[Preview of the Weekly] The Profound Impact of ChatGPT: It's Time to Seriously Discuss the Future of Education (AI Translation)

【周刊提前读】ChatGPT冲击深远：是时候严肃讨论教育的未来了

Everyone is talking about artificial intelligence (AI)—at the end of November 2023, over the course of two days, the World Innovation Summit for Education (WISE) was held in Doha, the capital of Qatar. The summit brought together more than 300 speakers from around the world in education, business, and academia, with thousands of attendees. AI emerged as the undisputed focal point. Although participants might not be familiar with Spanish, Portuguese, or Arabic, they could frequently discern keywords like "ChatGPT" from another unfamiliar language. This astonishing chatbot has been around for a full year now, and the tremors it has sent through global education continue unabated.

It's hard not to be captivated by such an atmosphere. "It makes me think of the Renaissance," one of the WISE speakers, Isabelle Hau, told Caixin. She likened the current scene to Europe's most famous cultural movement in modern history: "It feels as though we are re-entering a period of great questions and explorations. The wave of AI has reignited contemplation on the fundamental issues of education—what is learning? What is education? Which skills and capabilities are uniquely human? What makes us human?"

Isabella Hao currently serves as the Executive Director of the "Stanford Learning Accelerator" at Stanford University's Graduate School of Education. This interdisciplinary center aims to collaborate with researchers, educators, entrepreneurs, and others to develop learning solutions while greatly benefiting from the new attention and opportunities brought by ChatGPT. "GPT has given us a huge boost," recalls Isabella Hao, "The number of AI courses on campus has doubled, and all AI courses are popular with students." To explore innovative applications of generative AI in education, the Stanford Learning Accelerator has partnered with the Stanford Institute for Human-Centered Artificial Intelligence (HAI) to distribute a total of $625,000 in seed funding to 16 teams. "All researchers are working on projects, and students are full of interest."

This can be seen as a snapshot of the "AI education boom." Over the past year, teachers, schools, and governments around the world have gone through stages of caution, skepticism, acknowledgment, and acceptance towards ChatGPT. Some countries and regions have already decided to fully embrace AI within their education systems.

In February of this year, Singapore announced plans to gradually integrate ChatGPT into its school education system, becoming the first country to open the doors of education to generative AI. In June, Hong Kong, China, introduced new junior high school AI teaching materials that include content from ChatGPT and Midjourney, promoting them to over 400 public middle schools. In October, Australia's public education system reversed its previous stance of exclusion and announced that generative AI tools will be allowed in classrooms starting next year. The country's Minister for Education candidly stated: "AI is not going away; we must learn how to deal with this new technology."

Educators and researchers have come to realize that AI will not only affect "how we teach" but also challenge "what we teach," with a mix of hope, anticipation, and urgency. "Imagine a future where AI is ubiquitous; do we really want students to spend a lot of time learning things that calculators can do or teaching them knowledge they won't need in the future?" asks Conrad Wolfram, a British technology expert who has long focused on math education. Speaking at the World Innovation Summit for Education (WISE), he fiercely criticized traditional math curricula as being "80% wrong," having strayed far from real-world applications and needs.

On the other hand, no matter how exhilarating AI may be, people are still trapped by traditional educational challenges: a shortage of teachers, compromised educational rights for disadvantaged groups, and a disconnect between higher education and labor market demands. During the opening ceremony of WISE (World Innovation Summit for Education), Sheikha Moza bint Nasser, Chairperson of Qatar Foundation, lamented that since the outbreak of the Israeli-Palestinian conflict, 36 Palestinian schools supported by Qatar's "Education Above All" Foundation have been partially or completely destroyed, wiping out hard-earned educational achievements in an instant. How to envision the future of education in a world rife with "old problems" such as conflict and injustice? This serious question continues to challenge educators and policymakers.

Change and transformation continue to ferment, and from a global education event that brought together various countries and stakeholders, one might glimpse the vision and direction of education under the technological revolution.

"Omnipresent"

The most significant impact of ChatGPT on education has already surpassed the generative AI it represents. Multiple interviewees at a conference told Caixin that GPT has unprecedentedly advanced a consensus: we will live in an era of AI. "Ubiquitous" is a term frequently used by speakers when discussing the role of AI in the future world and in future education.

Such a consensus has formed the underlying theme of global educational development and innovation over the past year, leading to a unique phenomenon: even less "technical" educational projects have begun to explore the potential application of AI technologies such as machine learning and large language models (LLMs).

The 2023 WISE Education Award winner, Safeena Husain, founded Educate Girls in India as a testament to the impact of such initiatives. In the past, Educate Girls had to visit households one by one in Indian villages to bring girls who had dropped out or never attended school back into the educational system. Today, Educate Girls is collaborating with IDinsight, a global development data analysis organization, to leverage machine learning technology for more accurately pinpointing the distribution of out-of-school girls. IDinsight has cleaned and integrated data from over 200,000 villages in northern India and spent nearly a month developing a proof-of-concept tool designed to predict the number of out-of-school girls in these villages.

The effectiveness of this "cross-sector" collaboration exceeded expectations: before partnering with IDinsight, the "Girls' Education" program was able to identify an average of 10 to 15 out-of-school girls per village; with the aid of algorithms, the program can now find an average of about 46 out-of-school girls per village. By such estimates, solely through AI targeting, "Girls' Education" could cover an additional 600,000 dropout girls within the same budget.

"In the first 10 years of founding 'Girls' Education,' we successfully rescued 340,000 out-of-school girls. At this rate, it would take me 45 years to get 1.5 million girls back to school; but now, we can accomplish in 5 years what would have taken 45," Hussein marveled. "In the next decade, our goal is to reach 10 million girls. AI can help us find girls in difficult situations more quickly and provide education that is scalable and of high quality."

Lasse Leponiemi, co-founder and executive director of the Finnish education organization HundrED, has also taken note. Each year, HundrED collects and recognizes 100 global educational innovation cases. In 2022, there was much discussion about the challenges that the COVID-19 pandemic brought to families and schools; by 2023, the backdrop has quietly shifted from the "post-pandemic era" to the "AI era." Among the thousands of case applications received by HundrED, AI/EdTech (Artificial Intelligence/Educational Technology) has become the most frequently occurring keyword.

"After ChatGPT, we began to see many different types of educational solutions that make full use of AI in various ways," Lasse Leponiemi told Caixin. "Although the majority of the educational innovations recognized by HundrED are not purely technological solutions or applications, they generally incorporate online or technological elements." He also mentioned that this year, many early-stage innovative projects failed to make the list because they did not meet the requirement of "submitting one year's data." It is expected that in the coming years, more educational technology innovations will be featured in HundrED's recognition cases.

Educators are rapidly becoming more receptive and enthusiastic about AI, with the importance of "learning AI" being increasingly emphasized. At WISE, an information technology education project from the small Eastern European country of Moldova has garnered attention. Initiated by the Moldovan Association of Information and Communications Technology Companies (ATIC), the project aims to provide Moldovan students with a variety of free courses geared towards the IT industry, including AI, programming, graphic design, and entrepreneurship courses. It has already reached nearly 400 schools within Moldova, benefiting over 50,000 students.

Ana Chirita, the Strategic Project Director of ATIC, explained that in addition to offering free courses, the project also organizes competitions every year to help students apply the skills they have learned in practice. For example, students from rural areas can develop applications for local restaurants, schools, and shops and profit from them. The project has received support from the Moldovan government in terms of platform and content. Chirita stated that the government views students' information technology capabilities as key to enhancing future human capital competitiveness: "As a landlocked country in Eastern Europe, we don't have natural resources; we don't have oceans; we don't have mountains. Apart from human capital, we have nothing."

Data, both big and small, more vividly illustrates the ubiquitous resonance of AI and education throughout the year: A survey by educational technology company Chegg this year, which covered university students from 15 countries, showed that 40% of students have already started using AI, and 65% wish for AI-related training to be included in their courses; a summer survey by public opinion research firm Impact Research found that nearly everyone knows what ChatGPT is, with 63% of teachers reporting they have used ChatGPT in their work, and acceptance of AI in the classroom even surpassing that of students; meanwhile, according to a report by third-party research firm MRFR, as more educational institutions adopt AI services, the market size for AI in education is projected to reach $23.82 billion by 2030, with a compound annual growth rate of 38%.

Let's return to the "tipping point" itself, ChatGPT and generative AI. Over the past year, what kind of impact has it had on businesses, schools, and educational systems?

The Turbulence in EdTech

Educational technology companies have been particularly active at this year's WISE, ranging from well-known education firms such as Duolingo, VIPKID, and Taleemabad to startups like TeachPitch and LiveTech.AI. Developers and entrepreneurs with a keener sense of technology and faster adaptability are drawing widespread attention for how they will apply and develop ChatGPT in the field of education.

Duolingo is one of the earliest educational enterprises to integrate with GPT-4. On the same day that GPT-4 was released, Duolingo announced the launch of Duolingo Max, which is based on GPT-4 and features two new functions: "Role Play" and "Explain My Answer." "When GPT first appeared, our CEO paid attention to it earlier than most people. He believed in the potential future encapsulated in AI, especially generative AI, and wanted to incorporate this technology into Duolingo," Uriel Kejsefman, Senior Product Manager at Duolingo, told Caixin.

In fact, long before the emergence of ChatGPT, educational technology companies had already invested considerable effort in AI. Among the most well-known product strategies is "adaptive education," which involves collecting and analyzing dynamic and static information about learners, integrating relevant knowledge graphs, and customizing the difficulty level and learning pace of instruction to achieve what is referred to as "personalized" and "precision learning." Duolingo is no exception—Michael Korda told Caixin that before launching Max, Duolingo had already established a relatively complex machine learning model. "(The model) intelligently learns from the mistakes users make during their study process, thereby pushing new material and exercises that are most useful to the user."

The advent of GPT has once again changed the rules of the game: in certain scenarios, users can now take a more proactive role in deciding what to learn and how to learn, rather than simply receiving "smart" suggestions from software. "Now we can have a conversation that is completely driven by your responses. When you mention that you're about to travel to Qatar, the entire conversation will revolve around Qatar. The level of personalization is astonishing," said Ke Daoming. After integrating GPT-4, Duolingo's "role-playing" has become one of the new features with the most positive user feedback. "It allows conversations to appear as if they are happening in the real world, an effect that only generative AI can achieve."

However, the immense potential of generative AI not only heralds new opportunities but also brings more intense developmental pressure. On one hand, the "large model fine-tuning" approach supported by GPT significantly shortens the development cycle and reduces human resource investment for educational applications, intensifying competition among edtech companies in product research and development as well as iteration. On the other hand, ChatGPT's own impressive performance could substantially squeeze the existing business models of edtech enterprises: a WISE attendee told Caixin that after using ChatGPT, her frequency of using writing aids like Grammarly and Quillbot has drastically decreased.

Among the early adopters of GPT-4 in the education sector, the rise and fall of online tutoring company Chegg reflect the double-edged sword effect of this revolutionary technology: Although Chegg has strived to demonstrate that its AI learning companion powered by ChatGPT, CheggMate, is more effective than using ChatGPT alone, users are more inclined to turn directly to ChatGPT and stop paying for Chegg's homework help services. On May 1st this year, Chegg CEO Dan Rosensweig acknowledged during an earnings call that ChatGPT "has had an impact on the rate of new customer growth." The following day, Chegg's stock price plummeted by 48%, nearly halving in value; stocks of Duolingo, Pearson, and Udemy were also affected during the same period, falling between 5% and 15%.

Some young edtech entrepreneurs are forging new paths. Younes Bensouda Mourri, a Stanford University alumnus and founder of the educational startup LiveTech.AI, initially aimed to provide live tech courses for aspiring software engineers. However, with the advent of GPT, Mourri quickly pivoted, embarking on the development of a K12 school management system powered by generative AI. "Education department officials, principals, teachers, students, and parents can all log into the system and operate on different dashboards—teachers can create lesson plans and grade essays; students can receive instant feedback and personalized assignments; schools can review recent performances of various teachers; education departments can generate analytical reports based on specific regional data, schools, courses, and subjects," Mourri told Caixin.

The combination of "Education SaaS + GPT" is not unusual in itself, but Mu Li has not entered the highly competitive English market where large models perform best. Instead, he has turned to smaller language nations in search of opportunities. This year, LiveTech.AI has collaborated with King Abdullah University of Science and Technology in Saudi Arabia to build AI-driven education across all levels and disciplines within the country. "Large models work well in English, but for other languages, such as Arabic, the performance of large language models is still not satisfactory. Language disadvantages can affect the performance of corresponding countries in applying large models in fields like education, healthcare, and military. What we aim to do is help these languages keep pace," he explained.

The considerations extend beyond that. Murli is ambitious and openly expresses his interest in the advantages of Middle Eastern countries like Saudi Arabia, Qatar, and the United Arab Emirates, which differ from those in Europe and America—particularly their stronger capabilities for centralized infrastructure development. "This includes the construction of infrastructure and centralization, such as establishing data centers, cooling systems, reserving GPUs, and connecting GPUs with school software systems," Murli observes. He believes that the future competition among AI education enterprises will be a "rich man's game," which makes him place special emphasis on B2B, especially B2G business. "Only with access to large-scale GPUs and infrastructure can one perform better."

This may reflect another surging undercurrent in the field of educational technology: an increasingly broad and deep collaboration between private enterprises and public institutions. Leveraging cutting-edge AI technology, companies aim not only to provide personalized educational services to consumers but also aspire to play a role as "education solution providers" in the government-led digital transformation, seeking their share of the pie. In interviews, Murry repeatedly expressed his attention to the development of China's education digitization, particularly the construction of digital foundations at various levels of education, noting that "if used properly, these accumulated education data could generate tremendous value."

However, all applications and promises related to the combination of education and technology must ultimately withstand the test of the educational front line—the classroom.

The Game in the Classroom

In contrast to the ever-changing EdTech market, the classroom presents a different scene. Faced with a plethora of educational applications, teachers' feedback is often divided. Joe Clement, one of the hosts of WISE and a high school teacher in the United States, has complained that free time in school schedules has shifted from lively conversations between students and teachers to silent screen reading. During conference discussions, many teachers frequently express their confusion: "This idea sounds good, but can it really be implemented in the classroom?" "Educational products are becoming more powerful in terms of their assistive features, but how can we ensure they don't harm students' original spirit and creativity?"

It is noteworthy that while ChatGPT has taken the world by storm, the United Nations Educational, Scientific and Cultural Organization (UNESCO) expressed support for a global ban on smartphones in primary and secondary schools in its "Global Education Monitoring Report" published in August this year. "Not all educational technologies have had positive effects," the report stated, citing an analysis of a study on an AI learning and assessment system. It pointed out that although more than 25 million students in the United States use this system, it is no better than traditional classroom teaching in terms of improving grades.

"According to a British survey, only 7% of educational technology products have undergone randomized controlled trials that can provide data to prove their effectiveness. Meanwhile, a 2017 U.S. survey indicated that only 11% of education decision-makers consider whether there is reliable evidence attached to an educational tech product before purchasing it," Isabella Hao told Caixin. "This means that 89% of the purchasing decisions are influenced by peer recommendations, market evaluations, or other factors, rather than being based on reliable evidence."

Earlier this year, Stanford University and Google hosted a seminar on generative AI on campus, with about 50 Google engineers and 50 computer scientists from Stanford participating. "I listened to everyone's presentations, and throughout the entire day, not a single word was mentioned about education," recalls Isabella Hao, who was struck by this detail. She felt a divide: "Education is not at the core of large language models; our tech companies and scientists do not prioritize education and have not received any training in learning sciences. On the other hand, educators have accumulated much knowledge proven effective for students through research but are unable to apply such knowledge to technological practice."

A major criticism of various AI educational technology products is their excessive focus on efficiency. Functions such as "precision problem recommendation" and "accurate assessment," which were widely marketed even before ChatGPT, mostly optimize learning outcomes and efficiency through algorithms built upon existing accumulations of factual data: they enable students to master more knowledge points in less time and achieve better grades. "If we only care about efficiency, we will miss a huge opportunity. The ultimate goal of AI should be to help students fall in love with learning," insists Isabella Hao.

At this edition of WISE, Isabella Hao and her team at the "Stanford Learning Accelerator" garnered significant attention—In February this year, the accelerator launched the "Generative AI for Future Learning" seed fund, which has already amassed $625,000 to support research by 16 project teams on how generative AI can advance classroom teaching. The projects aim to design tools based on real teacher needs, validate their effectiveness through methods such as controlled experiments, and continuously adjust based on teacher feedback to ensure that the tools truly serve "the purpose of education."

Some research attempts are quite innovative, such as whether AI can be used to promote human-to-human interaction in the classroom. "The human brain is a social organ; we learn through coexistence with others, by imitating and listening to them. There is already ample evidence proving the impact of social connections on learning effectiveness. However, most educational applications based on ChatGPT are still limited to one-on-one interactions between humans and machines," explains Isabella Hao. One of the 16 projects funded by the accelerator aims to explore the potential for generative AI to support human-to-human dialogue and foster a more personalized collaborative experience. Interestingly, researchers involved in the project have also launched an autumn course titled "Designing for Learning: Generative AI for Collaborative Learning," which is open to both undergraduate and graduate students without any prerequisites, inviting them to participate in brainstorming sessions on AI innovation in education.

Another research project named M-Powering Teachers supports providing teachers with timely curriculum feedback in K12 classrooms. This tool displays the ratio of teacher/student talk, analyzes which questions students respond to most enthusiastically, and suggests to teachers how to use more positive, supportive language in conversations with students. "In traditional classrooms, some students speak less while others speak more, and teachers are unable to fully engage in group learning. We hope to use artificial intelligence to enhance teaching skills and student engagement, ultimately fostering student learning," said Isabella Hao.

In addition to promoting interpersonal collaboration, the Stanford Accelerated Learning Lab is also particularly focused on the potential of AI in fostering inclusive learning. One team is attempting to build an AI system that generates descriptions for data visualizations, helping visually impaired students better engage in STEM education; another team has developed a Rapid Online Assessment of Reading (ROAR) tool, designed to identify and intervene early with children who have reading disabilities. Furthermore, one team is focusing on "neurodiversity," trying to use generative AI to enhance the learning experiences of children with autism, cognitive impairments, ADHD, and more. There are also teams concentrating on "linguistic diversity," looking into how native speakers of non-dominant languages interact with large language models. "How to use AI to build more inclusive classrooms is also a direction we are very interested in," said Isabella Hao.

A phenomenon often mentioned by teachers is the friction that occurs between education and technology in real classrooms. "Some products have good concepts and look fancy in design, but either teachers don't have the conditions to use them, or they find them not so user-friendly. And when schools really want to undertake some reform exploration, they often can't find tools that match their needs," a high school principal once described to Caixin.

Ben Nelson, founder of the Minerva Project, argues that what matters is not the technological solution itself, but the curriculum and teaching solutions supported by technology. As an innovative school without a physical campus or classrooms, offering courses entirely online, Minerva University has developed its own learning platform. During online classes, the platform records the duration of each student's participation and uses different colors to remind professors to call on students who speak less frequently; it also supports real-time voting, document collaboration, instant feedback, and automatically tracks student tardiness and attendance. "At Minerva University, this system ensures that students are fully engaged in class; during the COVID-19 pandemic, some traditional universities began trying our platform as well. However, they did not use our curriculum and teaching methods which resulted in disastrous outcomes—not because the technology didn't work but because the quality of their courses was poor," says Nelson. "Before technological transformation can occur, there must be changes in teaching practices, curricula, and assessment first."

This tension leads educators back to the fundamental logic of technology application: In the age of AI, what exactly should we be teaching for?

The Reshaping of Education

In March this year, Zhang Zhi, the director of the Baoshan District Education Bureau in Shanghai, summarized the comprehensive impact brought by GPT at a salon titled "ChatGPT and Future Education" with concerns about knowledge obsolescence, ineffective educational tools, and a confused capability model: "We do not know what kind of talents we need to cultivate for the future or what core competencies they should possess. Will the abilities we have valued in the past such as listening, speaking, reading, writing, reasoning, calculating, and imagining still be useful? Will these ability models be less important than before?"

At WISE, "What to Teach in the Age of AI" remains one of the hottest and most divisive topics among educators from various countries.

Conrad Wolfram, the chief designer of the renowned mathematical software Mathematica, can be described as a "radical" on this issue. In his speeches and interviews, he asserts that "80% of what is taught in mathematics is wrong," fiercely criticizing that traditional math curricula are severely disconnected from practical applications and needs. "What we're doing in education now is teaching people how to do manual calculations—things that computers have long been able to do—rather than learning how to solve problems at a high level. Why do we pretend as if computers don't exist when teaching math?" He calls for a shift in math curricula away from focusing primarily on arithmetic and towards fostering "computational thinking" as the core objective. Computational thinking refers to enhancing the combined capabilities of students and computers to tackle real-world problems and challenges. "Learners may not need to know every step required to solve a quadratic equation, but they should understand what a quadratic equation is, how to verify results, and most importantly—when and why an equation needs to be established."

"In the 1950s, if you attended a private school in Britain, Latin occupied most of your academic schedule. But rapidly, Latin receded from the core curriculum, and mastery of it was no longer seen as an essential educational skill," Wolfram illustrated. "Mathematics could face a similar fate if taken lightly." He envisions a new math curriculum that resembles project-based inquiry where students go through four steps: defining the problem, abstracting it into a computable form, calculating the answer, and interpreting the results to tackle real and complex issues such as "Should we shut down nuclear power plants?" or "What are the impacts of pandemic lockdowns on health and economy?" "Asking the right questions," he suggests, "will be much more important than computation."

Clearly, not all educators are in complete agreement with such a disruptive proposition. Dr. Ng Pak Tee from the National Institute of Education at Nanyang Technological University in Singapore argues that "fundamentals" will still be important in the future. "Although some fundamental changes are about to occur, I do not believe that a solid foundation of basic knowledge will become obsolete anytime soon. An airplane with autonomous piloting can make navigational decisions using AI. But if the computer malfunctions, passengers would want a pilot on board who can manually operate and handle emergencies. Wouldn't such a pilot need to possess all the basic skills and have extensive flying experience?"

Professor Huang Bozhi, who teaches leadership courses at his school, has trained approximately 75% of the current primary, secondary, and high school principals in Singapore to date. He told Caixin that instead of directly answering "what to teach for the future," he prefers to pose an open-ended question to educators: If "artificial intelligence" is artificially produced (artificial), then what constitutes real human intelligence (real intelligence)? "Education should enhance the learner's true human intelligence. The challenge for educators is to figure out how to effectively use AI so that learners can better utilize their real intelligence," he said.

In the view of Huang Bozhi, educators must also possess the ability to cope with "paradigm shifts." "Previously, teachers assigned topics and students spent weeks writing essays. Now, students can generate a paper with a click of an AI button. Can the method of assigning homework change? For example, could we have students act as teachers to grade papers generated by AI?" Faced with new changes brought about by AI, educators should be able to continuously reflect on and modify course content, teaching methods, and assessment approaches in order to better support and motivate student learning.

Nelson believes that the purpose of education, particularly the core purpose of higher education, is to teach students the ability to think systematically in the age of AI. "We need to teach students transferability, which is the ability to move knowledge from one domain to another and to view the world through various analytical perspectives and modes of thinking." Minerva University has established four cornerstone courses that explicitly teach habits of mind and foundational concepts under the four core competencies: critical thinking, creative thinking, effective communication, and effective interaction. In Nelson's view, these core competencies will become even more important in the future. "If universities continue to be limited to teaching compartmentalized, atomized knowledge, the disconnect between higher education and labor market demands will worsen in the AI era."

Several interviewees emphasized the importance of paying special attention to ethical issues in the application of AI in education. According to the aforementioned Global Education Monitoring Report, during the pandemic, out of 42 countries providing online education, 39 engaged in practices that threatened or violated children's rights; only 16% of these countries have laws explicitly protecting data privacy in education, and 29% have implemented related policies, mainly in Europe and North America. "Developers bear a significant responsibility to incorporate ethical considerations into the design and development of AI technology, including algorithm transparency, fairness of AI impacts, and actively preventing any form of discrimination while strictly protecting user privacy," said Fatima Nacada from the Department of Computer Science at Qatar University during WISE (World Innovation Summit for Education).

Hu Xuan also contributed to this article

This article is selected from the upcoming December 11th issue of Caixin Weekly, originally titled "Caixin Weekly | ChatGPT Forces a Rethink in Education">