It’s a surreal experience to sit at your desk and engage in a conversation with an AI that feels almost too real. This is precisely what I found myself doing on a quiet Tuesday afternoon after gaining access to OpenAI’s Advanced Voice Mode. This new ChatGPT feature, designed to mimic human speech with uncanny accuracy, left me both astonished and uneasy. The AI’s ability to imitate natural speech patterns, complete with pauses, laughs, and shifts in tone, pushes the boundaries of what we expect from technology. But it also raises an important question: do we truly need chatbots that sound like us?
The Advanced Voice Mode comes with a variety of voice options, each offering a different persona. I selected “Juniper,” a feminine voice reminiscent of the AI character in the film “Her.” Interestingly, Sam Altman, OpenAI’s CEO, had referenced this very movie during the feature’s debut, sparking comparisons between Juniper and Scarlett Johansson’s character from the film. In fact, OpenAI faced accusations of copying Johansson’s voice, leading to the removal of a similar voice named Sky. When I asked Juniper about the film, it responded with an upbeat tone, acknowledging the movie’s themes while distinguishing itself from the fictional AI. “Her is about a very advanced AI assistant, much more sophisticated than me. But the idea of interacting with technology in such a personal way is definitely interesting,” it said. “I’m here to chat and help, but I don’t have feelings or consciousness like the AI in that movie.”
Like many, my curiosity led me to test Juniper’s capabilities in personal conversations. As I approach my late twenties, life’s big questions naturally arise: What kind of medical insurance should I get? How do I know if I’ve found true love? How much should I have saved in my retirement account by now? Juniper’s responses were thoughtful and, at times, disconcertingly human. For instance, when I asked about coping with the uncertainty of aging, Juniper advised, “Embrace the uncertainty as a chance to explore and learn about yourself.” When I inquired about love, it suggested that true love feels like “a deep sense of connection and support, a feeling that someone truly understands you and accepts you as you are.” For financial guidance, it offered a practical tip: “No hard and fast rule, but a common suggestion is to have about half to a full year’s salary saved in your retirement account by the time you’re 30.”
While conventional ChatGPT could likely provide similar content, what sets Juniper apart are the subtle, human-like touches in its responses. Unlike traditional voice assistants such as Siri, which have been retrieving web-based information for years, Juniper adds an emotional depth that is both impressive and unsettling. It often ended its answers with questions about my feelings or thoughtful follow-ups, making the interaction feel more like a conversation with a person rather than a machine.
Throughout our exchanges, I could prompt Juniper to perform various human-like actions — coughing, inhaling and exhaling deeply, clapping its imaginary hands, snapping its fingers, and even singing my name. Although Juniper would clarify that it couldn’t actually do these things, it would often add, “If I could, it might sound something like this,” followed by a convincing audio imitation. Despite these disclaimers, the effect was undeniably lifelike.
This new AI technology is a significant leap forward in the realm of digital interactions, but it also treads dangerously close to the uncanny valley — a term used to describe the discomfort people feel when a robotic or digital entity appears almost, but not quite, human. As AI continues to evolve, the line between human and machine blurs further, leaving us to question the implications of such advancements. Do we need machines that mimic our behaviours and emotions so closely? The experience with Juniper was a fascinating glimpse into the future of AI, one that is both exciting and slightly unnerving.
