The Bowtie Pattern
There is a recurring pattern in learning systems, often referred to as the "Bowtie," where a large amount of information is compressed into a smaller representation and then expanded back into a larger set of information.
In neural networks, this occurs in what are known as encoder-decoder architectures. Input sequences are encoded into a latent space, then decoded into outputs that resemble the input but appear in a new form or context.
This pattern recurs across multiple scales. For example, large language models (LLMs) can be seen as a compression of the web – distilling petabytes of text into gigabytes of floating-point numbers. Each word on the web is itself a compression of human-perceived reality, packaged into a neat semantic structure.
Another good example of the knot part of a bow tie is standards. HTTP is a classic compression point through which many different types of data pass and from which many things emerge. With the rise of LLMs, we’re now seeing new standards emerge - for example, the Model Context Protocol (MCP), a protocol designed to facilitate communication and data exchange between AI models and their clients, or the newly announced Agent2Agent Protocol from Google.
Learning occurs through this compression process because useful compressions inherently involve generalisation. If a compressed representation of a large set of information can apply to many other situations, it is valuable precisely because it captures and generalises an underlying pattern.
The "Bow Tie" for your organisation is your ontology. You aim to compress useful abstractions into core ontological classes. This is a human-in-the-loop, collective learning process - aided by machine learning, but grounded in continuous interaction with people across your organisation. An active feedback loop exists between humans and machines, mediated by the ontology.
For all the reasons mentioned above, it is important that your ontology is based on standards – and fortunately, the Semantic Web already provides many of the standards we need.
⭕ What is an Ontology: https://www.knowledge-graph-guys.com/blog/what-is-an-ontology
⭕ Bowtie Concept in Cognitive Systems: https://www.youtube.com/watch?v=wQHaGhOrnug
⭕ The Great Compression: https://www.knowledge-graph-guys.com/blog/the-great-compression
⭕ Human In The Loop: https://www.knowledge-graph-guys.com/blog/human-in-the-loop
⭕ LLMs + Ontologies: https://www.knowledge-graph-guys.com/blog/llms-ontologies
