As I began working on a book on intergenerational trauma, I interviewed descendants of Holocaust survivors about their experiences growing up with survivors. I recorded the interviews on Zoom and used AI software called Fathom to transcribe them.

What I quickly discovered was that machines cannot pick up all the subtleties of the English language, accents, and how to spell names properly. Background noise impacts an AI transcription. As a result, I have been constantly going back to the recording to double-check what was said. It’s become time-consuming and frustrating. I also understand the context of what people are saying in their interviews, something the machines fail to do.

A good writer friend, Ellie Leonard, has a transcription business called Red Pencil Transcripts. Her clients have tried to replace her services but keep running back because Ellie does things that a computer does not do, such as checking the spelling of names and acronyms to get them right. The New York Times published an article on AI featuring Ellie several years ago. The points made in the article are still valid in 2024. Machines can try to duplicate what we do, but like self-driving cars, they can easily be confused.