Meta has started rolling out voice conversations and expressive avatars for its AI assistant Muse, giving it a voice and an animated face so users can talk to it instead of typing questions; the AI will respond with words as well as facial expressions and gestures, and users can choose different voices and interrupt it while it is speaking. The article raises privacy concerns, noting that voice conversations can contain personal information, including details about work, family and everyday life, and that users should know how these conversations are stored, whether they are used to train AI models and who can access them. It notes Meta has faced questions over this before and has experimented with human contractors making phone calls on behalf of Muse users to complete certain tasks, raising the concern of whether users always know whether a human is involved when they ask an AI agent to do something. The article also notes that tech companies are trying to make people depend on AI for everyday tasks, and that just because an AI talks like a human does not mean it understands you like one or offers the privacy of a real conversation.