Symbolic & Connection - Building AGI

About time!

For most of your who has been following my blog for a while, or know me in person, I have been sharing my thoughts on how to build Artificial General Intelligence or AGI in short.

Even though deep learning and machine learning has provided the breakthrough on AI, together with the vast amount of data and seemingly infinite computation power brought about by the cloud architecture. It has a huge downside to it. The downside has been revealed for each and every iteration of AI, for instance, adversarial attack on neural network, hallucination by Large Generative Models. It will keep on happening unless we are able to understand how neural network, together with the different architecture of it, learns and set the rules and decision boundaries. I doubt that can happen given how intricate and precise the learning can be. As such, deep learning models will definitely violate one of the principles of ethical AI, imo, and that is “Transparency”.

Consider making a “book" donation to support and share my readings and research!
Link at the bottom of the newsletter or here. :)

Given all these, it is with strong belief that building the next level of intelligence will definitely involve putting together connectionist models (from machine learning) and also symbolic manipulation models such as Knowledge Graphs or Rules-Based. So here begs the question, how should they come together then? I have my own theory, which is published in an older blog post of mine. I will share it below. There are many ways they can come together for sure which is a research topic I am keen to work on but paying the bills comes first though.

One thing I like to point out is that I believe the research folks are also starting to explore putting these two schools of models together. One of them is adding knowledge graphs on top of LLMs, etc. This is definitely going to be another interesting research frontier and hoping to see more results towards building better Artificial Intelligence.

Your thoughts? Please comment below or share them via LinkedIn. :)

Recommended Content

Thoughts on Building AGI (Part 1) <Post> (Part 2) <Post>

Incorporating Knowledge Graphs on LLMs <Arxiv>