The AI-powered Rethink of Education

Introduction

As a high school student, the chatter among my classes, and in the teacher’s lounge, is converging on one topic: Artificial Intelligence. Generative AI systems have disrupted all corners of my school: ChatGPT wrote my principal’s winter address, teachers use it to design assignments, and my classmates use it to complete them. As AI evolves, it’s high time to rethink the fundamentals of high school education.

What impact will AI have on education? Some, such as Khan Academy founder Sal Khan, envision a world where AI enables low-cost personal tutors, democratising access to high-quality education. Others are concerned about the prospect of bias, misinformation and cheating. Among my teachers and friends, most admit they don’t know.

Understandably so. Among experts, there is little consensus on decision-relevant questions, such as whether, given AI progress, any familiar jobs will remain. Friends who have been coding for many years are wondering whether their skills will still be an advantage in the labour market, or even in demand at all.

Regularly, when I am discussing AI and education, such as in this ITU panel on ChatGPT in the Classroom, I realised it is easy to feel helpless in terms of the knowledge and skills that are fundamentally necessary. GPT-4, OpenAI’s most recently released AI large language model, can perform to about the 90th percentile on a host of standardised tests, from the Law School Admission Test to AP Art History to the GRE.

Given this, it is perfectly understandable to assume that most knowledge or skills imparted at a high school level that can be assessed will be, in the near future, completely automated by AI. The education function in high school education (without considering other functions like structure, social development, and signalling to universities and employers) is to impart knowledge, facts and information, and skills, the ability to perform a task. What happens when that is no longer necessary?

A framework

I argue that the digital revolution, therefore, caused a differentiation between ‘redundant’, ‘accessible’ and ‘foundational’ knowledge, and propose that the AI revolution will herald a similar differentiation for skills.

• Redundant knowledge/skills refer to pieces of information or abilities that are no longer valuable or relevant for individuals and society. They typically include elements that are phased out or replaced due to technological advancements, changes in societal structures or requirements, or shifts in a particular field of study.
• Accessible knowledge/skills refer to information that can be readily retrieved or tasks that can be performed using modern tools. The term "accessible" signifies that while this knowledge or these skills are useful or even necessary in society, they don't need to be memorized or learned in depth by individuals.
• Foundational knowledge/skills refer to the essential knowledge and skills that are critical for an individual's deeper understanding and personal growth in a particular field. Despite the fact that modern tools will be able to perform many tasks relating to this knowledge, it is crucial for students to learn and master this knowledge.

To illustrate, consider the following examples.

This is simply a table of examples. These, like most things, are not neat categories, but spectra. For example, charitably, memorising the periodic table confers some benefit in generating an inherent understanding of the structure of the organisation of the elements. However, AI still shifts how each of these skills appears on the spectrum.

These categorisations are helpful in framing debates. In my parent’s generation, rote memorisation was common. My principal spoke about how he was taught to memorise the periodic table, while today, most modern examinations now provide them. However, most primary schools still encourage students to memorise their times tables, because it is still fundamental to have a fluent command of more advanced mathematical knowledge. In both of these examples, the internet could be used to retrieve the knowledge needed, however, one is not worth memorising, and the others, many argue, are. Explaining exactly why can be narrowed down to making the distinction between memorising the periodic table as “accessible” and times tables as “foundational.”And, as we watch AI evolve, particularly with the recent advancements in generative models that can write persuasively, craft photorealistic images, and even code competently, it's clear we're on the brink of a skills-rethink, much like we faced with the internet and knowledge.

What should we do?

In my experience, there have been significant amounts of attention to how we teach in the classroom. My teachers regularly use AI to design activities. Khan Academy has released a GPT-4 powered chatbot, Khanmigo for teaching. But we will need to fundamentally rethink what we teach in the classroom.

Deciding how we do this is difficult. Many claim that "the most important skill will be learning how to get AI to be useful to you, like prompt engineering and learning how to integrate AI into your workflow." It may be tempting to hastily mandate lessons on prompt-engineering, however, one reason this might be premature is that pretty quickly because we have AIs which are able to understand poorly-optimized prompts well and respond. There seem to be strong financial incentives for AI developers to solve this problem for us such that normal people can just use natural language.

The first step is understanding. We all know the teacher who insists or refuses to adopt digital and classroom technologies. Avoiding these advancements does a disservice to their students, who need this to be. School administrators, curriculum designers, and educators have a responsibility to meaningfully engage with cutting-edge technologies available today and stay informed about the potential developments of tomorrow. Educators, I urge you to explore tools like ChatGPT—not only to aid in tedious tasks such as individualised report comments but to understand the limits and potential of such technology in shaping education.

Take the initiative to explore the capabilities of these AI tools. Just recently, I demonstrated to a geography student friend how the Wolfram plugin on GPT-4 could conduct, explain, and visualise an independent t-test—a process that would have taken hours of manual work with spreadsheets. If you find yourself at a loss, engage your students in this exploration. It's highly likely some of them are already experimenting with AI tools in their own time.

A skills rethink is upon us. Educational institutions, like many other organizations, often struggle to adapt to unseen future challenges. Allocating scarce resources, leadership focus, and the valuable time of teachers towards anticipating and preparing for these changes is a daunting task. But it's a necessary one.

• Redundant Knowledge/Skills: Once identified, these aspects of the curriculum should be phased out. Education time and resources are finite, and dedicating them to outdated or irrelevant content makes little sense. Moreover, focusing on redundant information might hinder students' ability to learn more current and applicable knowledge and skills. For instance, learning how to navigate a physical library catalogue may consume time that could be better spent teaching students digital research skills, which are far more relevant today.
• Accessible Knowledge/Skills: This category requires a shift in teaching strategy. Instead of encouraging rote memorization or in-depth learning, educators should focus on teaching students how to access and use this type of knowledge and skills efficiently. For instance, coding could be considered an accessible skill due to the rise of AI-powered coding tools. It may lead to a shift from teaching specific programming languages to teaching foundational computational thinking skills, understanding of algorithms, and efficient use of AI coding tools.
• Foundational Knowledge/Skills: This category remains crucial for in-depth learning. Foundational knowledge and skills form the building blocks of advanced understanding and should be a central part of the curriculum. Here, the goal should be to ensure students gain a strong understanding, allowing them to apply these skills creatively and adaptively in various scenarios. For instance, instead of relying solely on AI tools for writing and communication, students should be taught the fundamental principles of effective communication, critical thinking skills, and how to construct logical and persuasive arguments.

The lessons here apply beyond the classroom, too. In any place where you are asking the question: “Should I allow the use of generative AI tools here?”, consider whether the skills and knowledge you are trying to impart are accessible or foundational. Even though I’ve been arguing for a rethink of the rules, you should still write out your policy explicitly – ambiguity in whether AI tools can be used or not incentivises those most willing (or able, in the case of some paid tools) to skirt the rules. For example, if you are running a high school essay contest, consider breaking down the skills at each stage of the academic essay-writing process into these categories. Maybe you’ll consider instigating a ban on the use of tools to write out essays, but otherwise require participants to declare the use if it has helped them with research or ideation. And even if you do not think believe in shaping the future through skills rethink in service of students and the education system, practical enforcement difficulties in preventing students from using these tools means that this responsibility is still one you’ll have to confront.

Finally, if you are a student (or curious outside observer), ask yourselves the same questions and try out the technology. Stay by the rules, yes, but also push the boundaries. In 1968, at the age of 13, Bill Gates was programming computers. The only way to avoid being swept up in the coming AI revolution of education will be to stay ahead of it.

Acknowledgements. Sincere thanks to Aaron Sher, Moritz Wallawitsch, Chet Katu, Nick Alchin, David Bacchoo, David Kann, Kelly Smith, James Camacho for feedback and suggestions. All mistakes are my own.