The origin of Interleaf, and our vision of how artificial intelligence can be integrated into academic education in the near future.
Introduction - Education must evolved to be tailored and adaptive
The journey of education is as diverse as the human mind. As a Cambridge graduate and nearly a decade of tutoring experience, I’ve personally witnessed and experienced a spectrum of learning styles and capabilities - from ones who seem to grasp concepts effortlessly with almost photographic memory to those who struggle despite their best efforts. This variance isn’t merely innate, rather it’s a reflection of the intricate tapestry of human cognition, highlighting the importance of nurture in education. Here, one might expect the cliché mention of the stagnation of the current education system; yet, quite the contrary, education has evolved and improved significantly from an objective perspective. To a large extent, especially, its accessibility. The Internet has revolutionised access to information, making knowledge not only more readily available but also enabling its dissemination at an unprecedented speed. Nevertheless, education still seems outdated and inefficient. Perhaps the problem is much more fundamental and convoluted, where the solution does not lie in continually expanding the breadth of knowledge and its reach. Instead of weaving a wider tapestry, we should focus on each individual thread.
One reason I suppose that gives the perception of a sluggish progress in education could simply be the accelerated pace of virtually everything else in comparison. It is like a steam engine in the world of bullet trains and ever changing terrains; the solution isn’t to add more engines, but to teach its passengers to build a motorbike while equipping them with a navigation system, enabling them to chart personalised, agile routes, at their own pace. Especially with how the value of knowledge is changing, education is no longer merely about hoarding information, but to cultivate curiosity, critical thinking, and the ability to learn and adapt. Its experience should parallel those enlightening late-night chats with your wisest friend who somehow understands you in a very personal way – leaving you invigorated and inspired each time. This quality of personalised education should and must, therefore, be small, tailored, and adaptive.
The problem - Quality education is just not so scalable...
The problem with such a style of education is that it is rare and hardly scalable, which I have personally experienced myself. Since my days of tutoring my high school roommate, I have developed a talent with my naturally strong intuition that not only helps me pinpoint where students stumble, but often deducing their underlying logic and thought process that led them astray. This was subconsciously done through picking up subtle hints from their language and word of choice, while meticulously analysing their errors and engaging in probing conversations. The results were evident, at least in this current domain of discussion about academic education, with my students going into top universities worldwide such as Harvard, Oxford and of course my own alma mater, Cambridge. Yet, lurking in the background of my mind was a challenge I had been hesitant to confront, scalability. For private tutors like myself, or even school teachers, maintaining such a high level of personalised instruction realistically means teaching classes of only one to three students. Even if one were to teach just one subject and hold three classes a week, one would reach the capacity with just 30 unique students each year. The inevitable consequence of this is a tier of exclusive and expensive academic education services. I soon realised that to address this issue, revolutionary changes needed to be made, perhaps with technology, I suppose.
At around the same time, Oscar, my life-long friend from middle school saw a different yet related problem. He was doing his Masters degree in Cognitive Psychology and considering doing a PhD on educational psychology, when he realised a lot of evidence-based cognitive learning theories, despite being well proven, were often ignored when it comes to actually implementing them in practical educational settings. “It was shocking to see how little has been done out there when there is so much proven research here,” he said, “it’s as if academia and practical education were communicating through a dead letter box.”
One of his favourites is interleaving, a learning technique that involves mixing different topics or forms of material within a single study session. Since it was first proposed in the 1970s, numerous rigorous scientific research has demonstrated its effectiveness and benefits compared to the traditional blocked practice where students learn one topic at a time. Not only it was proven to improve long-term retention of information; learning a mixture of topics also requires students to differentiate between concepts and choose different strategies appropriate to the task, enhancing their problem-solving skills while building better neural connections.
One reason he thinks on why theories such as interleaving was not implemented despite its proven results, was its difficulty to do so. “Successfully implementing such a technique requires careful planning to ensure that different topics or skills are mixed in a way that makes pedagogical sense, as well as a personalised approach that is tailored to each students’ unique thinking pattern,” he suggested, “which quite frankly speaking, is not practical in a large scale.”
Another significant factor is due to the inertia of traditional education models - our education systems are big, well-established and therefore very sticky when it comes to big changes such as methods of teaching and curriculum designs. At the same time, it might not align well with standardised testing and uniform education standards. “The solution,” as Oscar later suggested, “is to do it privately, to design a tool that is agile enough to complement our existing education system, perhaps through technology, I suppose.”
Technology to the rescue - Artificial Intelligence
After one of our late night poker sessions back in 2020, Oscar and I were talking about possibly building something that would reshape education through technology, specifically artificial intelligence; though each from our own perspective and inspiration.
Reflecting on my teaching methods, I've come to realise what I once thought of as a teacher’s intuition is, in essence, a highly refined form of pattern recognition through years of experience. Interestingly, this resembles the capabilities of today’s neural networks. With the recent explosive advancement in deep learning and neural networks during the 2010s, such sophisticated pattern recognition isn't just a theoretical possibility but an emerging reality. More specifically in technical terms, throughout the past decade, significant progress has been made in key algorithmic innovations, such as improvements in backpropagation, convolutional neural networks (CNNs), and recurrent neural networks (RNNs), particularly Long Short-Term Memory (LSTM) networks. Combined with another area of breakthrough in hardware where the rise of Graphics Processing Units (GPUs) and later Tensor Processing Units (TPUs) provided the necessary computational power to handle complex calculations at speed, making deep learning practically viable. More importantly, advances in open-source frameworks like TensorFlow and PyTorch, along with increased academic and private investment, accelerated AI research and broadened its application scope.
This has given me a glimpse into a future where such personalised education, facilitated by artificial intelligence, can be massively scalable. As these systems integrate predictive analytics and natural language processing capabilities, they would bring a paradigm shift towards an educational framework where quality and personalisation are delivered on a mass scale, mirroring the intuition of an experienced educator but with the scalability of digital technology.
For Oscar, the perspective was different where he wasn’t focused on the delivery and exchange between the teacher and the student; rather, he envisioned harnessing students’ data to drive the creation of tailored content and curricula. He wanted to leverage the power of neural networks to effectively create a much more optimal curriculum sequence and personalised learning path, unleashing the potential of true interleaving.
Interleaf Technology Limited
All these would mark the beginning of Interleaf, an Education Technology Startup that missioned to reshape academic education through combining artificial intelligence with cognitive learning theories.
In the domain of academic education, we imagine a near future where education is seamlessly integrated with artificial general intelligence, creating a learning ecosystem that is both profoundly personalised and scalable. In this landscape, AI doesn't replace educators but amplifies their reach and effectiveness, offering each student a customised learning path.
In and out of the classroom, through both daily and academic interactions, neural networks would analyse individual learning styles, capabilities, and preferences to tailor the curriculum in real-time. Learning environments are dynamic, adapting to the student’s grasp of the material, ensuring that no concept is too challenging or repetitive. This creates an environment where students are continually challenged just enough to stay engaged and motivated without becoming overwhelmed. On top of that, it can help students align their strengths and interests with potential future careers, based on evolving labour market trends analysed by AI, ensuring that students are prepared for the jobs of tomorrow, not yesterday.
On the other hand, highly academic focused learning platforms would provide instant feedback, not only correcting students but explaining why an answer is incorrect, just like a personal tutor would. Based on individuals' thinking patterns and habits, these systems could even anticipate misconceptions before they become ingrained and offer corrective lessons immediately, with tailored analogies that would fit the individual interests. Not only will this drastically improve the efficiency of knowledge transfer, it could even fundamentally shift the way we understand while addressing learning difficulties.
This integration of AI into academic education promises a democratisation of learning opportunities, making high-quality, individualised education accessible to students regardless of geographic or socioeconomic constraints. As a result, we might see a generation of learners better equipped than any before to tackle the complexities of the modern world, empowered by an education system that understands, adapts, and grows with them.
Education beyond the academics
While Interleaf’s mission is rooted in enhancing the domain of academic education, I believe education itself extends far beyond academic subjects traditionally taught in school settings.
Education encompasses a vast array of domains that are critical to personal growth, these areas of learning are crucial for developing well-rounded individuals who can navigate the complexities of life beyond the classroom. By instilling social and emotional intelligence, we have to prepare learners to manage their emotions and build healthy relationships. Meanwhile, moral reasoning fosters a sense of ethics and responsibility, while financial literacy equips individuals with the knowledge to make informed economic choices. Health and wellness education encourages a lifestyle that values physical and mental health, contributing to a more productive and fulfilling life. After all, education itself is a life-long process and these are only just some aspects among so many others.
What are we working on right now (31 Oct)
As we forge ahead toward the future we envision, we are experimenting with different helpful tools that could empower educators across the globe. As of the time of writing this blog (31 October 2023) one of our favourite ideas is a generative AI tool to create educational exercises. With the current breakthrough development of Large Language Model (LLM) and generative AI, we are eager to harness these technologies to contribute to the way educational content is crafted and delivered.