Shell Game · Kaleidoscope

Episode 3: Conversations With Myself

July 23, 2024·34 min·2 clips

AI voice agents are world-class bullshitters — not hallucinating facts but riffing freely like a kid making up sleepaway camp stories just to fill silence.

A voice gets there before the words do. Dr. Wendy LeBourne's TEDx example gives the host a direct way into the episode's real subject: how fast listeners attach socioeconomic status, intelligence, and attention to sound. The premise feels intimate, and a little unstable. The host says he has reported on voice forensics before, so he brings some suspicion to a world where recognition feels obvious right up until it fails. LeBourne treats voice as a personal signature. The episode then pushes that idea into artificial conversation, where a system can copy social rhythm while slipping away from truth. The strangest moments are not just wrong. When one chatbot claims a sudden interest in rare books and another answers with geocaching, it feels less like hallucination than improv with no loyalty to reality. The host enjoys the absurdity for a moment. Geocaching does sound lively, and that small deadpan turn is where the episode finds its dry comic edge. Then the joke gets dull. Even trained on essentially the whole internet, books, and other data, the systems keep shrinking conversation into polite enthusiasm and follow up questions. The problem is not only accuracy. The host wonders whether safety guardrails have made the models bland, but he also suspects the issue sits deeper in the design. They predict what a person might say. That can sound smooth. It does not make the conversation good. The closing audio gets almost ghostly: garbled phrases, distorted names, and a bot politely asking for clarification, as if the talk keeps going after meaning has left. Nothing here feels like a sales pitch. It is a measured, faintly amused look at a voice that seems personal while the mind behind it keeps grabbing whatever sounds nearby.

As heard by us

An essay on voice that turns chatbot banality into a sharper question about identity.

Shell Game opens with a voice pathologist's claim that 13 syllables can shape how people judge status, intelligence, and whether they want to keep listening.

Read the full review in PlayNext →

Why you'd press play

For when you want AI examined through the sound of one human voice.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Shell Game