By Anagh Prasad, Investor, Accel
Since Benjamin Bloom’s seminal study on methods of instruction came out in 1984, the ‘Two Sigma Problem’ has become one of the most recognizable concepts globally in the theory of education. The famous study alludes to a set of experiments that conclude that one-on-one tutoring results in the average student performing two standard deviations (sigma) above the average from a classroom setting.
The beauty of the Two Sigma problem theory lies in its immediate intuitiveness. Across nations and societies, personal tutoring is a clear preference for anyone who can afford it. In India, the concept of a revered ‘Guru’ who enlightens his disciples with knowledge through super-personalized attention finds numerous references in both history and mythology. As the Industrial Revolution proliferated, coinciding with the invention of the printing press, education moved from being a noble person’s privilege to a more accessible bridge to opportunity. It marked the introduction of the classroom method, which continues to be omnipresent centuries after its introduction.
Indian classrooms, in particular, have continued to be characterized by large average sizes – predictable for a country with over 250M actively enrolled students and much less than 1/100th the number of high-caliber teachers. Even among the most recent attempts at delivering education, aka ed-tech, one-to-many interactions dominate all viable business models. Bloom’s notes highlighted this challenge very well: he concluded that one-to-one tutoring is “too costly for most societies to bear on a large scale” and challenged education researchers to find methods of group instruction as effective as personal tutoring.
We at Accel in India believe that the long-standing two-sigma problem might finally have gotten a promising solution from technology: Generative AI. If we pan out a bit to observe what LLMs are good at – analyzing a tonne of unstructured information and creating personalized answers – their applicability to powering learning tools becomes quite apparent. Unsurprisingly, many students are among the most regular users of ChatGPT, and global ed-tech giants like Duolingo and Khan Academy have already seen remarkable feedback on their recently launched AI tutor features.
We predict that, over the next few years, Generative AI will be applied to Indian education extensively in the following three ways:
Hyper-personalization: While AI-based recommendation systems have long powered surfacing relevant content to users (ed-tech included), generative AI goes one step beyond by not just recommending but creating highly contextual content. Customized learning paths for every student are poised to be a reality as the marginal cost of creating new content further descends to near zero.
In the short to medium term, many emerging personalization features will become omnipresent in ed-tech: AI-generated micro lectures, 24*7 available QnA bots, personalized test feedback, and instant practice problems, to mention a few. In the long term, AI-generated customized learning paths can help us move from the current top-down, relatively rigid design of pedagogy in India to a more fluid one – with more weightage to original expressions (e.g., AI-art) and intersectional topics (e.g., computational chemistry vs chemistry and computer science).
New intuitive interfaces: Many, if not most, of today’s popular digital learning applications look similar: live/recorded video lectures, a comments-like section for QnA, and practice problems/tests. However, in a world where computers come with inbuilt capabilities to understand our language, many assumptions around the suitable learning interface should be (and are being) questioned. The best ed-tech interfaces of this decade will create original designs that combine multi-modal input capabilities (chat, voice, image, symbols) with immersive, generative UI spanning audio, video, text, and 3D.
AI for teacher productivity: Within a few quarters of LLMs going mainstream, we’ve already seen ‘co-pilots’ pop up for many professions – the most notable being software development. The 85M+ global teacher population will likely be one of the following early adopters for specialized teaching co-pilots. From helping craft classroom lessons to auto-grading assignment submissions, such applications can free up teachers’ busy work to help them focus more on nurturing student relationships, a significant driver for delivering learning outcomes.
We believe that Generative AI offers a once-in-a-generation level of transformative potential for education that will continue to manifest itself for many years into the future. We are particularly excited about the chance to create a more equitable world with accessible quality education. Secondly, we hope that unlocking true personalization in education with AI will lead us to celebrate every learner’s individuality rather than suppressing it, culminating in more original thinkers who propel humanity forward.