The Problem with Evolutionary Art Is - Galanter (2010)

The Problem with Evolutionary Art Is…

Author: Philip Galanter — Department of Visualization, Texas A&M University Published: EvoApplications 2010, LNCS 6025, pp. 321–330, Springer-Verlag

Summary

Galanter identifies three structural problems with evolutionary art (EA) after 20 years of practice: the lack of a robust automated fitness function, genetic representations that prevent multi-level emergence and innovation, and the absence of coherent art theory. He argues most EA is self-contradictory — it claims to be inspired by evolution’s bottom-up process but is designed top-down — and proposes “truth to process” as a corrective aesthetic principle, where the evolutionary process itself becomes the content.

Key points

• The fitness bottleneck: Interactive Evolutionary Computing (IEC) is rate-limited by human judgment — populations stay small and generations few. Automated aesthetic evaluation (CAE) remains unsolved because it must ignore culturally determined aesthetic values that shift over time. Crowdsourcing (e.g. Sims’s Galapagos) trends toward mediocre aesthetic mean.
• Four levels of genetic representation (in increasing complexification capacity):
  1. Fixed parametric — genes directly map to fixed traits; no innovation possible
  2. Extensible parametric — more genes, still constrained
  3. Direct mechanical — genes describe drawing machines; single layer of emergence only
  4. Reproductive mechanical — machines create machines; multiple layers of emergence, like DNA → proteins → organelles → cells. This is what nature does, and what EA mostly ignores.
• Effective complexity (Gell-Mann): complex systems live in tension between order and disorder, allowing emergence across multiple scales. EA systems with single-level genotype→phenotype mapping can’t produce this — they plateau into sameness and never innovate like Pollock or Guston did.
• The incoherence of typical EA: natural evolution is bottom-up and non-teleological. Most EA is top-down (genetic representation pre-loaded to produce a desired result) and teleological — a direct contradiction of what it claims to be inspired by.
• Truth to process: borrowing from “truth to materials” in architecture and abstract expressionism — the essential nature of EA is the process of complexification, not the artifact. Aesthetic value should emerge from displaying that process honestly. Verbs over nouns; the artifact may be irrelevant.
• Generative art framing: generative art (including EA) is defined by the artist ceding control to an autonomous system — this is as old as art itself (symmetry, tiling, chance procedures). What’s new is the use of complex systems rather than simple ones.

My notes

Add your own notes here.