# Genetic Algorithm **Domain:** Evolutionary Computation / Optimization **Doc Type:** Canonical Concept Node **Maturity:** Foundational ## Definition A **genetic algorithm** searches a space by maintaining a population of encoded candidate solutions and repeatedly applying evaluation, selection and variation such as mutation and recombination. ## Distinctions Genetic algorithms abstract selected features of biological evolution. Candidate encodings are not genes in the full biological sense, fitness is usually an engineered objective, and generations may be centrally scheduled. Their value lies in search over irregular spaces where direct construction or gradient information is unavailable. [[articles/The Evolutionary Roots of Silicon Valley|The Evolutionary Roots of Silicon Valley]] uses the ST5 antenna as the canonical local artifact. One Ames approach varied a real-valued parameter vector and selected candidates that better met radio and geometry requirements. ## Governance Context The objective function governs the population. A system can discover unexpected ways to score well, so constraints must include externalities and validation outside the simulator. ## Key Insight **A genetic algorithm automates variation and retention; it does not automate the legitimacy or completeness of fitness.** ## Sources / Provenance - John Holland, _Adaptation in Natural and Artificial Systems_, 1975. - NASA ST5 antenna report: https://ntrs.nasa.gov/search.jsp?R=20040152147 ## See Also [[wiki/Evolutionary Algorithms|Evolutionary Algorithms]], [[wiki/Genetic Programming|Genetic Programming]], [[wiki/Selection Gradient|Selection Gradient]], [[wiki/Evolved Antenna|Evolved Antenna]]