Learning Occurs In Networks

All learning occurs in networks. This ranges from gene expression networks to networks of cells, from neural networks in brains to artificial neural networks in large generative models like Chat GPT.

Mathematically, we can conceptualize these networks as graphs, comprised of nodes that are connected together by edges. For example, in an artificial neural network, backpropagation adjusts the weights of the edges that connect the nodes together, enabling the model to learn. Some form of this process occurs in all other types of network, even the most basic ones. Consider a physical system where blocks are interconnected by a series of springs. If this system is manipulated over time, the direction in which the springs are most frequently stretched will cause them to become progressively less tightly coiled. In this way, the system adapts and learns from the environmental perturbations it experiences. Even this simple network is learning!

💡 In an age driven by AI innovation; everyone should know that learning thrives in networks 💡

At a simplistic level, we can say that AI=DATA+COMPUTE, so let’s think about DATA and COMPUTE in terms of networks:

✔  The good news is that the latest machine learning models already possess a graph structure with weights based on neural networks. Furthermore, most of these foundational models are being developed externally by a few large organisations with access to massive compute and data. The majority of organisations can simply buy in these foundational models and use them internally.

✖  The bad news is that organisations must supply their proprietary data themselves, and most of that data is currently held in relational databases. This siloed relational data is not currently organised in one networked graph structure. Without this network configuration, the data is not in an optimal structure for learning.

So, if all learning occurs in networks and most organisational data is held in relational databases, then there is one simple takeaway for organisations wishing to prepare themselves for the age of AI: transition your data into a graph structure and do it as quickly as you can!