🧬 kbird.ai

Black Boxes

4 min read

Black Boxes

Understanding systems through input-output behavior

The Concept

A black box is a system whose internal workings are unknown or irrelevant — we understand it only through its inputs and outputs. Cybernetics treats many systems as black boxes, focusing on behavior and function rather than internal mechanism.

“The black box is a system whose internal structure is unknown or irrelevant to the investigation.” — Ashby

Why Black Boxes?

Epistemological Necessity

We can never fully know another mind, a complex machine, or a social system. We only see:

  • What goes in (inputs)
  • What comes out (outputs)
  • How they relate (transfer function)

Practical Utility

  • Don’t need to understand brain to predict behavior
  • Don’t need to know code to use software
  • Don’t need to see inside to control

The Observer’s Position

All observation is from outside. Even when we “look inside”:

  • Microscope reveals new black boxes (cells)
  • Cells reveal organelles (more black boxes)
  • It recurses indefinitely

Understanding Black Boxes

The Ashby Method

  1. Apply inputs systematically
  2. Observe outputs carefully
  3. Map input-output relations (transfer function)
  4. Build a model that reproduces the behavior
  5. Test predictions

You never know if your model matches the internal structure — only if it predicts behavior.

Example: Understanding a Person

  • Input: Say “hello”
  • Output: They smile, say “hello” back
  • Model: “This system responds greetings with greetings”
  • You don’t know their internal experience
  • You might be wrong (maybe they’re smiling from gas pain)
  • But the model works for prediction

Isomorphism

Two black boxes are isomorphic if they produce the same input-output behavior, regardless of internal differences.

Examples

  • Calculator and abacus: Different mechanisms, same math
  • Brain and computer: (possibly) different substrates, same function
  • Two people: Different neurons, similar responses

Functionalism

The philosophical view that what matters is function, not substrate:

  • Mind is to brain as software is to hardware
  • Consciousness is organizational, not material
  • A perfect simulation would be the real thing

The Black Box Problem in AI

The Opacity Problem

  • Large language models are black boxes (billions of parameters)
  • We don’t know how they produce outputs
  • We can only test inputs and observe outputs
  • This creates safety concerns

Alignment

How do we ensure black box AI behaves as intended?

  • Test extensively: But can’t test all inputs
  • Interpretability research: Open the black box (partially)
  • Constitutional AI: Define desired input-output patterns
  • Human feedback: Shape behavior through interaction

Are We Black Boxes?

To an outside observer, human minds are black boxes. We:

  • Receive sensory input
  • Produce behavioral output
  • Have internal processes we can’t fully articulate

The simulation argument: If we can’t tell if others are conscious, can we tell if AI is?

White Boxes vs. Black Boxes

Black BoxWhite Box
Input-output onlyInternal structure visible
Functional understandingMechanistic understanding
PredictionExplanation
EngineeringScience
AI/ML modelsTraditional algorithms
Other minds(none — always black boxes)

Most systems are gray boxes — some internal visibility, some opacity.

Applications

Reverse Engineering

Treat competitor’s product as black box. Reproduce function without copying mechanism.

Medicine

Treat patient as black box when mechanism unknown. Does treatment X produce outcome Y?

Psychology

Behaviorism treated mind as black box. Skinner: “The causes of behavior are outside the organism.”

Engineering

Test components as black boxes. If input A → output B, use it. Don’t need to know how it works.

References

  • Ashby, W.R. (1956). An Introduction to Cybernetics (Chapter 6)
  • Ashby, W.R. (1962). Principles of the self-organizing system
  • Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine
  • Braitenberg, V. (1984). Vehicles: Experiments in Synthetic Psychology

The box is black. The behavior is visible. The mechanism is mystery. We work with what we have. 📦

Graph View

Backlinks