All stakeholders, from educators and parents, seem to be aligned on the objectives:

Ensure Quality Education and Skill Development: Every student, irrespective of their background, should have access to education that equips them with the necessary skills for the future.

Provide Accreditation and Certification: A streamlined system that certifies knowledge acquisition and allows transitions between curricula is essential.

Personalize the Educational Journey: Recognizing that every student is unique, a tailored educational experience is the ultimate goal. This ensures that learning is not just effective, but also engaging and relevant.

The Traditional Education Conundrum: A System Awaiting Disruption

Traditional education has grappled with challenges it often couldn’t surmount, chiefly the trade-off between mass education and limited resources. This system, as described in the “Grammar of School,” has been primarily organized around age-based cohorts, standardized curricula, and assessments, leading to a one-size-fits-all approach. However, the diverse needs of key stakeholders — be it governments, schools, parents, students, or publishers — make it hard to customize solutions for each learner. Generative AI presents an opportunity to redesign this system, considering the nuanced demands of each stakeholder, and ensuring a more holistic and adaptive educational experience for all involved. It can customize learning materials for individual students, predict areas where they might struggle and suggest corrective actions, and even help in automating administrative tasks, freeing educators to focus on what they do best — teach. So let’s take a deep dive into what is working on the market as of today and the lesson learned.

1. The Future of Gen-AI in Education is already in progress: panorama of initiatives happening globally:

Content Creation:

With LLMs (Language Learning Models) in play, AI’s role in educational content creation has evolved. From speedy creation to quality content, AI-generated material spans from simple exercises to comprehensive teaching materials with image, video, etc. For example, Mindshow, a California start-up, offers a Gen-AI based tool to generate powerpoints or slides for educators in just a few clicks. Another example mentioned by OpenAi is a professor at Universidade da Coruña in Spain, Fran Bellas, who is harnessing chatGPT for quizzes and lesson plans. But beyond individual educators, ed-tech giants like Kahnmingo and Doulingo now offer material generation services on their platforms. Crucially, possessing extensive content repositories, like publishers, brings an edge, as illustrated by ZuoYeBang, who leverages its vast repository of exercises and the responses from students since its inception in 2014 to train its LLM, ZYB-Galaxy.

Tutoring and Assistance:

Tutoring remains the dominant AI use case as of today embraced by startups. An AI tutor enhances engagement, facilitates question-driven learning, and even gamifies education. While ed-tech firms are jumping aboard the AI-tutoring bandwagon, refining answer quality is paramount. This is achieved primarily by feeding the AI with vast educational data. For example, Gaotu, a major online tutoring company in China, is experimenting with LLMs to make “duplicates’’ of its famous teachers; Henan Radio and Telecommunication University in China, has already an AI teacher “HeKaiKai”, though its performance is yet far from satisfactory as a primary source of teaching. A novel approach by the free app Sizzle AI in the USA involves a “step-by-step” methodology, boosting both student understanding and LLMs’ reasoning capabilities. It’s becoming typical for students to use AI-powered learning apps, like ChatGPT, to instantly get answers without learning anything. However, Sizzle doesn’t simply provide solutions to the problems. The app acts as a tutor chatbot, guiding the student through each step. Students can also ask the AI questions so they can better understand concepts. “Unlike humans, AI based systems like Sizzle can be infinitely patient, understanding, and personalized. They can interact with each learner 1 on 1 and answer as many questions or go in as much detail as they need. The intent is not to replace teachers or classrooms, but to provide to all learners the kind of assistance that only the most privileged had access to in the past.” said Jerome Pesenti, founder of Sizzle and ex VP of AI at Meta. Similarly, Khan Academy’s Khanmigo aids students during live classes, and Haoweilai’s MathGPT showcases strength in the tricky realm of mathematics.

Sizzle

Assessment and Feedback:

Automating assessment alleviates educators from the tedium of grading or providing immediate feedback to the students. AI can provide explanatory feedback, ensuring students grasp their mistakes. Duolingo Max, powered by GPT4, excels at this, bridging human-like conversation with feedback and the App developer has priced it at $168 a year, nearly 60% higher price than its Super Duolinguo $108 yearly package. In China, iFlyTech has employed AI to evaluate exam essays since 2015, which is a significant revenue source for the company, and it has embarked on a journey with its LLM, Xinghuo, to offer advanced, personalized feedback.

https://blog.duolingo.com/duolingo-max/

Personalized Learning:

Envision an educational landscape tailored for each student. While this remains an aspiration, AI has the potential to rewrite the “grammar of school”. NewClassrooms.org from New York, USA pioneers this frontier, offering a comprehensive AI-powered educational package, truly tailored for each student. In this program, students from the same “class” do not learn at the same speed, neither with the same teaching materials nor evaluated with standardized exams. Of course, to make it real, we need both powerful AI and innovative adoption to the existing educational system. Europe’s iRead initiative in 2020 heralds this evolution, delivering a bespoke AI-infused learning suite, intricately catering to each learner. Here, students within a cohort progress at their unique pace, with distinct educational resources and non-uniform assessments. The transition from aspiration to reality hinges on robust AI and its creative integration into the prevailing education framework. In Finland, Claned Group melds AI with Finnish educational expertise to offer a personalized online learning platform. This platform devises individual learning paths and promotes social interaction among students, which keeps them engaged. The machine learning system gradually comprehends each student’s learning behavior, giving tailored recommendations for study materials and alternative learning paths, thus embodying the personalized learning approach.

Administrative Tasks:

In the backdrop, AI simplifies school administration, tackling tasks from admissions to grading. Azure OpenAI, aims to support customized learning and can be employed for new reading assignments, showcasing an effort to aid in student administration and grading through AI technology. Also, some existing ed-tech specialists like Turnitin are trying to leverage AI to better assess plagiarism but as the LLM can both detect fake but also produce content it is creating some challenges. Tech behemoths like Microsoft and Google have started to acknowledge that they have to address the issues of plagiarism “Think of the fight against AI-assisted cheating like a battle between Sherlock Holmes and Professor James Moriarty, trying to outsmart each other”. Universities, such as Tsinghua in China, have developed tools to detect if the text is generated by AI or by students. Whereas these tools need to be updated frequently so as to catch up with the fast progress of new LLMs. Haoweilai, a Chinese edtech company, has developed a teacher evaluation system based on students’ participation in class and results of exercises/exams.

2. Takeaways from the first successful players in Generative AI EdTech:

1. Focus on Genuine Pain Points of users:

Successful GenAI ed-tech initiative focuses on identifying and addressing user needs. For example: all over the world parents search for a suitable tutor for their child because research consistently showcases the positive influence of mentors on student outcomes, but scarcity and costs of qualified tutors often restrict access. This is where Khan Academy’s Khanmigo created a difference. In a pioneering move, Khan Academy collaborated with OpenAI and integrated the prowess of the GPT-4 model to furnish students with personalized AI mentoring. Khanmigo’s ingenious solution doesn’t just offer generic assistance to questions; it engages students by adapting to the style of the learner and proposes discussion with virtual mentorship from AI renditions of iconic figures’ insights. This innovative approach not only elevates the educational journey but also boosts user adoption by offering a blend of personalized learning and illustrious guidance. The adoption has been emulated by others such as Brainly — Ginny, A Polish AI-powered learning platform, that combined GPT-4 with their accumulated knowledge base of educational resources, where Learners can ask for simplified answers to their questions or opt for more in-depth solutions that explore complex topics. They can also ask follow-up questions for more information, clarification of a response or a step-by-step solution to a problem. All these approaches have in common that they speak to the pain of the user by proposing an adaptive solution.

2. Harness Available, Relevant Content:

Successful GenAI ventures don’t stray far from the established curriculum, quite on the contrary they leverage the very granular local curriculums. For instance, in the US, Amplify, is an ed-tech company founded in 2000 that provides digital curriculum and assessment tools for K-12 schools. Amplify not only taps into language learning but also aligns its content closely with national and regional linguistic syllabi. Their products are designed to help teachers personalize instruction and improve student outcomes. Amplify’s curriculum is aligned with state and national standards, ensuring that it is relevant and applicable to the established curriculum of public schools. This ensures AI-generated content is not just impressive but immediately relevant and applicable. Many other startups are following steps, such as Curriculum Associates an ed-tech company that offers regional based programs in reading, language, and math, along with assessment tools. ​This in turns facilitate adoption by local schools and local teachers.

3. Continual Model Refinement:

The elite in the edtech space know the worth of consistent AI improvement. In China, Haoweilai’s MathGPT stands out here, even if it has not yet attained perfection. By gathering continuous feedback from users, they’ve fostered a sustainable data in-flow, enhancing the model’s capabilities while simultaneously erecting a barrier for potential competitors. In the USA, Dreambox Learning, founded in 2004 is an adaptive learning platform that provides personalized math instruction for K-8 students. In 2016, a study from Harvard University revealed mixed yet encouraging results regarding DreamBox Learning’s impact on student achievement. Key findings include greater achievement gains for students who spent more time on the software or followed its lesson recommendations. However, most students didn’t reach the recommended usage levels, and the variation in usage was mainly due to teacher- and school-level practices rather than student preferences. Based on those findings and the power of Model refinement, Dreambox Learning is better equipped to use continual model refinement to provide better engagement.

4. Associate with Public Authorities:

Initiating collaborations with educational regulators is crucial when delivering the national curriculum. For instance, in the US, Carnegie Learning was selected by the Connecticut State Department of Education for High-Dosage Tutoring, integrating directly into classroom activities. This early endorsement from educational authorities significantly enhanced their credibility, streamlined customer engagement, and hastened adoption. Similarly, in South Korea, the government’s proactive involvement in ed-tech has fostered a conducive environment for such integrations. Their National e-Learning Strategy has propelled initiatives like EBS Online Class, where digital learning materials align with the national curriculum, promoting a seamless fusion between traditional and digital educational frameworks. Through these early engagements with public authorities, both Carnegie Learning and Korean ed-tech initiatives have managed to align with national education goals efficiently, accelerating their adoption and impact.

5. Prioritize Customer Acquisition via Schools:

Behind every successful GenAI edtech business is a meticulously crafted acquisition strategy. Many startups assume that the customer acquisition cost can be high because the lifetime value of the customers is potentially very long over the course of learning a new skill. However, multiple examples have shown that customer engagement is rather low, unless schools make edtech mandatory. Hence the need to have the school recommendation to limit Cost of Acquisition or to carefully balance the costs of acquisition with an average LTV of six months. iFlyTech in China provides whole packages for schools and governments, which includes: exam essay evaluation for schools, teaching software with AI material generation (incl. Images, video & games), student learning software with exercise and AI feedbacks, digital whiteboard in classroom, which directly recognises and converts teacher’s writing/drawing into the teaching software, etc. This approach has turned a technological solution into a ubiquitous educational tool and maximized the company’s return on investment.

3. The Roadblocks to Wider GenAI Adoption in EdTech: Building an Effective EdAI Enterprise:

1. Imperative of Flawless Delivery on tasks that are otherwise challenging:

In the podcast Freakonomics, Dr. Mitchel Resnick (Professor of Learning Research at the MIT Media Lab) stated “Schools end up focusing on the things that are most easily assessed, rather than focusing on the things that are most valuable for kids and valuable for thriving in today’s society. So what we need to do is to focus more on trying to assess the things we value rather than valuing the things that are most easily assessed.” For this precise reason, GenAI and edtech present an opportunity to emphasize often-overlooked soft skills, offer deeper insights into student thought processes, and integrate real-world contexts into learning, enhancing relevance and engagement for students. In a research paper “Realizing the promise: How can education technology improve learning for all?’’ The Center for Universal Education reviewed ed-tech interventions from 37 studies across 20 countries, emphasizing the importance of accuracy and technology’s potential in enhancing education quality, especially in hard-to-staff schools or regions with frequent teacher absences. It discusses the positive impact of pre recorded lessons on student learning, yet also mentions failures where such interventions didn’t improve or negatively impacted learning outcomes because of its lack of accuracy on focusing on the task that are otherwise “challenging”.

2. Recognizing the Significance of Hardware:

While software or SaaS solutions are transformative, hardware remains central to initiate revenue and promote adhesion, particularly in places like China or underdeveloped regions. Parents often exhibit reluctance towards paying for services that don’t offer tangible products. Public education is also lure into digitalizing the classroom with Digital Boards, Cameras, and cloud systems. Huawei Digital Education initiatives already implemented in China and around developing countries, is using hardware to facilitate adoption. Additionally whilst trying to be implemented in regular settings, many classrooms across the globe have stringent policies against smartphones and tablets, yet still allow traditional tools like pens. Integrating AI capabilities into these tools is often a game-changer. Products like the Neopen, Youdao Dictionary Pen, which combine traditional writing with digital capabilities, offer a glimpse into this fusion. The pen enables translation & exercises. In light of these successses, it’s evident that while GenAI advancements are crucial, the tangible impact of hardware in edtech cannot be understated, serving as a bridge between traditional methods and the digital future of education.

3. Defining a Unique Value Proposition:

With the rapid progress of Open AI ChatGPT, Antropic Claude, Perplexity, Google Bard and MIcrosoft Bing, it’s essential for new entrants in the EdAI space to clarify their unique selling points. With the capacity of the Institution to define very powerful models, even using ‘only’ 7 billions parameters using OpenSource such as those hosted on HuggingFace, it is essential to differentiate. Differentiation is not necessarily in the information managed nor its volume of parameters. It can be the user interface, specialized curriculum alignment, or the integration of images or experiences via animation. Companies need to articulate what sets their product apart in a market teeming with Gen AI- socratic conversational agent (ChatGPT4 alike). For example, the main page of MathGPT from Haoweilai, is directly relevant to parents and students because its topics are arranged using the Chinese education curriculum. Similarly Khanmingo and DuoLingo Max are keeping records of previous conversations with students to depen the mentorship of the student.

4. Understanding the Shift in Revenue Expectations:

While many businesses anchor their expectations in the Lifetime Value (LTV) of a customer, the EdAI industry is a different playing field. On the face of it, one could imagine a very long LTV. However, with academic cycles and swift tech evolution, a six-month Return On Investment (ROI) is proving a more reasonable metric. The initial prospespect of long LTV approach has created an Ed Tech bubble in China in 2020–2021 and pushed the acquisition cost up to 200 USD/user, which was unsustainable, since many customers change the platform after the first payment cycle (1 semester). In the USA, the findings are not different, the NASDAQ-listed Online Program Management (OPM) firm from the US, 2U, reported a loss of $43 million (EBITDA) against a revenue of $410 million, despite being operational for 11 years. Similarly, Chegg, established in 2005, incurred a loss of $7 million (with a revenue of $320 million) in 2018, and it only turned profitable as of the second quarter of 2023, with a reported net income for Q2 of $24.6 million.This focus on ROI planning for a short term LTV ensures products are designed with immediate effectiveness in mind, aligning with the rapid cadence of academic terms and customer ability to change provider if the provider is not mandatorily imposed by the school or learning program.

5. The Age of the AI-Powered Personal Assistant in Education:

As we look to the horizon of Edtech, the rapid progress of GenAI points to personal assistants elevated by function calling and multimodal capabilities. These abilities will be available to all just using general search / nowedge tools such as ChatGPT, Claude, Poe, Bart, You or Perplexities. The new progresses are currently centered around two main areas of development:

Function calling, in the context of AI, refers to the ability of the AI system to execute specific tasks or operations in response to a user’s input. Instead of just providing information or answering questions, an AI with function-calling capabilities can actively perform tasks. A notable example in EdTech is at Georgia Tech, where an AI assistant named Jill Watson helped manage a master’s-level AI class’ online message board, handling repeated queries with a 97% success rate. This AI system, capable of parsing query context and responding accurately, illustrates function calling by autonomously handling specific tasks, easing the load on human teaching assistants​.

Multimodality in AI refers to the integration of multiple types of input and output modes in an AI system. This can mean the combination of text, voice, images, videos, and other sensory data. For instance, a student could show the AI a picture of a historical artifact, and the AI could provide information about it verbally, supplemented by visual aids and even interactive 3D models. This isn’t just about answering questions but also executing tasks and integrating various functionalities. Imagine a tool that doesn’t just provide information but can also schedule study sessions, curate relevant study materials based on a student’s progress, or even initiate multimedia educational experiences.

AI in education isn’t just about answering questions. It’s about AI systems that can actively assist in the learning process by performing tasks and offering diverse, rich, and interactive learning experiences that cater to the varied needs and styles of students. This makes learning more tailored, efficient, and engaging, fundamentally transforming the EdTech landscape.

Conclusion: Unlocking the Future: Bridging Generative AI with Modern Education

In a world where technological advancements often outpace our ability to fully grasp their implications, the education sector finds itself at a crossroads. The leverage of Generative AI by the educational frameworks heralds a new era, one that promises to redefine the traditional “conveyor belt” model of education that has endured since the 19th century. The maturation of GenAI now offers solutions to old educational limits to facilitate the endeavors of educators, parents, schools, and public authorities, aiming to pivot from a one-size-fits-all approach to a more tailored, effective learning experience.

GenAI Edtech is not restricted by technological limitations but rather by the current organizational structure of education systems. These systems, ingrained with “rites of passage” markers such as exams and accreditations, find it challenging to disrupt the status quo. The myriad of authorities involved, coupled with the quest for a sustainable Long Term Value (LTV) of solutions within organizations, add layers of complexity to this transformation.

On the global stage, different nations exhibit varying degrees of adaptability and strategy in integrating GenAI within their educational landscapes. Western countries often delegate the choice of GenAI EdTech partners to regional schools and parents, fostering a culture of individual choice. On the contrary, nations like China, Finland, and South Korea exhibit a more centralized approach, albeit with distinct methodologies.

China and other developing nations often couple EdTech hardware with GenAI solutions, a strategic move to offset infrastructure challenges and elevate school systems’ capabilities. This dual-pronged approach provides a perfect setting for a digital transformation, yet the system remains the same for now, based on the national curriculum for all.

Meanwhile, Finland and South Korea are reshaping their educational organizations to offer personalized curriculums and learning paths for students. This approach, more systemic in nature, aims to align the educational framework with the personalized capabilities of GenAI, ensuring a more holistic and adaptive learning experience.

Generative AI, despite its nascent stage, is on a trajectory that will soon provide generalized access to sophisticated and personalized agents for all. This progression necessitates that EdTech solutions orient themselves towards adapting to local educational systems before revolutionizing them.

As GenAI capabilities and education alliance, the road ahead is offering a potential for a positive disruption yet is paved with challenges. The ability to navigate these challenges, adapt to local educational landscapes, and continuously evolve with the rapid advancements in AI technology will delineate the trailblazers from the bystanders in this journey. The fusion of GenAI with education is not merely a trend; it’s an inevitable transformation, a cornerstone for a changed society where quality education is accessible to all, irrespective of their geographic or economic standing. Through strategic partnerships, continuous innovation, and a nuanced understanding of the complex educational ecosystem, GenAI stands to redefine the essence of learning in the modern era.