AI in Informationization 2.0 Action informationization segment, education education. From 2016 to Q1 2019, Plan officially kicked off the phase giants and VC including TAL Education there were 25 investment deals in transition and upgrade of education and Xinrong Qihang are beginning to "AI + education" in foreign language informationization. Meanwhile, this show interest in the segment. Although training, of which over 65% of the favorable policy will help the education most of the investment is of the early companies have entered round A informationization segment become stage, the investment amount reached or subsequent rounds. This shows the target of capital pursuit. From 2016 RMB10 million. that start-up projects in the foreign to Q1 2019, a total of 26 investment language training segment have deals were completed in AI-based AI-based foreign language training entered the rapid expansion stage. education informationization. In has entered rapid expansion stage: Leveraging technological advances and 2018, investment and financing in the popularity of foreign language continuous aggregation of user data, education informationization increased training in the education sector has companies will rapidly expand the substantially, accounting for 65% of extended to the "AI + education" market with large-scale financing and the total investment deals for the four sector. With the rapid development of build up brand advantages, gradually years. In terms of rounds of investment, voice recognition and machine vision forming an industry barrier. 57.7% of investment during the four application in China, the teaching
41 Development of AI-based education in China | 4. Trends of investment in AI-based education
Figure 4-8: Distribution of AI-based foreign language training between 2016 and Q1 2019 by round of financing
4
a l sta e ound A in t ansition sta e o th sta e
Source: Public materials, Deloitte Research *Note: Rounds in early stage (seed, angel, Pre-A); rounds in transition stage (A, A+); rounds in growth stage (round B and subsequent rounds)
AI-based quality education is "AI + quality education" segment saw to technology and the sector. Their becoming the blue ocean for an upsurge in 2018. The number of application remains focused on investment: Following the launch financing deals completed in 2018 relatively vertical fields. Over time, of a series of policies, rules and alone already accounted for 78.2% of the vertical market segments of regulations in the second half of 2018, the total financing deals completed companies would further deepen. In investors have become relatively by the sector since 2016. In terms of 2019, companies will achieve multi- cautious about their investment in development stage, most financing dimensional technological penetration, AI-based education. Quality education projects of "AI + quality education" and segments will integrate together related segments have begun to grow. are still of rounds in early stage and on a large scale. Since AI-based education emerged in transition stage. 2016, the quality education segment Investors of education companies represented by STEAM from Europe 4.3 AI-based education are divided into three main types. and the United States has already investment and integration trends The first type is leading companies completed 23 financing deals. The for the next step and institutions in AI technology development of the quality education During 2018, China's AI-based such as Tencent, iFlytek and Chinese segment was supported by the education sector attracted Academy of Sciences Venture Capital state's efforts in promoting quality investments from leading players in Management Company. The second education. In addition, segments AI technology, VC/PE and education type is renowned VC/PE such as focusing on K12 and English language giants, including K12 giants like ZhenFund and Sinovation Ventures. teaching have already entered the red New Oriental and TAL Education, The third type is education giants such oceans. Leveraging new education internet enterprises like Tencent, as TAL education, New Oriental and philosophies, rapid development as well as renowned capital firms Net Dragon. in digital technology, and strong like Sequoia Capital, Matrix Partners support from relevant policies, quality China, ZhenFund and Blue Elephant education has become the new blue Capital. At present, AI-based education ocean for the education sector. The companies are not fully opened up
42 Development of AI-based education in China | 4. Trends of investment in AI-based education
The investment of leading players 30%, 10%, 10% and 20% respectively. in AI technology is inclining With abundant technologies and funds, towards more mature start-up leading players in AI technology also projects focus on overseas "AI + education" The investment made by AI technology market. In 2017 and 2019, Tencent companies such as Tencent and iFlytek invested in Indian AI-based education as well as technology investment company Byju's. In contrast to Tencent, funds supported by Ministry of Science AI technology company iFlytek focused and Technology has penetrated in all their investment on the early stage. segments in AI. In terms of investment Among iFlytek's investment in education areas, projects chosen by investors are between 2016 and Q1 2019, almost in various stages of their future industrial all companies were "AI + education" strategic plan. In terms of rounds of companies. In terms of segment, their investment, rounds in the growth stage investment mainly focused on education (post-round B) have absolute dominance informationization and quality education. over other rounds. Among Tencent's In addition, projects iFlytek invested in nine investment deals in AI-based were mostly start-up projects incubated education between 2016 and Q1 2019, under the company. round C, round D, round E, round F and strategic investment accounted for 30%,
Figure 4-9: Percentage of rounds of investment in "AI + education" by Tencent between 2016 and Q1 2019
Round C
Round D
Round E
Round F
Strategic investment
Source: Public materials, Deloitte Research
43 Development of AI-based education in China | 4. Trends of investment in AI-based education
VC/PE investment became more to incomplete statistics, among the 38%, 21% and 25% respectively. In rational with a focus on projects investment in education made by terms of investment amount, the with strong technology and renowned VC/PE between 2016 and amount of investment made in round implementation 2018, the percentage of investment A and subsequent rounds by VC/PE 2018 marked a different year for in "AI + education" varied between exceeded RMB10 million. In the future, venture capital and private equity 5% and 50%. The percentages as the market gradually turns rational, investment. Compared to the of investment in "AI + education" companies with advantages in previous investment landscape of made by investment firms including technology, data and business model AI projects, VC/PE investment has Sequoia Capital, Fortune Capital are more likely to attract to VC/PE. become more rational. According and Sinovation Ventures were 47%,
Figure 4-10: Statistics of frequency of investment in education made by major VC/PE between 2016 and 2018
4 4 4 4 hen und lue inovation at i e uoia ID o tune hun ei lephant entu es a tne s Capital Capital und Capital China
Number of deals of investment in education Percentage of investment in "AI + education"
Source: itjuzi.com, Deloitte Research
44 Development of AI-based education in China | 4. Trends of investment in AI-based education
Education giants accelerate their plans in the white space, bringing about the era of extensive investment and M&As of education giants in the "AI + education" sector From 2004 to 2018, a number of iconic adaptive learning start-ups worldwide have been acquired by large international education groups. Such trend has reached a peak in 2018. In particular, the investment and financing trends of adaptive learning pioneering countries around the world also have certain indicative effect on the future development trends of China's AI-based education sector. In the future, as competition of AI in education in China enters the red oceans, China's education sector may continue to scale up investment while carrying out M&As on companies in certain segments, just as what education giants in Europe and the United States did.
Figure 4-11: Major acquisitions of overseas adaptive learning companies between 2013 and Q1 2019 (incomplete statistics)
Intelligent Adaptive learning Location Year of acquisition Acquirer company
Lexia Learning System United States 2013 Rosetta Stone
Grockit United States 2013 Kaplan
ALEKS United States 2013 McGraw-Hill Education
Area9 Denmark 2014 McGraw-Hill Education
LearnBop United States 2014 K12
Think Through Learning United States 2016 Imagine Learning
Gradescope United States 2018 Turnitin
Fishtree United States 2018 Follett Corporation
Knowre United States 2018 Daekyo Investment
Grovo United States 2018 Cornerstone OnDemand
Carnegie Learning United States 2018 CIP Capital
StudyBlue United States 2018 Chegg
Waggle Practice United States 2019 Houghton Mifflin Harcourt
Knewton United States 2019 John Wiley&Sons
Source: Crunchbase, Deloitte Research
45 Development of AI-based education in China | 4. Trends of investment in AI-based education
Among investment in "AI + education", K12. In contrast, TAL Education the investment will expand the market investment of traditional education has focused their investment on for other participants and enable giants has played an important part. education informationization. In synergistic collaboration with the According to statistics, investment of addition, New Oriental also have traditional business of the company. New Oriental and TAL Education in invested in the K12 and language In the future, as competition of AI in AI-based education accounted for 11% related segments. In terms of rounds education in China enters the red and 21% respectively of investment of investment, post-round B projects oceans, China's education sector may in the education sector as a whole. in the growth stage are more likely continue to scale up investment while "AI + Educatioin" segments that to attract education giants. Overall, carrying out M&As on companies New Oriental are gaining footholds before making investment, both New in certain segments, just as what in include informationization, early Oriental and TAL Education have education giants in Europe and the childhood education, language and taken into consideration of whether United States did.
Figure 4-12: Percentage of investment in "AI + education" by education giants between 2016 and Q1 2019
4
e iental A ducation
Investment in education Percentage of investment in "AI + education"
Source: crunchbase, itjuzi.com, Deloitte Research
46 Development of AI-based education in China | 5. Future challenges, prospect and thinking of AI-based education
5. Future challenges, prospect and thinking of AI-based education
With learning experience being used to form effective assessment "teaching" to "educating" as well as reshaped by AI and the gradual to drive learning improvement. The communicating on an emotional level. formation of a new education acceptance of implementation of From the perspective of people's system, the development of education scenarios by the traditional livelihood, the development of AI- China's education sector is education system will become a based education will facilitate the fair approaching towards an intelligent tough issue for "AI + education". For allocation of education resources, era. Nevertheless, AI-based example, in the K12 segment, students an insight already perceived by the education is still faced with spend most of their time learning at government and education decision- various issues in the course of its school. Schools under the traditional makers. From 2016, a series of new development. For "AI + education" to education system are sticking to the education policies on intelligentization be fully implemented, there remains a traditional learning model. Therefore, and informationization are driving the number of challenges: implementing "AI + education" development of AI in education. requires schools, teachers, parents First, there is insufficient data in AI- and students to work together. The data required for machine based education. Data and algorithms learning is usually highly customized. are the core of AI-based education. The development of AI in When used for assessing students' Currently, the gap in core algorithms education is still seeing bright performance, data security may technology among AI-based education prospects. To better implement become the key bottleneck of the use companies is not significant except AI-based education and make of AI. The development of AI-based individual teams. Data is the major it the future of education, we education in China has been relatively issue encountered by the vast majority need to rethink the role of AI and lagged behind partly due to the high of education companies at present. AI education in society: requirements on algorithms and requires massive precise and labeled the huge R&D investment required data. However, AI-based education Educators and decision-makers have in the early stage. However, the companies cutting in from different to understand AI in a broader learning "marginal cost" is not high in actual education segments all encounter context. From the perspective of implementation. Once the issue on different degrees of data insufficiency. educators, AI-based education brings algorithms is solved, precision data will In addition, data is not effectively challenges to teachers' development become a key factor of whether AI can assessed. The key obstacle of "AI mechanism and changes in their be practically implemented. With the +education" is the absence of a closed future roles. Of the teaching aspect support of advanced algorithms and loop of learning data in the education of education in the future, teachers data in the future, AI-based education sector. Certain important parts such as are likely to be completely replaced will achieve the level of the special- learning process data and knowledge by AI. However, in terms of educating grade teacher, and the efficiency of mastery data are still missing. people, it is not possible for AI to knowledge acceptance by users will be Therefore, at this stage, AI cannot be replace teachers. By then, the role significantly enhanced. of teachers is likely to change from
47 Development of AI-based education in China | 5. Future challenges, prospect and thinking of AI-based education
In the future, the role of AI in supporting human development will be increasingly important, bringing fundamental changes. Like every scientific and technological progress that was made, legal issues around such technology would emerge accordingly. The issue on formulating laws, rules and regulations in the development process of AI is also a challenge faced by the industry. Gartner predicted that by 2020, AI and machine learning would create 2.3 million jobs while eliminating 1.8 million. In this case, whether job will be eliminated or created largely depends on the industry. For example, healthcare, education and the public sector would be benefited. To maximize value, we should focus on the AI industry to enrich the industries of the future, and reset existing rules and policies to create new industries. Changing our own culture will allow us to quickly adapt to opportunities brought by the AI science and technology.
48 Development of AI-based education in China | Deloitte China Contacts / Whitepaper Editorial Board
Deloitte China Contacts
Taylor Lam Frank Li Deloitte China Technology, Deloitte China Technology Media & Telecommunications Sector Leader Industry Leader Tel: +86 10 8520 7290 Tel: +86 10 8520 7126 Email: [email protected] Email: [email protected]
Charlotte Lu Roger Chung Deloitte China Education Sector Deloitte China Technology, Media Leader and Telecommunications Industry Tel: +86 21 6141 1801 Research Director Email: [email protected] Tel: + 86 21 2316 6657 Email: [email protected]
Whitepaper Editorial Board
Steering committees: Deloitte China, Intelligent Education Special Committee of Automation Association
Editorial Board member: Taylor Lam, Charlotte Lu, Frank Li, Wang, Feiyue, Wang, Wanliang, Roger Chung, Iris Li, Liu, Xiwei, Gong, Xiaoyan
49 Office locations
Beijing Hefei Shenzhen 12/F China Life Financial Center Room 1201 Tower A 9/F China Resources Building No. 23 Zhenzhi Road Hua Bang ICC Building 5001 Shennan Road East Chaoyang District No.190 Qian Shan Road Shenzhen 518010, PRC Beijing 100026, PRC Government and Cultural Tel: +86 755 8246 3255 Tel: +86 10 8520 7788 New Development District Fax: +86 755 8246 3186 Fax: +86 10 6508 8781 Hefei 230601, PRC Tel: +86 551 6585 5927 Suzhou Changsha Fax: +86 551 6585 5687 24/F Office Tower A, Building 58 20/F Tower 3, HC International Plaza Suzhou Center No. 109 Furong Road North Hong Kong 58 Su Xiu Road, Industrial Park Kaifu District 35/F One Pacific Place Suzhou 215021, PRC Changsha 410008, PRC 88 Queensway Tel: +86 512 6289 1238 Tel: +86 731 8522 8790 Hong Kong Fax: +86 512 6762 3338 / 3318 Fax: +86 731 8522 8230 Tel: +852 2852 1600 Fax: +852 2541 1911 Tianjin Chengdu 45/F Metropolitan Tower 17/F China Overseas Jinan 183 Nanjing Road International Center Block F Units 2802-2804, 28/F Heping District No.365 Jiaozi Avenue China Overseas Plaza Office Tianjin 300051, PRC Chengdu 610041, PRC No. 6636, 2nd Ring South Road Tel: +86 22 2320 6688 Tel: +86 28 6789 8188 Shizhong District Fax: +86 22 8312 6099 Fax: +86 28 6317 3500 Jinan 250000, PRC Tel: +86 531 8973 5800 Wuhan Chongqing Fax: +86 531 8973 5811 Unit 1, 49/F 43/F World Financial Center New World International Trade Tower 188 Minzu Road Macau 568 Jianshe Avenue Yuzhong District 19/F The Macau Square Apartment H-N Wuhan 430000, PRC Chongqing 400010, PRC 43-53A Av. do Infante D. Henrique Tel: +86 27 8526 6618 Tel: +86 23 8823 1888 Macau Fax: +86 27 8526 7032 Fax: +86 23 8857 0978 Tel: +853 2871 2998 Fax: +853 2871 3033 Xiamen Dalian Unit E, 26/F International Plaza 15/F Senmao Building Mongolia 8 Lujiang Road, Siming District 147 Zhongshan Road 15/F, ICC Tower, Jamiyan-Gun Street Xiamen 361001, PRC Dalian 116011, PRC 1st Khoroo, Sukhbaatar District, 14240- Tel: +86 592 2107 298 Tel: +86 411 8371 2888 0025 Ulaanbaatar, Mongolia Fax: +86 592 2107 259 Fax: +86 411 8360 3297 Tel: +976 7010 0450 Fax: +976 7013 0450 Xi’an Guangzhou Room 5104A, 51F Block A 26/F Yuexiu Financial Tower Nanjing Greenland Center 28 Pearl River East Road 6/F Asia Pacific Tower 9 Jinye Road, High-tech Zone Guangzhou 510623, PRC 2 Hanzhong Road Xi'an 710065, PRC Tel: +86 20 8396 9228 Xinjiekou Square Tel: +86 29 8114 0201 Fax: +86 20 3888 0121 Nanjing 210005, PRC Fax: +86 29 8114 0205 Tel: +86 25 5790 8880 Hangzhou Fax: +86 25 8691 8776 Zhengzhou Room 1206-1210 Room 5A10, Block 8, Kailin Business Center East Building, Central Plaza Shanghai No.51 Jinshui East Road No.9 Feiyunjiang Road 30/F Bund Center Zhengdong New District Shangcheng District 222 Yan An Road East Zhengzhou 450018, PRC Hangzhou 310008, PRC Shanghai 200002, PRC Tel: +86 371 8897 3700 Tel: +86 571 8972 7688 Tel: +86 21 6141 8888 Fax: +86 371 8897 3710 Fax: +86 571 8779 7915 / 8779 7916 Fax: +86 21 6335 0003
Harbin Shenyang Room 1618, Development Zone Mansion Unit 3605-3606, Forum 66 Office Tower 1 368 Changjiang Road No. 1-1 Qingnian Avenue Nangang District Shenhe District Harbin 150090, PRC Shenyang 110063, PRC Tel: +86 451 8586 0060 Tel: +86 24 6785 4068 Fax: +86 451 8586 0056 Fax: +86 24 6785 4067 w w w About Deloitte Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited (“DTTL”), its global network of member firms, and their related entities. DTTL (also referred to as “Deloitte Global”) and each of its member firms and their affiliated entities are legally separate and independent entities. DTTL does not provide services to clients. Please see www.deloitte.com/about to learn more.
Deloitte Asia Pacific Limited is a company limited by guarantee and a member firm of DTTL. Members of Deloitte Asia Pacific Limited and their related entities, each of which are separate and independent legal entities, provide services from more than 100 cities across the region, including Auckland, Bangkok, Beijing, Hanoi, Hong Kong, Jakarta, Kuala Lumpur, Manila, Melbourne, Osaka, Shanghai, Singapore, Sydney, Taipei and Tokyo.
The Deloitte brand entered the China market in 1917 with the opening of an office in Shanghai. Today, Deloitte China delivers a comprehensive range of audit & assurance, consulting, financial advisory, risk advisory and tax services to local, multinational and growth enterprise clients in China. Deloitte China has also made—and continues to make—substantial contributions to the development of China's accounting standards, taxation system and professional expertise. Deloitte China is a locally incorporated professional services organization, owned by its partners in China. To learn more about how Deloitte makes an Impact that Matters in China, please connect with our social media platforms at www2. deloitte.com/cn/en/social-media.
This communication contains general information only, and none of Deloitte Touche Tohmatsu Limited, its member firms, or their related entities (collectively the “Deloitte Network”) is by means of this communication, rendering professional advice or services. Before making any decision or taking any action that may affect your finances or your business, you should consult a qualified professional adviser. No entity in the Deloitte Network shall be responsible for any loss whatsoever sustained by any person who relies on this communication.
©2019. For information, contact Deloitte China. Designed by CoRe Creative Services. RITM0347185