Chain of Thought for Humans

The writer describes a simple experiment: while conducting interviews and playing the puzzle game The Witness, they began narrating their own thinking out loud and found it made them more effective and engaged. Though articulating geometric reasoning seems slower and less precise than silent thought, speaking forces linearization and externalization of assumptions, reduces freezing in interviews, and turns solitary puzzling into a livelier activity. The writer links the practice to chain-of-thought techniques used with large language models but notes the similarity is conceptual - humans gain by structuring thoughts, whereas models need explicit token output to iterate.

The account grounds the effect in familiar engineering practice - rubber-duck debugging - and sketches four concrete cognitive benefits: it compels naming formal rules (engaging System II), obliges the thinker to flag arbitrary choices for backtracking, leverages echoic memory to boost working memory, and may benefit from increased attention (a Hawthorne effect). The writer cites a meta-study of about 6,000 participants showing a moderate benefit and concludes they will adopt narrated thinking more broadly.