Predisposed and learned preferences for multipoint visual statistics in visually naïve newly hatched chicks

Recent studies have revealed that human and non-human animals (rats) can detect luminance distribution and correlations between pixels in an image (ranging from 2-point to 4-point). This sensitivity is believed to stem from optimization processes in the visual system that operate through efficient coding mechanisms to extract the most informative image statistics from the environment. However, it is yet to be determined whether this optimization is evolutionarily given by inborn mechanisms or shaped by visual experience. Here we report that newly-hatched visually naïve domestic chicks spontaneously prefer to approach luminance, 2-point and 4-point correlation patterns (respectively, horizontal lines and rectangular patterns), while showing no preference for 3-point correlation over white noise controls. This parallels the ranking observed in adult humans and rats, thus suggesting that evolutionarily given biological predispositions largely drive efficient coding of natural images. We also found that learning by exposure to visual stimuli, as occurs naturally during visual imprinting, induced a preference for white noise over point correlation patterns in chicks exposed to 3- and 4-point patterns. We hypothesize that this behavior could reflect chicks' preference for stimuli of lower statistical complexity.