The article “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task“ offers an elaborate empirical study comparing how students write essays with three different modes of support: no tools, traditional search engines, and Large Language Models (LLMs). The authors track neural activity using EEG, analyze the resulting text with NLP tools, and conduct post-task interviews. Their conclusion is that using LLMs weakens memory, reduces ownership, and lowers cognitive engagement. They refer to this condition as cognitive debt.

The data is valuable. The interpretation is not.

The entire study is structured around a deficit model. If an LLM reduces internal cognitive effort, the authors assume it must therefore reduce learning, retention, and agency. They treat the LLM as an external tool and assess it with pre-AI rubrics: memory recall, essay quoting, structure, ownership. This would make sense if writing still consisted of a solitary subject composing from memory. But that world no longer exists.

What the study documents is not a decline. It is a shift in the medium of writing—from literature to something else. But the authors remain attached to a legacy cognitive paradigm and misinterpret this shift as degradation. The writing they observe in the LLM condition is less quotable because it is less linear, less mnemonic, and less individuated. But that does not mean it is less intelligent. It means it emerges from a hybrid system.

The study’s fundamental flaw is conceptual. It does not ask what kind of medium LLMs represents, nor what kind of authorship it enables. It remains stuck in the vocabulary of past practices, applying analog measures to a computational phenomenon. That is why its conclusions fall short. A different framework is required—one that does not assess human-machine writing with the metrics of pre-machine writing.

In a follow-up post, I introduce such a framework. I call it Literatronica. It is not a metaphor. It is a name for a mode of composition that is already here, one that neither the study nor most of its commentators have adequately recognized.

More about this idea here: https://biomathematicus.me/writing-has-left-the-page/