Color feels objective. We point to the sky and say it's blue, and everyone agrees. But the agreement is largely linguistic rather than perceptual — the actual experience of color is remarkably variable between individuals, and in some contexts, dramatically so. The viral "dress" phenomenon of 2015, which appeared definitively white-and-gold to some viewers and definitively black-and-blue to others, was a perfect illustration of how differently human visual systems can interpret identical physical stimuli.
Color perception begins in the retina with three types of cone photoreceptors, each maximally sensitive to different wavelength ranges: roughly corresponding to short (blue), medium (green), and long (red) wavelengths. The signals from these cones are combined and processed through opponent channels — red versus green, blue versus yellow — and eventually reach the visual cortex where color is interpreted in the context of the entire scene.
Genetic variation in the photopigments within cones produces real differences in color perception between individuals. Normal color vision spans a range, with some people more sensitive to fine distinctions in certain hues and less in others. Approximately 8% of men and 0.5% of women have some form of color vision deficiency, representing a significant variant in the distribution of color perception.
The more striking source of color variation is contextual. Your visual system doesn't report raw wavelength — it attempts to identify the actual surface color of an object independent of the illumination falling on it. This process, called color constancy, normally works well, but it requires the visual system to make assumptions about light sources, and different viewers make different assumptions.
The dress illusion worked because the image provided ambiguous lighting cues. Viewers who unconsciously assumed the dress was in shadow interpreted the ambiguous colors as white and gold. Viewers who assumed daylight illumination interpreted the same ambiguous colors as black and blue. Neither group was wrong — they were simply completing an underspecified problem differently.
For designers and AI systems processing visual content, the lesson is that color is not a property of surfaces but a construction of visual systems interacting with light. AI color analysis systems must account for illumination, context, and viewing conditions — not just raw pixel values — to produce results that match human perception reliably.