I’ve always thought that Neal Stephenson’s The Diamond Age was his best book. One might claim that Snow Crash was better (it was the book that introduced the concept of the Metaverse), but there was something about The Diamond Age that stuck in my head for years afterwards. This was a fictional world where nanotechnology reigned supreme – and interestingly, where usage of private decentralized currencies caused the collapse of most nations. But neither of those points was what made this book interesting to me.
At the heart of the book was the “Young Lady’s Illustrated Primer,” an interactive book that played an important role in the growth and development of the story’s main characters, Nell and Elizabeth. The Primer, an educational tool powered by advanced AI, provided individualized tutoring and guidance – and it adapted to each girl’s unique circumstances and learning needs. It became the girls’ life-long companion – and it not only enriched their lives but also empowered them to become the heroines of their own stories.
I can’t help but think about how recent advancements in AI, such as ChatGPT, could help us build something like the Primer now. As a father with two young children, I’ve been thinking a lot about what made the Primer so powerful. For me, it all comes down to the idea of individual tutoring. Numerous studies have highlighted the significant impact of individual tutoring on educational outcomes. One of the most influential studies in this field is the ”2 Sigma Problem” by educational psychologist Benjamin Bloom. Bloom’s research demonstrated that students who received one-on-one tutoring performed two standard deviations better than their peers who learned through conventional classroom instruction (this is the difference between 50th percentile and 98th percentile!). More recently, in his study titled “The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems”, psychologist Kurt VanLehn indicated that modern intelligent tutoring systems could achieve effects that were very close to human tutors – albeit closer to one standard deviation instead of the two sigma Bloom indicated.
Individual tutoring can be game-changing – but has historically been unscalable. As a family with two working parents, making time for personalized study sessions with our children isn’t always an option (it doesn’t help that our kids seem to never listen to us 😊). Additionally, finding tutors in many non-urban locations can be difficult and even if you do find one, you must make it work in limited time windows. Financial access is another barrier to entry – you must have the means to afford these tutors – a proposition that has largely been limited to the ultra-wealthy. This is frustrating because the findings on individual tutoring underscore the profound impact of personalized attention on students’ learning progress. What could be the answer?
To me, it’s technology. I believe ChatGPT has opened the door to the creation of individual tutoring applications that can provide personalized, adaptive learning experiences for students. But I see potential for enhancing educational interactions far beyond text-based AI, especially when I consider how generative AI models could revolutionize the way we interact with digital learning environments. For example, models like OpenAI’s DALL-E can generate realistic images, while AI-driven voice synthesis technologies, such as WaveNet, can create lifelike voices.
Combining these advancements with language models like ChatGPT could pave the way for the development of truly immersive learning experiences. Students could engage with AI-powered tutors that not only adapt to their learning needs but also manifest as realistic characters with images and voices, making the learning process more engaging, relatable, and effective. I can’t help but be excited about the potential of using these technologies.
I envision a range of features that would be essential to bring these AI-driven “Primers” to life:
- Integration of cutting-edge AI models, such as ChatGPT for natural language understanding, DALL-E for image generation, and WaveNet for voice synthesis
- Training of the AI models on current school curriculums, pedagogical best practices, and individual learning strategies to ensure alignment with educational standards
- Designing the Primer to run on a wide variety of devices, from low-end laptops to high-end tablets, making the technology accessible to students from all socio-economic backgrounds
- Prioritizing accessibility and adaptability, allowing the AI-driven learning companion to cater to each student’s unique needs and circumstances
There’s a lot that needs to happen before companies can deliver on this vision. I would be remiss if I didn’t note that these models are still far from perfect. Today, they still frequently deliver incorrect information and hallucinate details that potentially stretch the truth. Many topics the models could engage on are not appropriate for children, and there are questions about bias in these models that need to be addressed. But I believe these are all tractable challenges for founders looking to build a product in this space.