Welcome. This blog documents what I’m learning about knowledge graphs & LLMs. It’s also the seed of an AI+knowledge graph project that I believe could transform education.
If you’d like to know more about how this project started & where it’s heading, read on. Otherwise, you can just start learning.
I have specific interests & objectives re: AI.
I’ve been skeptical of many of the claims being made about the nature of AI tools and their relationship to human consciousness (& I sometimes write about these topics on LinkedIn)—but that doesn’t mean that many of these AI tools aren’t promising—indeed, thrilling.
My day job is in healthcare, and standard LLMs simply aren’t going to cut it from a quality perspective. When it comes to generative AI in heatlhcare, an AI tool’s output needs to be both verifiable (meaning it meets your requirements, including requirements for truthfulness and reliability) and validatable (meaning it meets your user’s needs).
Hence my particular interest in the combination of LLMs and knowledge graphs, or what Microsoft and others are calling “GraphRAG.”
GraphRAG is one way to combine the probability-based AI tools in the news now—which are amazing, but not always trustworthy—with what is called a “neuro-symbolic” approach to AI, which means an approach that maps specific knowledge relationships in a way that models truth. Ideally, in a graphRAG approach, we can get to the best of both worlds.
The question driving me is: how can teachers and authors easily generate knowledge graphs related to their work that could provide a foundation for chatbots that could supplement the learning experience? How could tools like this transform education by connecting verified knowledge with the remarkable agility of LLMs? How could this transformation bring world-class, personalized education to people around the world without requiring the debt loads that current higher ed often entails?
For a teacher, these AI agents could be reliable teaching assistants, trained on the learning objectives and grading rubrics of the teacher while also always available and immediately responsive to student questions and drafts.
In the case of a writer, the AI agents could play the role of facilitator and tutor, guiding readers through difficult books, providing encouragement and a repository of notes, highlights, and terms that could aid review and recall.
In both these cases, AI can supplement human aims, helping us work toward goals we’ve defined personally—promoting individuality, creativity, and autonomy.
All existing AI tools are only useful when they are given a telos by humans. Without an end to aim for, they don’t do anything. But as AI grows more powerful, this means it matters more than ever how we imagine our ends, our intentions, goals, & collective future. If we don’t define the telos, it will be defined by people who are indifferent or even hostile to our common humanity.
So, this project is an attempt to build what I believe, and explore how we could use AI to improve education.
I’ll share my code snippets here from the various mostly Python-based AI/ML projects I’m working on—be warned, things might get messy. I do try to annotate my code carefully to make sure I know what’s happening where. I also type out all code snippets myself when working through books, rather than copying & pasting—if you’re following along at home, I encourage you do to this too, along with annotating as you go. These practices, recommended by Zed Shaw, have helped me greatly in learning what little I know.
To join in the fun, see what I’m currently up to, and get some sense of where I’m heading next, check out the roadmap.
I’m creating a general glossary for all the concepts I’m learning, and different collections of posts for the different books and projects I work through, each of which will be documented in the bibliography.
I’d be very happy to hear your thoughts about what I’m up to, as well as your corrections & suggestions. Thanks in advance!