Education: Are We Doing It Right?

Of course not. Anyone you ask this question to will immediately point out several shortcomings in today's education - from an outdated curriculum to rigid teaching methods, a learning culture that is optimized for tests rather than understanding, and so on. Arguably, most of this is well-known, and attempts are being made to improve the situation. The world around us is changing rapidly, and improvements are being made rather slowly and certainly not at the scale required.

There is perhaps one particular development that is worth considering in more depth - computer science and AI. Computer science has grown rapidly in recent years, impacting everything from the intellectual world of sciences to the physical world of sports, from cutting-edge research to our everyday lives. In 2024, the Nobel Prizes in physics and chemistry were awarded to computer scientists, and of course, everyone today carries a networked computer in their pocket - a mobile phone.

So, what does computer science have to do with education, apart from being a popular field of study and an increasingly essential tool with widespread applications? The answer perhaps lies in the very foundations of computer science. Computer science at its core is about problem-solving - developing methods, algorithms and systems to meet various objectives. Professor Hal Abelson in his seminal course "Structure and Interpretation of Computer Programs" talks about the essence of computer science, which is profoundly insightful - computer science is not really about using computers, but it is really about us formalizing "how to" - how to do things, how to solve problems. Isn't education, at its core, about learning "how to" skills, how to solve various kinds of problems, and achieve our goals in life as well?

By getting computers to perform various tasks over these past years, what we have actually learned is various approaches to solve different kinds of problems. Computational thinking - the art of clearly defining a problem, decomposing it into manageable parts, and building solutions step-by-step in a methodical manner - is the key to problem-solving. In our education while we do learn how to solve specific problems in languages, mathematics, and sciences, general problem-solving skills and insights derived from computers are not perhaps taught as much. Instead, the focus remains narrow, for example, limited to coding specific algorithms in the current scope of computer science curriculum in basic K-12 education. Is there room to build more computational thinking competency and meta-solving skills? For long, computers have been a medium through which we discover insights and then apply them to various fields - but education has seen far less share of it.

There has been one spectacular recent development in computer science that is quite different from traditional computational thinking - Artificial Intelligence or AI. So somehow we have been able to teach computers how to solve problems without really laying out the step-by-step process, and this has worked out remarkably well in various fields, from self-driving cars to fairly generalized language understanding and reasoning abilities in LLMs.

AI models develop these non step-by-step way to solve problems - you can call it a form of understanding, implicit knowledge, statistical pattern recognition or what might loosely be termed intuition - and can solve problems that traditional computer science was unable to. In doing so, it has arguably come closer to human capabilities; after all, we too have these innate capabilities whether in language or through simple practice in areas like driving. However, AI can use this in areas where we as humans largely rely on step-by-step solutions - for example, current AI can do intermediate mathematics very well without any explicit reasoning. So should we try something similar in education, not focus too much on everything being explicit and step-by-step and instead try to nurture and develop such intuitions? Hmm, probably sounds absurd - after all, it seems natural that mathematics is solved using step-by-step reasoning at least for humans, even if AI can solve it implicitly. But then we did have individuals, human calculators like Shakuntala Devi who could do such mathematical computations using non-conventional methods. Are we missing out on developing intuitions in areas where we have historically relied on step-by-step solutions?

And even if we do not want to exclusively rely on a single approach - are they complementary? A great chess player can be more intuitive or more calculative - but they invariably possess both abilities. In fact, isn't it intuition that truly makes an individual expert across fields? Whether through years of experience or, in rare cases, natural prodigy, experts develop this special "intuition" that sets them apart. Such experts often have an uncanny ability to look at a problem and instantly sense the direction of a solution, even before they can explain why. So should the ultimate goal of education be to develop this so-called intuitive problem solving in one or more areas, and filling the interim gaps if any with step-by-step problem solving?

We can probably take a little inspiration from AI again here. Even if we cannot imagine solving everything by intuition (well at least until AI demonstrates it after achieving Artificial Super Intelligence or ASI level) - there is this holy trifecta in AI - training, data and inference which in fact represents the entire gamut of problem-solving. Training is the intuition or expertise, data is the knowledge, and inference is the step-by-step reasoning, and you perhaps need all these three; the balance may be different based on the maturity of the field and individual. The current education system, however, puts a lot of emphasis on knowledge and step-by-step solutions and leaves intuition to accidental emergence, that too in a narrow field of specialization after years of rigorous practice. Looks like something which can be done better? Can education focus more on building intuitive problem solving skills in addition to knowledge and step-by-step problem-solving skills?

There is another very important aspect to computer science in the context of education - learning by doing, where the digital canvas enabled by computers allows us to explore, experiment and express ourselves in ways not feasible earlier - personalizing and accelerating our learning. We have started to incorporate such computer-enabled experiential learning elements in the main curriculum - Programmable floor robots, where kids can code instructions to reach a certain position, or programming environments like Scratch where kids program interactive stories, games, and animations. This indeed is one area where a lot of emphasis of broader improvements in education has been focused; Montessori and other popular methods of education focus on natural exploration and activities of children instead of textbook instruction.

How do we go about developing the intuitive problem solving we talked about? Is learning by doing the key? Teaching kids something, probably step-by-step first and then letting them do tasks repeatedly, hoping they will acquire the intuition. But why do some people become experts while others do not, despite equal practice? Is there more to developing intuitions than what current natural exploration and repetition encompasses? Similar to LLM training, should we first pre-train and develop implicit skills in constituent areas? And then force development of intuition in the target area by presenting tasks and rewarding correct solutions but without focusing on the solution steps, only offering minimal guidance when needed? Should we become more tolerant of a few incorrect solutions early on, rather than insisting on correct answers for each and every problem before moving forward? Are we doing the right number of repetitions, at the right frequency? Or we are simply overfitting AI training methods on human learning.

There is perhaps a bigger change looming on the horizon that we need to consider - what will be the purpose of human life when AI can do everything better than humans, a world potentially devoid of physical and intellectual labor. After all, the purpose of education is to enable us to lead meaningful lives - many of us currently derive that meaning from the work that we do. What will we do when AI can do everything? And education should target the fields that will stay relevant for building skills, whether intuitive or methodical. Perhaps we will all become artists and creators in some way - pursuing creative writing, music or even building businesses. So should education work to fuel creativity, encourage independent ventures and deeper skill development? How do we teach for creativity - where there are no objective goals?

Even today there are two distinct paths and most of us choose somewhere between them. One is to select a specific field and focus on learning and developing expertise, a path followed by a small number of young achievers, from musicians to chess players. This is actually quite similar to how some of the world's most advanced AI systems are purpose-built, from chess engines to self-driving cars. The second approach, demonstrated by LLMs, is to develop broad foundational skills that provide a good base - in areas like language, reasoning, STEM knowledge - before fine-tuning for specific purposes. This is what most of us do today trying to make a living. In a world largely free of traditional work and to develop deeper expertise that distinguishes us from AI - it is likely we would shift more to the former path even if the degree and pace of that shift is unclear at the moment. While building some foundational skills may not be a bad idea to continue with, we may have to reconsider whether 15 odd years of formal schooling is the best approach. Should we instead create more opportunities for young people to explore and discover their calling? And as we've seen from both AI development and human learning, starting this exploration early often leads to better outcomes.

Alright, there seem to be many open questions: are we building the right skills in the right areas using the right approaches to achieve the right goals in education? While this seems like a lot, it is clear that education needs to rapidly evolve. Education is inherently future-focused - there are typically 15 years of education and when a kid from today completes education, AI with generalized superhuman capabilities will likely be present. Various estimates put it within a decade or so, more aggressive ones within a few years. If we can answer some or many of these questions, perhaps we can evolve education and humanity for the better. That is what Kyvl attempts to do - drawing inspiration and evidence from computer science and AI, with occasional sprinkles of imagination, to rethink the future of education.