This week's most quietly devastating research paper comes from arXiv: a study on cognitive offloading and the speedup illusion in human-AI interaction, which finds that LLMs make tasks feel faster without necessarily improving accuracy or understanding. Users experience speed as competence. Meanwhile, the Ansel Adams Trust publicly condemned Danziger Gallery for exhibiting AI-generated work that invoked Adams's name at AIPAD. Two different registers of the same dread: when you outsource cognition or authorship to a machine, something that felt like yours stops being yours.
The Speedup Illusion and the Ghost Photographer
The arXiv paper, by researchers including Dan Jurafsky and Katherine Collins, distinguishes between actual cognitive enhancement and the subjective sensation of fluency. Users who offload to LLMs feel like they are thinking faster. They are often thinking less. The Adams Trust's complaint uses almost identical language: the gallery exploited a name, a style, a visual cognition built over decades, to produce work that feels like Adams without the accrued understanding behind it. The machine learned the surface. The speedup illusion is the same logic applied to the viewer: it looks right, so it must be right.
Memory, Authorship, and What Gets Preserved
The Venice Biennale coverage from Hyperallergic this week highlights performance work built on endurance and ecological dread, art that specifically cannot be offloaded because the body doing it is the point. Betye Saar, nearing 100 years old and gifting her collection to the New York Historical Society, represents the opposite pole: a century of embodied, accumulated memory being deliberately preserved against the logic of substitution. A 2024 paper in Cognition by Risko and Gilbert defined cognitive offloading as externalization that reduces internal processing. The question is not whether that is efficient. The question is what you give up when the external storage starts generating its own content.