The Umwelt Representation Hypothesis: Rethinking Universality

Bosch, Victoria, Sommers, Rowan P., Doerig, Adrien, Kietzmann, Tim C., The Umwelt Representation Hypothesis: rethinking Universality, Trends in Cognitive Sciences, Aug. 8 ,2026, doi: 10.1016/j.tics.2026.07.004

Highlights: Artificial neural networks are increasingly used to study neural representations.

Recent studies have found a surprising degree of representational alignment between artificial neural networks and biological brains. In addition, broad alignment among ANNs of different modalities and architectures has been observed.

These findings have been interpreted as evidence for Universality: the idea that sufficiently capable systems all converge on a single shared representation of reality.

This interpretation has large implications for model comparisons that require meaningful differences between artificial neural networks: if all artificial neural networks eventually align with the brain, then creating new model classes leads to no new insights.

This opinion article argues that this inference is too strong and proposes the Umwelt Representation Hypothesis: representational alignment emerges from overlap in ecological constraints, not convergence to one privileged world model.

Abstract: Artificial neural networks are increasingly used to study neural representations.Recent studies have found a surprising degree of representational alignment between artificial neural networks and biological brains. In addition, broad alignment among ANNs of different modalities and architectures has been observed.These findings have been interpreted as evidence for Universality: the idea that sufficiently capable systems all converge on a single shared representation of reality.This interpretation has large implications for model comparisons that require meaningful differences between artificial neural networks: if all artificial neural networks eventually align with the brain, then creating new model classes leads to no new insights.This opinion article argues that this inference is too strong and proposes the Umwelt Representation Hypothesis: representational alignment emerges from overlap in ecological constraints, not convergence to one privileged world model.