If Writing Is Thinking

Central Question

The article explores the long-standing professional reality that few people thoroughly read long documents (memos, proposals, research papers). It then poses a critical question: What happens when generative AI, which both writes and summarizes content, is introduced into this environment? The author worries this will exacerbate the problem, creating a situation where not even the author has read or deeply understood the final text.

Key Arguments

• The Reading Problem is Universal: The author argues from personal experience as a senior executive that even critical strategy memos were rarely read in full. This issue extends beyond corporate tech to other fields.
• Examples Across Industries: The problem is evident in science, contributing to the reproducibility crisis as reviewers skim papers, and on Wall Street, where multi-billion dollar decisions are based on long analyst reports that are seldom read in their entirety.
• The AI Complication: AI introduces two new risks. First, when AI generates the text, the human “author” may not have a deep understanding of the content, meaning no one has truly read it. Second, when AI is used to summarize documents, the summaries can be lossy or even invent facts, leading to decisions based on flawed information.

Conclusion

Generative AI doesn’t solve the problem of people not reading; it threatens to make it worse. We may be entering a phase where critical information is created and consumed without deep human comprehension at any point in the process, amplifying the risk of miscommunication and poor decision-making.

Mentoring Question

Reflecting on your team’s workflow, how do you ensure that critical information in long documents is accurately understood? How might you adapt your processes to verify the integrity of content that is either generated or summarized by AI?

Source: https://hardcoresoftware.learningbyshipping.com/p/234-if-writing-is-thinking