Microsoft Teases Lifelike Avatar AI Technology but Withholds Release Date by Glenn CHAPMAN
Microsoft researchers have unveiled a cutting-edge artificial intelligence tool capable of generating remarkably realistic human avatars, yet they have refrained from specifying a release schedule. Citing concerns about the potential proliferation of deceptive deep fake content, they have opted not to disclose when the technology will be made available to the public.
Dubbed VASA-1 (visual affective skills), this AI model can produce animated videos of individuals conversing, complete with synchronized lip movements, using only a single image and a speech audio clip. However, disinformation researchers are apprehensive about the widespread misuse of AI-powered applications for creating deceptive images, videos, and audio clips, particularly in a critical election year.
“We are firmly against any activities that result in misleading or harmful representations of real individuals,” emphasized the authors of the VASA-1 report, released this week by Microsoft Research Asia. “Our commitment is to develop AI in a responsible manner, with the overarching goal of promoting human well-being.”
The Microsoft researchers highlighted the tool’s capability to capture a diverse range of facial expressions and natural head movements, enabling real-time interactions with lifelike avatars that mimic human conversational behaviors. VASA can handle various inputs, including artistic photos, songs, and non-English speech.
Despite its potential benefits, such as offering virtual teaching or therapeutic support, the technology is not intended for creating deceptive content, according to the researchers. They acknowledged that VASA-generated videos still exhibit “artifacts” that reveal their AI origin.
Ben Werdmuller, technology lead at ProPublica, expressed curiosity about the tool’s real-world applications, particularly in scenarios such as representing individuals in virtual meetings. Meanwhile, OpenAI recently introduced a voice-cloning tool called “Voice Engine,” capable of replicating someone’s speech based on a 15-second audio sample. However, OpenAI is proceeding cautiously with its release due to concerns about potential misuse.
Earlier this year, a consultant working for a lesser-known Democratic presidential candidate admitted to orchestrating a robocall impersonating Joe Biden, aiming to raise awareness about the dangers of AI. The incident underscored fears of an influx of AI-driven deep fake disinformation campaigns during the 2024 White House race.
