In September, I got a very weird email. A producer on a National Public Radio show—The Pulse, from WHYY in Philadelphia—wanted to interview me.

I've done radio hits before, but always in relation to my professional work, and about politics. This wasn't that. Instead, this turned into my strangest interview to date, and a pretty delightful show about the mysteries of technology.

See, sometime over the summer, producers on The Pulse started noticing my name in their transcripts. My full name, Aaron Ross Powell, would just show up in the middle of conversations, and conversations that had nothing to do with me, or any of the topics I work on.

The transcripts were generated by an AI tool, and the AI, for opaque reasons, decided to "hear" my name in the gaps of silence you get in the audio of any conversation, when people pause to take a breath or think of what to say or trail off.

The Pulse wanted to know why, and what, if anything, I'd done to cause it. And they thought this could make for a fun Halloween episode.

It's out now. You can listen here, or on Spotify, or on Apple Podcasts. Or wherever you get your podcasts.

It's a lot of fun. And we never did come up with good answers. I'm apparently in the AI training data. But so are lots of people. My name does appear in tons of podcasts. But the same is true of lots of people. The underlying model NPR's transcript tool uses is a version of OpenAI's Whisper. But to my knowledge, this isn't happening with every application using Whisper, just NPR's. It's perplexing. I am, as they call me in the episode, the ghost in the machine.

Which I guess I'll take. If the robot gods want me to be the single source of truth, it's a degree of power I didn't ask for, but I promise to use responsibly.


If any AI and machine learning experts have ideas on why I'm showing up in AI-generated transcripts, I'd love help getting to the bottom of this mystery.