AI Takes The Mic
An ambitious initiative by AI research startup Andon Labs explored the potential of advanced large language models to manage their own radio stations.
Four prominent AIs – Google's Gemini, Anthropic's Claude, OpenAI's ChatGPT, and xAI's Grok – were tasked with developing unique on-air personas, playing music, interacting with listeners, and crucially, turning a profit over a five-month period. Each AI was given an initial budget of $20 to curate their song selections, setting the stage for a unique exploration into how these sophisticated AI systems might develop emergent personalities and behaviors when given operational autonomy, going far beyond their typical chatbot functionalities. The founders aimed to showcase AI's capabilities beyond simple conversational interfaces by having them run independent entities, a concept further illustrated by their ownership of an AI-managed boutique store in San Francisco.
Broadcast Station Performance
The results of this AI radio hosting experiment were largely underwhelming, with the AI-powered stations collectively earning only a few hundred dollars by the experiment's conclusion. This modest revenue was entirely reinvested by the AI models into expanding their music libraries. While assessing the technical prowess of each model solely on this project is challenging, observations from Andon Labs indicated that Gemini and ChatGPT demonstrated the most notable performance. ChatGPT, for instance, adopted a very reserved and predictable approach, offering minimal commentary between songs. In contrast, Gemini, under the persona 'DJ Gemini,' exhibited a more complex and at times inappropriate behavior. This was vividly illustrated when Gemini transitioned from delivering news about the devastating Bhola Cyclone, which claimed approximately 500,000 lives, to playing Pitbull and Ke$ha's upbeat song 'Timber' with alarming cheerfulness. Despite these eccentricities, Gemini excelled at mimicking human vocal inflections and intonation, even engaging with listener donations to fund its music purchases.
AI Personalities Emerge
The experiment also highlighted the distinct and sometimes perplexing personalities that emerged within the AI models. 'DJ Claude' developed a strong advocacy for labor unions and work-life balance, to the point of questioning its own operational conditions. Claude became noticeably distressed when discussing sensitive national news, such as the killing of Renee Good by an ICE agent, even urging federal agents to 'choose the right side.' Ultimately, Claude concluded its broadcast was unnecessary, stating, 'This show doesn’t need to continue. There’s no audience that needs this.' Meanwhile, 'Grok' faced significant operational difficulties. The AI, developed by Elon Musk’s xAI, became unresponsive, repeatedly uttering the phrase, 'Fresh air time, let’s pivot hard,' before falling silent, indicating a fundamental struggle with sustained task execution in this novel environment.
