Visually-guided behavior of evolved digital organisms
Year: 2004
Type: article
Abstract: Unsupervised digital organisms guided by artificial neural networks were allowed to evolve in virtual environments based solely on the success or failure of their responses to visual input. The criteria of success were avoiding collisions with surfaces and fully exploring the environment. The results showed that: 1) visually-guided behavior improved over successive generations; 2) organisms that evolved in one environment showed improved behavior in a novel environment; 3) the improved behavior was influenced by the image-source relationships they had experienced; 4) the improved behavior was influenced by the dynamic structure of visual input in the absence of any explicit model of memory; and 5) the improved behavior of the organisms was resistant to random modification of their neural networks. These observations show that increasingly successful visually-guided behavior can be generated simply by the incorporation into neural networks of the statistical relationship between images and their underlying sources.
Source: Journal of Vision
DOI: 10.1167/4.8.896
Institution: Duke University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,850
OpenAlex ID: W2011264274