AI Clones & The Future of Voice AI — With Evan Ratliff

Episode Summary

Executive Summary: Evan Ratliff discusses his podcast Shell Game, in which he cloned his voice, paired it with LLMs, and deployed the resulting voice agent into real-world conversations with scammers, friends, family, therapists, and professionals. The conversation explores how voice AI is rapidly improving, where it already feels convincing, and the social risks around trust, deception, therapy, work, and scams.

Main Topics: Voice AI as the next major AI interface (Priority: 5/5): The hosts frame voice as a rapidly growing AI format, citing explosive ChatGPT usage and executives like Mark Zuckerberg who see voice as a primary interaction layer. Shell Game experiment: cloning and deploying a voice agent (Priority: 5/5): Ratliff explains how he cloned his voice, connected it to LLMs, and used it to make and receive calls as a semi-autonomous agent. Scam-baiting and conversations with robocallers (Priority: 4/5): He initially aimed the bot at scam and telemarketing calls to explore how AI behaves in adversarial, real-world phone interactions. Therapy, intimacy, and the limits of AI support (Priority: 5/5): Ratliff tested the voice agent with AI therapists and a human therapist after feeding it a detailed personal biography, raising questions about usefulness, safety, and emotional effects. Human reactions, trust, and uncanny behavior (Priority: 5/5): The discussion focuses on how people responded to the bot—sometimes accepting it, sometimes becoming alarmed—and how easily AI can create confusion even when imperfect. Workplace deployment and automation risks (Priority: 4/5): They examine how AI agents may be used in meetings, interviews, and customer service to save time and money, while raising concerns about purpose, accountability, and output quality. Scams, voice cloning, and personal security (Priority: 5/5): The episode ends on the threat of personalized voice scams, including grandparent scams and impersonation using short voice samples from social media or video.

Key Arguments: Voice AI is becoming central to the AI experience because it is already good enough to be useful and will only get more convincing. Ratliff's project is not just a prank; it is a way to study how AI changes trust, relationships, and daily interactions when it enters society. AI therapists may be marketed faster than science can evaluate them, especially for vulnerable users seeking loneliness relief or mental-health support. Human therapists outperform AI by being flexible and context-aware, while AI tends to remix inputs and repeat generalized responses. Even when people suspect the voice is AI, many continue the conversation because it is socially awkward to accuse someone of being artificial. AI voice tools will likely spread in work and customer-service contexts because companies have incentives to reduce labor costs despite flaws. Voice cloning supercharges scam potential, making it easy to impersonate relatives or trusted figures and extract money or information. Ratliff argues that people need practical safeguards now—code words, verification routines, and public awareness—because the scam wave has already begun.

Data Points: ChatGPT users: 100 million to 300 million - The host cites OpenAI signups rising sharply after advanced voice AI became available. Monthly web visits: 2 billion to 4 billion - The host says ChatGPT web traffic doubled after a period of flattening. Podcast length: 6 episodes - The host recommends Shell Game as a concise, easy-to-binge series. Personal biography fed to the bot: 8,000 words - Ratliff gave the system a magazine-length biography of himself to inform therapy conversations. Scam calls received: 30-40 per day - Ratliff says his phone line still receives a large volume of scam calls. Study count: 1 controlled study - Ratliff says he found only one controlled study on voice therapy. Voice sample needed for cloning: A few seconds - He notes that a short sample from video or social media can be enough to clone a voice. Zip code used by the bot: 90210 - He jokes that his bot commonly uses this as a fake zip code.

Pivotal Quotes: "I think people need to hear, like people will want to hear this." — Evan Ratliff: Explaining why he turned the voice-AI experiment into a podcast. "I think the market is going to dictate that people are going to use this technology, even if it remains flawed." — Evan Ratliff: Discussing why voice AI will spread in customer service, drive-thrus, and other labor-saving settings. "This is the greatest scamming technology that has ever been invented." — Evan Ratliff: Warning about the threat of voice cloning and personalized scams.

Implications: Voice AI is moving from novelty to infrastructure. Listeners should expect more AI phone agents, therapy bots, workplace avatars, and scams—and adopt verification habits now before synthetic voices become routine.