Umwelt Hypothesis: Rethinking universality in AI representations

burny_tech · x · 2026-08-20

Victoria Bosch et al. published an opinion piece in 'Trends in Cognitive Sciences' proposing the 'Umwelt Representation Hypothesis'. Challenging the notion of universal representations in AI, the hypothesis suggests model representations are constrained by training objectives and environmental experiences, similar to the biological concept of 'Umwelt' (self-centered world). The article calls for re-evaluating assumptions about 'general intelligence' and emphasizes understanding the specific boundaries of model capabilities.

Original post →

More from AGI Musings

AGI Musings channel →