Until about five years ago, whenever I interviewed someone for a writing project, I would be faced with the gruelling prospect of having to transcribe the recorded conversation, often running to many hours. This was exactly the sort of clerical tedium for which, in becoming a writer, I imagined I’d forgone financial security and societal respectability to avoid. The process was so grimly repetitive that there was usually a point, somewhere around the fourth or fifth hour of transcription, where I would resolve to give up writing entirely and get a proper job.
But then I started using an app that employed machine learning to convert recorded speech into text. The thing was surprisingly accurate and efficient, and I was impressed by its ability to automate a process I found incredibly dull and time-consuming. It wasn’t perfect; it struggled with strong accents, and muffled words, and it often made completely stupid errors for no obvious reason – but then, so did I, and it took me a lot longer to make them.
[ Meta wants to use your Facebook and Instagram data to train AI. Here’s how you can you stop it ]
After I finished the reporting for my last book, almost two years ago, I was doing other kinds of writing, and had no need for the transcription software. Then, a couple of weeks back, I started work on a long magazine piece, involving hours of interviews, and I began using it again. And I was, frankly, pretty amazed by how much the technology had improved in the time I’d been away. The whole “artificial intelligence” aspect of the thing had previously........