By G. Sawatzky, embedded-commerce.com
August 27, 2025
The field of artificial intelligence has experienced a shift that requires us to reconsider some established ideas about how we represent knowledge. Large language models (LLMs) and other modern AI systems demonstrate a powerful ability to connect concepts and recognize patterns, not through rigid rules and taxonomies, but through a more flexible mechanism that mirrors human cognition. This approach is reminiscent of philosophical concepts developed decades ago that challenged the very notion of perfect categorization.
For much of the computer age, the prevailing view of knowledge representation has been based on an unwavering need for precision. Systems were designed with exact database schemas and knowledge bases that required everything to fit into predefined categories. This approach, while providing a foundation for powerful tools, also created significant challenges. Real-world concepts often defy neat categorization, leading to logical paradoxes and complexity.
Classic examples highlight this issue: Is a tomato a fruit or a vegetable? Is a PhD student an employee, a student, or both? These questions expose the limitations of a purely rigid, taxonomic approach. The quest for perfect knowledge models becomes a struggle against the messy and interconnected nature of reality.
Philosophers have long explored this problem. The Austrian philosopher Ludwig Wittgenstein introduced the concept of "family resemblances" to describe how things in the world are related. In his view, a category is not defined by a single set of shared characteristics but by a series of overlapping similarities. Think of the members of a family: while no single trait is shared by all of them, they are connected by a web of common features like a similar nose shape, eye color, or speaking style.
Similarly, computer scientists like Joseph Goguen and John Sowa have argued for a more flexible approach to knowledge representation. Goguen, with his work on "fuzzy sets," proposed that categories could be defined by degrees of membership rather than a simple true or false classification. Sowa, a respected figure in knowledge representation, has long critiqued the limitations of purely formal logical systems, advocating for more human-centric models that can handle complexity.
These thinkers, each in their own way, challenged the idea that precision must come from perfect categories. Instead, they suggested that rigor could be found in a better recognition of similarity patterns, and that interoperability could come from intelligent interpretation rather than rigid standards.
Large language models have, almost by accident, leveraged this principle to solve a persistent technical problem: knowledge interoperability. They do not operate on a formal, rule-based logic. Instead, they process massive amounts of data to recognize patterns and statistical correlations between concepts. When an LLM understands that a "student information system" and an "educational management system" are conceptually related, it's not because they are defined by a strict rule. It's because the patterns of language used to describe them in its training data show a strong, overlapping similarity.
This ability allows AI to act as a crucial bridge between disparate systems that lack perfect, shared definitions. For example, a healthcare system that calls a record a "patient identifier" and a second system that calls it a "member ID" might be impossible to link through traditional, rule-based logic. However, an AI, by recognizing the family resemblance between the two terms based on their context and use, can establish a meaningful connection.
This approach brings us closer to a future where precision comes from recognizing patterns and connections rather than from the futile pursuit of perfect categorization. It means we might be entering an era where data systems and AI can work with knowledge in a way that aligns with how humans actually think about it: not as a perfectly categorized hierarchy, but as a rich, interconnected web of ideas that allows for seamless integration.
The argument for using family resemblances does not mean we should abandon foundational data modeling principles. The rigor of a well-defined conceptual model is still essential for building a robust and reliable system. A methodology like Object Role Modeling (ORM) is designed to create a single source of truth for an organization's internal knowledge, capturing business rules and relationships in an unambiguous, human-readable format.
The concepts discussed in this article apply primarily to the challenge of external interoperability. When two different systems, each with its own precise internal model, need to communicate, it is highly unlikely their schemas will align perfectly. Instead of the costly and difficult process of forcing a direct, one-to-one mapping between them, we can use AI's ability to recognize family resemblances. This allows for a more fluid and flexible connection, preserving the internal integrity of each system while enabling meaningful data exchange.
What if we could share these ORM-based models as lightweight JSON or YAML representations and let AI systems recognize their family resemblances automatically?
The quest for perfect knowledge representation might be giving way to something more useful: good enough knowledge recognition. This shift in perspective could provide us with the tools to handle the messy, ambiguous, and interconnected nature of information in the real world.