Artificial psychophysics questions classical hue cancellation experiments.

Area of Science:

• Computational neuroscience
• Color vision research
• Artificial intelligence in perception

Background:

• The classical hue cancellation experiment is a cornerstone of understanding human color vision.
• Opponent color theory, derived from these experiments, posits specific neural pathways for color processing.

Purpose of the Study:

• To investigate whether human-like opponent curves in hue cancellation experiments are an intrinsic property of color processing or an artifact of experimental design.
• To determine the role of artificial network architecture versus experimental parameters in generating these curves.

Main Methods:

• Artificial psychophysics simulations using diverse artificial network architectures.
• Theoretical analysis through a change-of-basis analogy of the hue cancellation experiment.
• Systematic variation of cancellation lights and network parameters.

Main Results:

• Human-like opponent spectral sensitivities emerge in artificial networks regardless of architecture, provided a tristimulus-like input and a hue cancellation task are used.
• The specific choice of cancellation lights, not the network's internal structure, is critical for generating human-like curves.
• Deviations from the classical cancellation light choices lead to significantly different spectral sensitivities.

Conclusions:

• Opponent curves in classical hue cancellation experiments may be a byproduct of photoreceptor properties and specific experimental choices, rather than direct indicators of downstream neural processing.
• The architecture of post-retinal networks appears irrelevant to the emergence of these curves in the classical paradigm.
• This challenges the conventional interpretation of classical hue cancellation results in human color vision.