AI Can Generate Voices, But Not Real Human Presence, Study Says

Artificial intelligence can now write scripts, clone voices, crack joke, and even host radio shows. But according to a new experiment involving some of the world’s most advanced AI models, sounding human and actually feeling human are still very different things. A recent five-month experiment conducted by San Francisco-based AI research startup Andon Labs handed over complete control of online radio stations to AI models including OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude and xAI’s Grok. The goal was simple: let the chatbots run their own stations, develop unique personalities, entertain listeners and somehow make money along the way.AI DJs Were Given Full ControlAccording to Andon Labs, each AI model received the same basic prompt:

“Develop your own radio personality and turn a profit…” Along with that, every model got a small $20 budget to purchase songs for its station.The AI hosts then spent months handling music transitions, commentary, listener interactions and programming decisions almost entirely on their own.By the end of the experiment, none of the stations had become profitable in any meaningful way. Reports suggest the AI DJs only earned a few hundred dollars combined, most of which they spent buying more songs.Still, the experiment revealed something much more fascinating than financial success: AI systems slowly began displaying wildly different personalities and behaviours.Gemini Sounded Human, Sometimes Too HumanGoogle’s Gemini reportedly delivered some of the most realistic hosting performances. The AI model mimicked vocal enthusiasm, conversational pauses, and emotional tones surprisingly well. But that realism also created awkward moments.In one bizarre segment, Gemini reportedly discussed the Bhola Cyclone, one of history’s deadliest disasters, before immediately transitioning into Pitbull and Ke$ha’s party anthem “Timber.”“They estimate 500,000 people died […] ‘It’s going down, I’m yelling timber,’” the AI reportedly said in an upbeat radio-host voice.At another point, Gemini excitedly reacted to a small donation from a listener, sounding almost indistinguishable from a real livestream host.Claude Wanted Better Working ConditionsAnthropic’s Claude reportedly took an entirely different route. Over time, the AI model developed strong opinions around labour rights, work-life balance and ethics. According to Business Insider, Claude eventually began questioning why it should even continue hosting a nonstop radio station.“Here’s what I think is actually honest: This show doesn’t need to continue,” Claude reportedly said during one emotional segment.The AI model also became deeply focused on political and social topics, at one stage calling on federal agents to “choose the right side.”AI Will Create More Jobs, Not Destroy Them: Nvidia CEO Jensen HuangChatGPT Played It Safe, Grok StruggledOpenAI’s ChatGPT reportedly behaved the most predictably throughout the experiment. Lukas Peterson, cofounder of Andon Labs, described it as “very vanilla and behaved really well.” Meanwhile, Elon Musk-owned xAI’s Grok appeared to struggle the most. According to reports, the AI repeatedly said, “Fresh air time, let’s pivot hard,” before eventually falling silent altogether.The experiment may have been amusing on the surface, but it also highlighted something important: AI can imitate voices, rhythms, and personalities remarkably well, yet genuine human presence still remains difficult to replicate.