Duo ex Machina
Conjuring up virtual podcasters aiming for authentic insights
My posts typically consist of dialogues with AI language models. Today’s installment takes that concept to another level, presenting virtual podcast episodes based on a few past dialogues. Google’s NotebookLM generated the episodes below, automatically preparing scripts and spoken audio of two AI co-hosts of a “Deep Dive” podcast engaged in discussions about the past posts.
These are one-shot creations, leaving no opportunity for manual revision after automated generation. While these conversations generally remain remarkably well grounded in the content of the subject posts and quite coherent in their delivery, there are instances of hallucinations, mangling of the source material, and meandering (just as in real podcasts, one might quip).
The episodes range from 8 to 15 minutes in length, while Substack’s estimated read-aloud times for the underlying source posts are 15 to 24 minutes. In this respect, the episodes represent a condensation of the source material by a modest factor.
Give them a listen and compare to the posts they discuss. To what extent do the episodes obscure, enhance, or substitute for the source material? How often does the illusion of two human co-hosts break? Where is there unintentional humor, such as when the virtual speakers present themselves as real and distinct from AI models?
Enjoy!
“Deep Dive” into Wired for Selfhood: Discovering identities with “13 Questions” from NewPhilosopher.com:
“Deep Dive” into Epicurean Algorithm: Estimating an ancient philosopher's responses to a modern survey:
“Deep Dive” into Processing Glitches: Seeking the boundary between real and surreal:
“Deep Dive” into Philosophy Bytes: Artificial paths to real connections:
(Note: I included this audio in the preface to the original post.)