Working memory demands of the mental imagery object reconstruction task

Determine to what extent solving the compositional mental imagery object reconstruction task requires human working memory, specifically whether successful performance depends on maintaining multiple components across steps or can be achieved by sequentially updating a single imagined scene that retains only the previous scene and newly introduced information.

Background

The authors note that humans may solve iterative imagery instructions by updating an imagined scene at each step, potentially reducing the need to retain many discrete elements simultaneously in working memory.

Clarifying the working memory demands of the task in humans would contextualize comparisons with LLMs and inform the design of more challenging instruction sets that isolate memory requirements from purely sequential updating.