Semantics: The Design System as a Knowledge Source for AI
Working Question
Today's design systems were created primarily as a means of communication between designers, developers, and other members of the product team. If AI is becoming an active participant in product design, implementation, and governance, then a design system must be capable of communicating knowledge not only to people, but also to autonomous systems.
Main Working Thesis
A design system is no longer merely a means of communication between people. It is becoming a source of knowledge from which autonomous systems can design, implement, validate, and evolve a product.
Questions We Want to Answer
- What information does AI actually need when working with a design system?
- How do the needs of humans and autonomous systems differ?
- What properties must a design system have in order to be interpreted unambiguously?
- How can the risk of misinterpretation be minimised?
- Which parts of the knowledge should be explicit, and which can be inferred?
Thesis 1
AI Needs Knowledge, Not Documentation
People can work with incomplete documentation, implicit rules, and examples. Autonomous systems primarily require explicitly represented knowledge and the relationships between it.
Thesis 2
Examples Alone Are Not Enough
Current documentation often explains design principles through examples. AI, however, needs to understand not only the correct solution, but also the reasons, conditions, and constraints that led to it.
Thesis 3
Design Intent Becomes Part of the Specification
Defining how a component looks and behaves is no longer sufficient. We also need to describe which problem it solves, when it should be used, and which goals it is intended to achieve.
Thesis 4
Every Rule Must Be Unambiguously Interpretable
A design system must minimise the room for multiple interpretations. This does not mean eliminating all exceptions, but rather defining their conditions and relationships explicitly.
Thesis 5
AI Changes the Quality Requirements for Design Systems
Ambiguities, duplication, and inconsistencies that were often tolerable when interpreted by humans can lead to systematic errors when interpreted by AI. The quality of a design system will therefore increasingly be judged by the clarity and unambiguity of its interpretation.
Open Questions
- How can we measure the interpretability of a design system?
- How should exceptions and context-dependent decisions be represented?
- How can flexibility be preserved without sacrificing clarity?
- Which information should be explicit, and which can safely be inferred?
- How can we verify that AI has understood the meaning of a rule correctly?
What Needs to Be Validated in Practice
- Which information AI actually uses when working with a design system.
- Which types of ambiguity lead to the most significant problems.
- Which knowledge structures produce the best results.
- Which parts of current documentation can be replaced by a more formal representation.
- How to design a design system that remains understandable to both people and future generations of AI.