Listen "S4E14: AI Art Turing Test"
Episode Synopsis
And finally, here's an episode generated by Notebook LM plus a little test you can take yourself.
I was inspired by an article shared by Justė Kostikovaitė following Mantas Velykis' post about an AI-generated Coca-Cola ad and the diverse reactions it sparked. The article discussed the AI Turing Test for art (here’s the link: How Did You Do on the AI Art Turing Test?). Notebook LM generated an impressive podcast between two virtual hosts in just five minutes using the link. While it can be a bit annoying how synthetic hosts try to complement each other, the podcast itself is decent, covering most of what was in the article. Of course, I wouldn’t want to hand over the podcasting business to AI so easily. But for repurposing content for personal use, like listening on your way to work - why not?
Let me intrigue you or spoil the surprise: we can only tell with 60% accuracy whether something was created by a human or AI. Even people who work with AI tools daily make mistakes! By the way, after reading the article, I wanted to take the test myself, but unfortunately, the study had already ended.
So, I hope Scott Alexander (the author of Astral Codex Ten) won’t mind that I went ahead and created (with some AI help) an HTML version of this test. It’s just the test, not the study. No data is being collected. The original study with 11,000 participants sounds representative enough. Want to give it a try? http://milkplayground.com/art_turing_data/
I was inspired by an article shared by Justė Kostikovaitė following Mantas Velykis' post about an AI-generated Coca-Cola ad and the diverse reactions it sparked. The article discussed the AI Turing Test for art (here’s the link: How Did You Do on the AI Art Turing Test?). Notebook LM generated an impressive podcast between two virtual hosts in just five minutes using the link. While it can be a bit annoying how synthetic hosts try to complement each other, the podcast itself is decent, covering most of what was in the article. Of course, I wouldn’t want to hand over the podcasting business to AI so easily. But for repurposing content for personal use, like listening on your way to work - why not?
Let me intrigue you or spoil the surprise: we can only tell with 60% accuracy whether something was created by a human or AI. Even people who work with AI tools daily make mistakes! By the way, after reading the article, I wanted to take the test myself, but unfortunately, the study had already ended.
So, I hope Scott Alexander (the author of Astral Codex Ten) won’t mind that I went ahead and created (with some AI help) an HTML version of this test. It’s just the test, not the study. No data is being collected. The original study with 11,000 participants sounds representative enough. Want to give it a try? http://milkplayground.com/art_turing_data/
More episodes of the podcast ADC*talks
S5E14: Dizaino diena su Radio Vilnius
28/12/2025
S5E13: Dizaino apžvalga #4
27/12/2025
S5E12: Dizaino apžvalga #3
26/12/2025
S5E11: AI-crafted Videos with Salma Aboukar
17/12/2025
S5E10: In Barcelona with Fera and David
14/12/2025
S5E9: Dizaino apžvalga #2
12/09/2025
S5E8: Ko kūrybininkai neišmoko mokykloje?
09/07/2025
S5E6: Mental Health in the Creative Industry
23/06/2025
S5E5: ADC*E Leaks with Kevin Swanepoel
16/06/2025
ZARZA We are Zarza, the prestigious firm behind major projects in information technology.