What We Call Intelligence (Inference Systems)
- Jun 8
- 3 min read
A Reframe
What we often call “intelligence” is not a single, general-purpose ability. It is better understood as a collection of inference systems—mechanisms that take in information, generate predictions, and update based on feedback.
Each system operates in a different domain:
the body in space (movement, balance, force)
objects in motion (trajectories, timing)
causal systems (how things work)
social systems (what others think, intend, or believe)
These systems are not independent. They interact continuously. But they are not equally developed in every person.
Calibration, Not Categories
The difference between individuals is not that one person “has” a type of intelligence and another does not. The difference is in calibration. Each inference system can be more or less well-tuned to its environment. Each domain draws on multiple inference systems, but in different proportions.
A weightlifter relies heavily on calibrated models of force, balance, and timing, but also on causal understanding of technique and feedback from coaches.
A quarterback builds on similar sensorimotor foundations, but adds highly tuned models of trajectories, timing, and coordination with other players under pressure.
A caddie may rely less on execution, but draws on environmental prediction, probabilistic reasoning, and strategic modeling of outcomes over time.
A navigator (whether in the wilderness or a city) depends on spatial models, memory, and continuous updating based on partial information often under uncertainty.
A lawyer uses many of the same systems, but applied to language and social inference: modeling how arguments will be received, how others will respond, and how to shape those responses.
An accountant applies causal and pattern-recognition systems to financial records, identifying inconsistencies, reconstructing flows, and bringing noisy data into coherent structure.
The systems are the same. What differs is how they are combined, calibrated, and applied.
What appears as “talent” is often a system whose predictions align closely with reality where errors are small and correctable.
Why We Are Drawn to Certain Domains
People tend to gravitate toward activities where their inference systems produce reliable predictions. When prediction error is manageable align with outcomes:
actions feel natural
feedback is clear
progress is reinforcing
When prediction error is high:
actions feel uncertain
feedback is confusing
disengagement is more likely
Interest, in this sense, is not random. It is partly a signal of where our current models are already working and where they can be improved.
Learning as Error Reduction
Learning can be understood as the process of reducing prediction error over time. Each attempt produces feedback. That feedback updates the model. The goal is not perfection, but better alignment between expectation and reality. Repetition contributes to this process, but it is often slow and uneven. More effective is deliberate engagement:
targeting specific errors
receiving clear feedback
adjusting models in response
This is why expert training environments produce disproportionate gains. They accelerate the rate at which models are corrected and refined.
Movement Through Model Space
Learning is not just repetition. It is exploration. Each learner is moving through a space of possible models, searching for those that reduce prediction error in a given domain. This search is:
noisy when done alone
more efficient when guided
dramatically accelerated when social
Other people provide shortcuts through shared models, corrections, and perspectives that reduce the cost of exploration.
Cultural learning is a force multiplier on individual learning.
Implications
If intelligence is a collection of inference systems that improve through feedback and interaction, then learning environments must:
expose students to varied domains of inference
provide frequent, meaningful feedback
allow for repeated iteration without penalty
leverage social interaction to accelerate model refinement
This shifts the focus from measuring static ability to developing systems that improve over time.
An adaptive learner is not someone who knows the most. It is someone whose inference systems are well-calibrated, who can recognize when their predictions are wrong, and who has the tools to reduce that error. They can navigate uncertainty rather than avoid it.
