Writing as Computation: A Turing Machine for Human Thought
Writing functions as a computational amplifier for human cognition. Drawing on Manuel Blum’s technical insight: without written externalization, thinking hits the limits of a finite automaton; with pen and paper, you gain Turing machine-level capability. Language allowed humans to share thoughts, and writing let us preserve and build on them across time. Large language models inherit substantial intelligence directly from text, suggesting they partly embody human thought already encoded in language.
The mechanism mirrors out-of-core computing: writing offloads cognition onto a medium, letting you manipulate larger conceptual problems than working memory allows. When drafting a paper review, you discover mid-writing that your initial understanding was shallow. Connections surface only through the act of composition itself. Written work becomes a persistent memory prosthetic; rereading your own writing reliably reconstructs ideas far better than mental notes, which evaporate. This is why researchers benefit from treating their drafting process as a lab notebook.
LLMs create a peculiar inversion: they generate sophisticated text without any cognition behind it, potentially making personal writing seem unnecessary. But this actually flips writing’s status. As AI handles surface-level text generation, maintaining your own writing practice becomes a competitive edge—you keep thinking while others outsource it. The practical approach: write messily without self-editing until you have material to work with, then shift to ruthless cutting and shaping. Keep prose direct; complexity masks fuzzy thinking. Write for intrinsic motivation rather than audience approval, since that sustains genuine cognitive work.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.