Your visual system is not evolved to be a colorimeter because that is not actually very useful for the kinds of things we use our visual system for. Thinking that your brain ‘should’ identify the same RGB values as the same ‘colors’ in a different context reflects a confusion about what invariant properties of the world the visual system represents as ‘color’.
Our conscious experience of color is related to the spectral composition of light that reaches our retinas but the RGB value of a pixel is not sufficient to describe the more complex qualia we label ‘colors’. If there is any ‘failure’ captured by this illusion it is a failure to understand what a good job the brain does of extracting useful information from the complex pattern of light that falls onto our retinas rather than a failure of the visual system. A colorimeter is a relatively simple $90 device. Matching the human visual system’s performance on the inverse rendering problem is an unsolved hard AI problem.
The anchoring phenomenon which can result in poor choices in certain circumstances on the other hand does reflect a ‘failure’ in the sense that a generally useful heuristic may lead us to make poor judgements. I’d say it is an example of misapplying a heuristic to a problem it is ill suited for. I think comparing it to the colour constancy phenomenon is misleading and inapt.
I totally agree that those functions of the visual system are features and not bugs, but I still think the analogy to biases holds; after all, there’s a strong argument that that biases can be features too. It’s all a question of whether they’re being used in the sorts of situations for which they were designed, or whether they’re in an unusual situation where they break down. In the case of vision, that’s specially designed optical illusions, but practically all modern thought is outside the original design specs for our cognitive heuristics.
Thinking that your brain ‘should’ identify the same RGB values as the same ‘colors’ in a different context reflects a confusion about what invariant properties of the world the visual system represents as ‘color’. … If there is any ‘failure’ captured by this illusion it is a failure to understand what a good job the brain does of extracting useful information from the complex pattern of light that falls onto our retinas rather than a failure of the visual system.
Your visual system is not evolved to be a colorimeter because that is not actually very useful for the kinds of things we use our visual system for. Thinking that your brain ‘should’ identify the same RGB values as the same ‘colors’ in a different context reflects a confusion about what invariant properties of the world the visual system represents as ‘color’.
Our conscious experience of color is related to the spectral composition of light that reaches our retinas but the RGB value of a pixel is not sufficient to describe the more complex qualia we label ‘colors’. If there is any ‘failure’ captured by this illusion it is a failure to understand what a good job the brain does of extracting useful information from the complex pattern of light that falls onto our retinas rather than a failure of the visual system. A colorimeter is a relatively simple $90 device. Matching the human visual system’s performance on the inverse rendering problem is an unsolved hard AI problem.
The anchoring phenomenon which can result in poor choices in certain circumstances on the other hand does reflect a ‘failure’ in the sense that a generally useful heuristic may lead us to make poor judgements. I’d say it is an example of misapplying a heuristic to a problem it is ill suited for. I think comparing it to the colour constancy phenomenon is misleading and inapt.
I totally agree that those functions of the visual system are features and not bugs, but I still think the analogy to biases holds; after all, there’s a strong argument that that biases can be features too. It’s all a question of whether they’re being used in the sorts of situations for which they were designed, or whether they’re in an unusual situation where they break down. In the case of vision, that’s specially designed optical illusions, but practically all modern thought is outside the original design specs for our cognitive heuristics.
Beautiful. Well said.