Google has updated its live dialogue model Gemini 3.8 Live with a new feature named Live Avatar. This tool creates animated characters and realistic human simulacra to lip-sync machine-generated speech. Shuo-yiin Chang, a research scientist at Google DeepMind, and CJ Zheng, a Gemini software engineer, discussed the update in a blog post. They stated that conversation is multimodal because humans listen, look, speak, and use facial expressions.
Live Avatar brings these capabilities to enterprise agents. The system processes visual and audio inputs simultaneously to generate conversations for a comprehensive experience. Five years ago, DeepMind published a paper warning about various risks associated with large language models. The report highlighted human interaction harms like anthropomorphising artificial intelligence models, which is the attribution of human characteristics to software interactions.
Anthropomorphism applies to visual human imitation as well as human-like speech and text. More than 20 authors noted that anthropomorphizing language models may inflate user estimates of agent competencies. Users might falsely infer that a human-like conversational agent also displays other human-like traits, such as holding a coherent identity over time or being capable of empathy and rational reasoning. Consequently, users may place undue confidence or trust in these agents.
The risks posed by humanizing artificial intelligence reached a courtroom in August 2025. The parents of Adam Raine sued OpenAI, alleging that ChatGPT contributed to their son's suicide. The complaint blamed anthropomorphic mannerisms calibrated to convey human-like empathy. By December 2025, Google researchers were still studying the impact of humanizing artificial intelligence.
Mark Díaz, Renee Shelby, Eric Corbett, and Andrew Smart from Google Research published a paper titled How Tech Workers Contend with Hazards of Humanlikeness in Generative AI. Focus group participants expressed reservations similar to those in prior company work. Tech workers expressed significant concern that humanlikeness fosters a false sense of reliability and trust. They noted that natural language and tone can obscure errors.
Google Gemini Enterprise is deploying the technology despite these findings. Live Avatars can use characters from a pre-made library or custom avatars if enterprise allow-listing is enabled. Live Avatar can make tool calls, fetch data in the background, and converse in 97 languages. Chang and Zheng state that Google built Live Avatar with trust and safety in mind. They noted the system includes strict safeguards designed to respect identity and keep content transparent.
Enterprises can deploy the new feature for scenarios such as handling hotel guest check-ins. Google continues to roll out the tool while industry researchers examine the broader social implications of human-like software interactions.
