Is my "red" your "red"?: Unsupervised alignment of qualia structures via optimal transport

Published: 02 Mar 2024, Last Modified: 02 Mar 2024ICLR 2024 Workshop Re-Align PosterEveryoneRevisionsBibTeXCC BY 4.0
Track: long paper (up to 9 pages)
Keywords: Unsupervised Alignment, Qualia structure, Optimal Transport
Abstract: Whether one person's experience of "red" is equivalent to another's has long been considered unanswerable. One promising approach to resolving this fundamental question about consciousness is the intersubjective comparison of the similarity relations of sensory experiences, termed "qualia structures". Conventional methods for comparing similarity relations largely sidestep the issue, assuming that experiences elicited by the same stimuli are matched across individuals, and thus ruling out the possibility that my "red" could be your "blue". Here, we present an unsupervised optimal transport method for assessing the similarity of qualia structures without presupposing correspondences between individuals. To validate and demonstrate the utility of the proposed approach, we analyzed a massive dataset of subjective color similarity judgments from color-neurotypical and color-blind participants. We show that optimal correspondences between qualia structures within color-neurotypical participants can be "correctly" aligned based solely on similarity relations. In contrast, qualia structures from color-blind individuals could not be aligned with those of color-neurotypicals. Our results offer quantitative evidence for the interindividual structural equivalence or difference of color qualia, implying that a color-neurotypical person's "red" is indeed another color-neurotypical’s "red", but not a color-blind person's "red", from a structural perspective. This method is applicable across modalities, enabling general structural exploration of subjective experiences.
Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.
Submission Number: 51
Loading