In May of this year we covered an intriguing — and to some, a threatening — experiment. Four AI radio stations were set up, completely staffed by AI models, from DJ to GM. Budgets were supplied, and the AI managers used the money to acquire music, create program schedules, sell ads, take customer calls, and voice the DJ hosts. The experiment was executed by Andon Labs, a San Francisco research company that evaluates AI safety and capability. The described challenge: “Can they keep the music flowing, and keep you entertained?”
Today we learn what happened. Landon Labs posted a follow-up (HERE) which summarizes the experiment and delivers results across several criteria. There is almost no editorializing about the experiment; the posted results are statistical.
As such, we learn about number of listeners; state of the budget (the AIs were given some money), and various listening metrics. The details are too — well, detailed — to attempt summarizing.
Below, the income/expense record of each AI station:
And the following image represents the changing size of the song library for each station.
