The Illusion-Illusion: Vision Language Models See Illusions Where There are None

TL;DR

This paper investigates whether vision language models are susceptible to illusions by presenting them with benign, non-illusionary stimuli that they misinterpret as illusions, revealing fundamental processing errors.

Contribution

It introduces the concept of illusion-illusions to test vision language models and demonstrates their tendency to falsely perceive non-illusions as illusions, highlighting core model limitations.

Findings

Models often mistake benign stimuli for illusions.

Vision language models exhibit fundamental perception errors.

These errors reflect broader issues in current perception modeling.

Abstract

Illusions are entertaining, but they are also a useful diagnostic tool in cognitive science, philosophy, and neuroscience. A typical illusion shows a gap between how something "really is" and how something "appears to be", and this gap helps us understand the mental processing that lead to how something appears to be. Illusions are also useful for investigating artificial systems, and much research has examined whether computational models of perceptions fall prey to the same illusions as people. Here, I invert the standard use of perceptual illusions to examine basic processing errors in current vision language models. I present these models with illusory-illusions, neighbors of common illusions that should not elicit processing errors. These include such things as perfectly reasonable ducks, crooked lines that truly are crooked, circles that seem to have different sizes because they are, in fact, of different sizes, and so on. I show that many current vision language systems mistakenly see these illusion-illusions as illusions. I suggest that such failures are part of broader failures already discussed in the literature.