There is no single, universally agreed-upon definition of intelligence. We often measure it as raw processing power, logical reasoning, or the amount of accumulated information. However, for complex systems and adaptive biology, the most useful definition is behavioral flexibility.
Popularized by neuroscientist Suzana Herculano-Houzel, behavioral flexibility is the capacity of an organism or agent to modify its behavior based on past experiences and the anticipation of future states.
This definition shifts our focus from what is inside a brain or a model to what the system does. Crucially, behavioral flexibility is substrate-independent. It does not require a complex brain or nervous system:
- Slime mold (Physarum polycephalum) possesses no brain, yet it displays behavioral flexibility by adapting its growth patterns to find food and avoid old, barren paths.
- An isolated LLM operates with fixed neural weights; it responds from the exact same state every time. It has very little behavioral flexibility.
- An AI agent (the LLM wrapped in a harness with persistent memory, tools, and feedback loops) can change its state, observe the consequences of its actions, and adapt its workflow across sessions.
In this light, intelligence is not a static property of a component. It is an emergent property of the entire system’s coupling with its environment, measured by its capacity to adapt.