Skip to content

Variation-Selection-Retention

Why It Matters

Human problem-solving defaults to planning: define goal, design solution, implement. This works for well-defined problems but fails when requirements are uncertain, search spaces are vast, or environments change. Variation-selection-retention provides an alternative: generate options, test them, keep what works, repeat. It's the engine of blind search that actually works.

The Key Move

When stuck on a problem where planning isn't working, ask: what's my variation mechanism (how am I generating options)? What's my selection mechanism (how am I testing them)? What's my retention mechanism (how am I preserving and building on what works)? If any element is missing or broken, fix it before trying harder at the other parts.

Classic Application

Genetic algorithms in computer science. Instead of programming a solution directly, you define a fitness function and let variation-selection-retention run. The algorithm generates random code variants, evaluates fitness, keeps the best, mutates them, repeats. This solves problems (antenna design, scheduling, game strategies) that are intractable to direct design.

Surprising Application

Career development. Planned careers ("I'll become X by doing Y") often fail because requirements and opportunities change. Evolutionary approach: generate varied experiences (variation), notice which produce results and satisfaction (selection), double down on those and eliminate others (retention). The best career "designs" emerge; they're rarely planned in advance. Explains why serendipity beats rigid planning.

Failure Modes

  • Inadequate variation: if all variants are similar, selection has nothing to work with. The process stagnates.

  • Weak selection: if fitness differences don't affect retention, variation accumulates as noise without direction.

  • Short retention: if successful variants aren't preserved long enough to be built upon, the process can't accumulate improvements.

  • Confusing evolution with optimality: evolution produces "good enough," not "optimal." It's path-dependent and constrained by history.

Go Deeper

  • Dennett, Darwin's Dangerous Idea - philosophical treatment of evolution as universal algorithm
  • Campbell, "Blind Variation and Selective Retention" - original articulation of evolutionary epistemology