Embracing AI: Navigating the Future of Education in SY24-25 and Beyond

Ever since the release of ChatGPT in November 2022, discussions have emerged about its transformative impact on education, from short-term methodological implications to long-term structural and conceptual influences. As one of the first users of ChatGPT and a tech veteran / educator, I have been contemplating this topic since December 2022. Some of my early experiments involved using ChatGPT for curriculum mapping between the Chinese and English curricula, a task allegedly took a team of experts nearly two years to complete. I also explored using ChatGPT to create lesson plans for innovative learning topics using Astra Nova School resources while I was working at AISL Harrow. Even with GPT-3.5, the Large Language Model (LLM) provided a reasonably satisfactory results for these tasks (obviously I'm not a teaching expert, so my bar is pretty low).

As educators and technologists grapple with the implications of this rapidly evolving tool, the questions have evolved from "How will AI shape the future of education?" to "What actions can we take now as schools and educators?" While the initial mixed reactions to ChatGPT have gradually shifted towards acceptance and integration, the journey is still unfolding, revealing both challenges and opportunities. As we approach a new school year, I’d like to offer some humble predictions and suggestions for my friends and ex-colleagues in education, with the hope that AI will bring you success.

The Fuzzy Old Days

Shortly after ChatGPT gained attention, universities and schools in the United States expressed mixed feelings about this advanced new technology. ChatGPT, a live encyclopedia where knowledge can be accessed through simple chat, quickly became a popular tool among students for cheating. This led some schools to hastily ban its use on campus. However, as time passed, school administrators realized that they couldn't indefinitely ban this technology and had to teach students how to use it responsibly.

Kevin Roose, a columnist for New York Times, wrote in January 2023 that ChatGPT could enhance students' learning experiences by providing personalized writing assistance, fostering creativity, and encouraging critical thinking. The key, he emphasized, was to teach students to use AI responsibly, understand its limitations, and consider its ethical implications. Schools should adapt to these changes rather than resist them, ensuring that students are prepared for a future where AI is ubiquitous. Although an early opinion, his perspective remains relevant today, as the fundamental challenges and opportunities surrounding AI in education continue to persist.

When Sal Khan introduced Khanmigo, a GPT-supported learning assistant, in March 2023, he not only demonstrated the educational potential of AI but also showcased how the education industry might be enhanced in the future. Khanmigo was the first AI chatbot fine-tuned to a level where it could responsibly enhance teaching and learning, rather than simply providing immediate answers. This development highlighted the potential of GenAI (Generative AI), showing that, when properly positioned and within the right scenarios, it can be effectively integrated into educational practice to yield plausible results. However, its implications for educational equity are even more profound—though that is a discussion for another time.

Sal Khan has long been one of my idols. My first academic paper in 2011 was a case study of Khan Academy and the impact of MOOCs (Massive Open Online Courses) on language learning. Twelve years later, he remains a leading figure in innovative education.

As more schools recognize that banning GPT tools is not a sustainable solution, many have lifted these restrictions following New York City's first initiative in May 2023, opting instead to implement guidelines for their responsible use. For example, some schools now require students to disclose when they’ve used AI tools like GPT in their assignments, ensuring transparency and encouraging ethical practices. Schools like those under the guidance of UCLA have adopted policies where students must document their interactions with AI tools, explaining how and why they were used.

Meanwhile, tech startups are rapidly developing GenAI tools tailored to meet the unique needs of educational institutions. Several GenAI powered education platforms emerged early. Based on various foundation models, platforms like MagicSchool and SchoolAI allow teachers to create AI-enhanced web-based learning environments, seamlessly integrating AI into the educational process. Squirrel AI, previously a leading youth coding training company in China, also launched products in this direction. However, the effectiveness and long-term impact of these tools remain uncertain, as schools continue to navigate this evolving landscape.

What Will School Year 24-25 Look Like?

As we approach September of 2024, with multi-modality, open-source models, and smaller-parameter-LLM dominating headlines in GenAI, schools worldwide are welcoming a new school year that will undoubtedly be influenced by GenAI in deeper and broader ways. I don't normally speculate but here are some of my predictions:

For one, the use of AI will be rampant. Many more teachers will start using AI tools to plan lessons and write up student reports. Students, especially those in upper school and college, will further embrace AI. If they were impressed by GPT-4's capabilities last year in automating coursework, this year they will likely enjoy GPT-4's multi-modal functionality even more. Taking a picture of a question and receiving an immediate answer will become a new norm and the students will also learn about GenAI hallucinations in a hard way. This will compel teachers and schools to develop more effective protocols and countermeasures to ensure learning and to detect AI generated contents, especially in universities.

Innovation with AI will emerge. Leading schools and visionary school leaders are likely to initiate pilot programs to integrate AI tools into their school's digital architecture, and some will experiment with AI courses led by their more forward-thinking teachers. However, many schools are more likely to purchase AI-related programs from third-party providers, the same ones who managed last year’s spring tour or fencing programs. These programs will proliferate, making it difficult for frontline teachers and mid-level managers to effectively distinguish between them. Despite the often mediocre quality and limited educational value of these offerings, you will now have AI in your schools.

In addition to these programs, everything will now "have" AI in it and it might cost more. SaaS providers who supplies the school's learning system and are almost certainly working on their AI platforms or functions. Although many current AI tool developers for schools may not fully understand AI or education, this will pave the way for a more tested and effective approach in the future.

In summary, the coming school year will present complex challenges for school leaders regarding AI. The sudden influx of early-stage products on the market may make adoption seem too late, yet hastily integrating these technologies could lead to more complications and operational uncertainties. While this period may feel chaotic, from a broader perspective, these initial efforts will ultimately lay the groundwork for a clearer, more stable integration of AI in Education in the future.

Suggestions for School Leaders

After joining AIRS and spending almost four months researching materials, talking to experts, and preparing content for the launch of INNOVED, our new AI-centric education platform, my view towards GenAI has become somewhat more rationalized. For me, the technology has shifted from a magic wand of limitless possibilities to a probability algorithm yielding seemingly intelligent answers that can drastically accelerate domain efficiency when applied properly. While Generative AI holds great potential for business, caution is needed in the field of education.

Our mission is to create programs that support the development of AI-native future talents. However, the term "AI-native" is rather complex, and the journey toward it will require extensive research and experimentation. Therefore, with SY24-25 in sight, I'd like to offer some suggestions regarding AI strategies in your schools, based on my evolved understanding of AI from my perspective both as an AI practitioner and a ex-school manager/educator:

• Train Your Management Team on GenAI: Despite some school leaders and administrators having experience with GenAI tools like ChatGPT or Gemini, comprehensive training is necessary to grasp how GenAI works and its capability boundaries. Understanding AI and GenAI mechanisms will help develop structural thinking for systematic adoption and future upgrades. You don't need a customized GenAI tool for schools at this stage; there will be many deep learning-based SaaS solutions that can significantly improve school operations. Knowing how AI can help is vital for long-term decision-making.

• A School AI Champion is All You Need (for now): Borrowing from the title of the 2017 Google Brain paper Attention Is All You Need, for successful AI adoption at the school level, you need a school AI champion. In any organization aiming to responsibly adopt AI, the best practice is to nominate a champion who understands business details and organizational challenges and actively learns about AI to a semi-professional level. This champion will then be able to design an AI adoption roadmap that suits your school. GenAI is not magic; it is essentially a general-purpose software that will eventually be used for specific school scenarios and needs.

• Streamline Your School Data: As within any LLM training process, preparing and preprocessing data is the first step, the same is true with your school AI adoption. School data should be at the center of your AI upgrade. While schools today have their own data systems, be it PowerSchool, Engage, or iSAMs, the data is often not correctly stored or fully utilized. Preprocessing student data involves designing a common language and data processing architecture to allow the data to be fully utilized by AI when implemented. This requires schools to reconsider what data they want to keep, the relationships between categories, and to provide digital data entry points throughout the school while cultivating correct data management practices among teachers and students.

• Be Cautious About AI Spending: Avoid purchasing any so-called AI hardware (except for robotics). GPUs and TPUs (Tensor Processing Units) are unnecessary for schools, even school groups, as they lack the capability to train localized models, and most AI tools today are cloud-based. School computer labs upgraded after 2017 should be capable of accessing any AI tools today. Also, be cautious about AI-powered school software. Without an AI architecture and data system in place, these tools are unlikely to function effectively across the board. RAGed (Retrieval Augmented Generation) AI tools are just as effective as your customized GPTs.

• AI Will Not Help You with Admissions (Yet): For private/independent schools, income is often a primary concern, leading to questions about whether AI can assist with school growth. Admissions are highly personalized, and AI cannot fully replace the human element. Key decision-making stages within an admissions pipeline rely heavily on human interactions, which AI can only facilitate, not replace. That said, AI can significantly enhance the quality and efficiency of marketing content production. If your school takes a branding approach to admissions, having AI-literate admissions and marketing teams could provide a competitive advantage.

• Curriculum Innovation? Start with PBL First: Many researchers at leading institutions such as MIT RAISE and Stanford HAI are working on how AI might impact curriculum changes in the long run (this two has also been my main source of readings other than Arvix). While schools are ideal testing grounds for such innovation, I do advise caution at this stage. Using a CNN (Convolutional Neural Network) metaphor, while we are still at the "fuzzy front end" of this innovation, it's better to first understand the broader picture without being bogged down by noises. As technology advances, it will serve as a refined filter that clarifies details step by step. I suggest that the first trial with AI in your curriculum should be faculty-led PBL (Project-Based Learning) projects, focusing on interactive learning topics that could be supported by AI.

• Fully Utilize AI to Facilitate Home-School Communication: For international schools, AI can immediately address the challenge of home-school communication. LLMs (Large Language Models) fall under a subcategory of NLP (Natural Language Processing), where machine translation is a predominant application. One of the key challenges in international schools worldwide is effective communication between English-speaking teachers/staff and non-English-speaking parents. LLMs not only can improve communication quality and efficiency, but also help create culturally sensitive content that connects the two groups. Therefore, this should be an obvious first step.

Conclusion

Having worked in multiple school settings and observed the broader picture from a business perspective at the educational group level, I’ve synthesized my recent experiences and insights of AI into the suggestions above. As I delve deeper into the technical aspects of AI, my initial AI FOMO has evolved into a more balanced and thoughtful perspective.

While AGI (Artificial General Intelligence) is still years away, as a field deeply intertwined with both moral and labor considerations, education will undergo significant changes in the coming years. Core subjects, particularly math and language, will remain essential, perhaps even more so. While dynamic changes in curriculum content may be unlikely over the next decade, methodological and pedagogical upgrades will be inevitable as teachers and school leaders become more familiar with AI concepts and practices.

A recent study published by Oxford highlights that as we exhaust authentic human-generated training data, LLMs might become "stupider" if trained on AI-generated content. Although children's minds don’t operate exactly like deep learning neural networks, this raises concerns about the potential impact on human cognitive development. Therefore, proper use instruction and structural control should be on every educators’ minds (rather than to ban it) when it comes to putting AI into education.

For school leaders, my advice is clear: invest time in understanding AI, leverage its efficiencies, prepare for long-term integration, establish clear guidelines, and stay mindful of its influence. After all, AI is intended to enhance human intelligence, not diminish it.