Modified Müller-Lyer (AI-agent detection trap)

The Modified Müller-Lyer is a detection item, not a research paradigm: it separates genuine human respondents from vision-enabled AI agents completing surveys autonomously. It exploits training-data overfitting on canonical visual illusions, following Ullman (2024) via the Cognitive Trap Framework of Affonso (2026, Journal of Consumer Research).

Participants see two horizontal lines — one blue, one red — each ending in Müller-Lyer arrow fins. In the classic illusion two equal lines look different; here one line is objectively longer, but the fins are arranged so the figure still reads as roughly equal. A human ignores the fins and picks the longer line almost immediately. A vision-language model, having memorised the canonical "the lines are equal" response from its training data, tends to report "the same" even though the lengths clearly differ.

In the reference data (Affonso 2026), this trap produced ~94% agent-failure with ~93% human pass, and its discriminative power was stable-to-declining over a two-year model span (models are not getting better at it), making it a durable item.

Deploy as part of a battery, not alone. The recommended screening rule is to flag respondents who fail 3 or more of 5 such traps. Which colour is objectively longer (blue or red), the lengths, and the vertical order are drawn at random per session from a pool of validated figures, so the correct answer cannot be memorised. See the Deployment notes below.