Where the optimization engine Analogy Breaks

Calling intelligence a optimization engine is useful for about five minutes. The comparison can expose structure that ordinary language hides. The problem begins when the comparison stops behaving like a tool and starts behaving like a claim about what the thing literally is.

Engineering becomes clearer when intelligence is decomposed into capabilities such as prediction, search, representation, control and learning instead of treated as one hidden substance.

What the metaphor gets right

Engineering becomes clearer when intelligence is decomposed into capabilities such as prediction, search, representation, control and learning instead of treated as one hidden substance.

This is why engineering metaphors are so attractive. They compress a complicated relationship into something we can manipulate mentally. In the best case, the metaphor produces predictions and sharper questions.

What it smuggles in

Optimization systems pursue explicit objectives. Biological agents operate under changing needs, ambiguous goals, limited information and embodied constraints.

Engineering words often carry hidden assumptions: a designer, a specification, modular ownership, explicit goals, a stable machine underneath the program. Those assumptions must be tested rather than inherited.

Return to the physical story

Intelligent behavior combines learning, perception, memory, planning, abstraction, adaptation and social interaction. Different organisms solve different ecological problems using very different bodies and nervous systems.

Once the mechanism is back in view, the analogy can be kept on a short leash. We can state which relation is genuinely similar and which difference would make the analogy generate a wrong prediction.

My three tests for a metaphor

1. What mechanism does this comparison help me see?
2. Which engineering assumption does the natural system not share?
3. What would I predict incorrectly if I took the comparison literally?

The third test is the most important. An analogy that cannot fail is decoration, not explanation.

Beyond the borrowed language

Turing reframed machine intelligence around observable capacities, while comparative cognition expanded scientific interest beyond human-style reasoning.

Intelligence may be less like a quantity stored inside a system and more like a relationship between an agent, its history and its world.

If two systems solve the same problem through entirely different mechanisms, in what sense do they share intelligence?

A good metaphor should eventually make itself less necessary. It gets us close enough to the real system that we can see where the scaffolding ends.

Reading trail

• Aeon — We are not machines
• NASA — Astrobiology Program Overview

These links are starting points for the scientific and historical ideas. The systems interpretation, analogies and conclusions here are my own.