'Creative Reading' Framework Challenges AI Summarizers' Design

Most AI reading tools, from summarizers to research assistants to paper-digesting chatbots, are built around the same quiet assumption: that reading is information extraction and faster extraction is better reading. A paper from researchers including Sophia Liu, Sarah Abowitz, Yijun Liu, Sarah Sterman, Shm Garanganao Almeda, and Max Kreminski, published on arXiv, puts a name to that assumption and argues it forecloses something important.

The authors call the dominant mode "reading to discard": "extracting the essential, actionable informational core of a document as quickly and efficiently as possible, so that the rest can be swiftly tossed aside." The critique is not that efficiency is useless, but that building tools exclusively around this mode embeds a design choice that most users never see made.

The alternative draws on a distinction between efferent reading, where readers focus on information "they will carry away from the text, such as facts, instructions, or claims," and aesthetic reading, which "emphasizes the lived-through experience of reading, including the feelings, images, and associations." The paper's central concept, "creative reading," treats the encounter between reader and text as something that can produce interpretations and readers who are themselves changed by the process, describing reading as creative "when the encounter between reader and text produces something that neither fully contains in advance."

To give designers something concrete to work with, the authors map reading augmentation systems onto two axes: transmission to transformation, and substituting reading to scaffolding it. Standard AI summarizers occupy the substituting-for-transmission quadrant. The concept they call "orienteering" captures what the other quadrants might offer: navigating complex terrain by following clues at intermediate steps, where "the route matters even when direct answers are available." The effort, in this framing, is not friction to be removed but part of what reading does to a reader.

What the paper does not give you is empirical evidence. There are no user studies here and no measured outcomes showing that transformative reading produces better scholarship or deeper understanding than efficient extraction. This is a design vocabulary and conceptual framework, not a tested system. That caveat matters for practitioners evaluating whether to act on it. For teams building reading tools in education, research, or knowledge management, though, the design space offers a useful diagnostic: where on those two axes does your product actually sit, and is that where you intended to land?