I think you should almost never use AI to write

The ability to think deeply while writing is now a unique human advantage that AI cannot replicate.

Author Erich Grunewald argues against using AI for substantive writing, including blog posts, reports, and emails. He believes the act of writing is intrinsically linked to the thinking process, suggesting that AI-generated text bypasses this crucial cognitive step. Grunewald posits that AI writing often contains subtle inaccuracies or vagueness that are difficult to detect. He also raises ethical concerns, stating that using AI for writing without disclosure is misleading and disrespectful to the audience. While acknowledging AI's utility in other aspects of research and editing, he specifically targets the creation of original text as an area where human involvement is irreplaceable. Grunewald's stance is based on current AI capabilities and anticipates similar limitations in near-future models, though he allows for the possibility of future AI systems that might genuinely master the writing process.

Grunewald's core argument is that writing is not merely output, but a fundamental part of cognition where ideas are formed, refined, and tested. By delegating writing to AI, individuals risk losing this critical thinking development. This perspective challenges the prevailing notion of AI as a productivity enhancer, reframing it instead as a potential detractor from intellectual growth and the creation of truly original thought. The subtle inaccuracies highlighted also point to a growing concern about the 'hallucination' or 'confabulation' problem in LLMs, suggesting that even with extensive prompting, AI outputs may lack the precision and nuance required for substantive communication.

The implications extend to how we value and assess written work. If AI-generated text becomes indistinguishable from human-authored content without disclosure, it erodes trust and the perceived authenticity of communication. Grunewald's ethical stance on labeling AI-assisted writing underscores a potential future debate about intellectual property, authorship, and the integrity of information. This also suggests that industries relying heavily on original written content may need to establish new standards for transparency and human oversight to maintain credibility and foster genuine intellectual engagement.