A History of Genetic Algorithms
— From Darwin to Holland

On the Origin of Species establishes the theory of natural selection: organisms with advantageous traits survive and reproduce more successfully, passing those traits to offspring. Over generations, populations adapt to their environments. Darwin provides the biological blueprint that GA would one day formalize.

Mendel's pea plant experiments establish the laws of heredity — that traits are passed from parent to offspring in discrete, predictable units (later called genes). Mendel's work provides the genetic mechanism that Darwin's theory lacked. Together, Darwin and Mendel lay the complete conceptual foundation for GA.

In "Computing Machinery and Intelligence," Turing asks: can machines think? He proposes the possibility of learning machines — computers that improve through experience. In a brief section, he speculates about "evolutionary search" as a possible path to machine intelligence: "Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's?" This is perhaps the first published suggestion that evolution could be implemented computationally.

Multiple researchers independently develop computational systems inspired by evolution:

• Box (1957): "Evolutionary operation" — a statistical process optimization method that uses mutation-like perturbations.
• Friedberg (1958): Evolves computer programs using random mutation.
• Bremermann (1962): Applies evolution to mathematical optimization, coining "evolution strategy."
• Rechenberg & Schwefel (1965): Develop Evolution Strategies (ES) at TU Berlin for optimizing fluid dynamics problems — a parallel development to GA that converges later.

John Holland begins developing the theoretical framework that will become Genetic Algorithms. He is motivated by the question: how can systems adapt to their environments in the same general way that biological organisms do? Holland's insight is that the key mechanism is not just mutation (random search) but crossover — the systematic recombination of partial solutions.

In 1975, Holland publishes Adaptation in Natural and Artificial Systems, the foundational text of GA. He introduces the Schema Theorem, which provides a mathematical basis for why crossover works: short, low-order, high-fitness patterns (schemata) grow exponentially across generations, while simultaneously the algorithm explores exponentially many hyperplanes in parallel (the implicit parallelism principle).

Holland's students — particularly Kenneth De Jong and David Goldberg — develop GA further. De Jong's 1975 thesis systematically evaluates GA on benchmark functions. Goldberg applies GA to gas pipeline control optimization, demonstrating real engineering utility. The field grows rapidly through the 1980s.

David Goldberg's Genetic Algorithms in Search, Optimization, and Machine Learning becomes the standard textbook. Clear, accessible, and comprehensive, it introduces GA to a generation of researchers across disciplines and catalyzes explosive growth in applications.

Karl Sims demonstrates evolved virtual creatures (1992) and evolved images using CPPNs (1991). William Latham creates "FormSynth" — evolutionary sculpture. These demonstrations show that GA can explore aesthetic and creative spaces, not just engineering optimization landscapes.

Hiroaki Takagi formalizes the field of Interactive Evolutionary Computation (IEC) — GA systems where humans serve as fitness evaluators. His 2001 survey in the Proceedings of the IEEE establishes IEC as a recognized research area, citing applications from music and art generation to industrial product design.

Secretan et al. launch Picbreeder — a web platform where users collaboratively evolve CPPN-based images by selecting from candidate offspring. Users build on each other's evolutionary lineages, creating a shared creative ecosystem. Picbreeder demonstrates that IEC can produce complex, beautiful results through sustained collaborative evolution. gene46's Mosaic mode is directly inspired by Picbreeder's CPPN approach.

GA designs an antenna for the ST5 spacecraft mission — a real engineering application deployed in space. The evolved antenna outperforms all human-designed alternatives. This becomes a landmark case study in GA's practical utility.

Neuroevolution — evolving neural network architectures and weights — becomes a productive research direction. Systems like NEAT (NeuroEvolution of Augmenting Topologies, Stanley & Miikkulainen 2002) and OpenAI's evolutionary strategy research show that GA and gradient-based learning are complementary. In 2022–2025, evolutionary methods are increasingly combined with diffusion models and large language models for creative applications.

gene46 launches as an interactive evolutionary generative art system combining IEC principles with modern web technology. Users evolve Bézier-curve Line Art and CPPN-based Mosaic art through swipe-based selection on mobile devices — making evolutionary art accessible to anyone, anywhere.