Ask for a definition of intelligence and you get one quickly: the ability to acquire and apply knowledge and skills. It is a reasonable start. Break it apart and three things fall out. Intelligence is an ability, so it belongs to a system. It concerns acquiring and applying knowledge. And applying skills hints that the system knows something about itself.
The definition I have used for years is shorter: intelligence is the ability to form, update and use concepts.
Data becomes information, relative to someone
Take intelligence as the conversion of data into information, where information is simply useful data. Usefulness is relative. A doctor and an engineer read the same report and come away with different information. So information is relative, and relativity implies context: every concept a system forms is bounded by its perception and its domain.
A musician and a chef both have excellent pattern-related abilities. Each is intelligent relative to a domain. Neither is intelligent in the abstract.
The mind as a system of relations
A mind can be described as a thinking system. Thinking is the use of the mind's data, with some intention, to produce thoughts in reaction to an action. Everything in a mind breaks down into relations: rules, facts, assumptions and procedures are each a relation or a group of them. A mind tries to bring order to what it gathers, and order is possible only when information can be categorised and a pattern found. That moment of finding the pattern is the feeling of satisfaction.
Those relations need a representation, and the representation is limited by what the system has: its senses and actuators, physical and mental. An agent with different organs forms different concepts from the same world.
Seven working propositions
Written as a theory, the picture looks like this.
- An intelligent agent is a system of sensory organs (internal and external), actuator organs, memory and processing.
- It can have many states, but it must begin with an initial state of information.
- Its degree of intelligence depends on its organs and its initial state.
- It is a responsive, data-perceiving system that works through an information protocol.
- All information can be represented with the fundamentals of mathematics, logic and linguistics, so such a protocol is achievable.
- Past some threshold of intelligence, an agent is bound to exhibit something like consciousness.
- Because intelligence is relative, it has personality.
Truth is relative too
If intelligence is how a system measures the truth it understands, truth is relative as well. Your truth and mine can differ, and the question of who is right gets harder than it looks. Software offers a cleaner way to say it. There is a logical truth, the model, and a perceived truth, the view of that model in some environment. Model and view depend on each other; they are entangled.
Why this matters for building systems
None of this is abstract to an architect. A system that is going to behave intelligently in a domain needs its concepts formed in that domain's terms, a way to update them from experience, and a measure of how well it uses them. Form, update, use. Those three words became the shape of the work: Domain Language Models that form concepts in the language of their world, structural learning that updates them without gradient descent, and a way to measure intelligence within a domain rather than in general.
Drafted in 2022
Updated for site in 2026