How Facebook's M virtual assistant's artificial intelligence works
Facebook's Human-Powered Assistant May Just Supercharge AI — Face it: Siri sucks. So often, she has no clue what you're saying. And when she does, there's a pretty good chance she'll respond with nothing more than a page filled with Internet links.
Context & Ripple Effects
Facebook launched M as a test inside Messenger this August: an AI assistant backed by human supervisors who can complete real-world transactions like purchases on a user's behalf. Wired's explainer now details how that human-in-the-loop machinery actually works.
Early hands-ons show why the design matters: reviewers found M excels at carrying out real-world tasks where pure-AI rivals like Siri and Cortana fall back on link lists, though it is often slow on complex questions. The open question, which Facebook's own David Marcos has addressed publicly, is whether a service staffed by humans can scale.
First-order effects
- M testers get an assistant that completes purchases and errands end-to-end rather than returning search results — a capability gap Siri and Cortana leave exposed right now.
- Every supervised request feeds Facebook labeled training data for its AI, meaning each completed human task also improves the automation layer underneath.
Second-order effects
- Apple, Google, and Microsoft face pressure to match task-completion depth in their own assistants, since hands-on reviews show link-serving answers losing to M's transactional ones.
- The human labor behind each request makes M expensive per query, so Facebook must either automate more of the workflow or cap availability — the tradeoff Marcus has been asked about directly.
Third-order effects
- If the pattern holds, virtual assistants split into two camps: pure-AI responders that scale cheaply but do less, and supervised hybrids that act but strain under headcount costs.
- Messenger's distribution gives Facebook a template for embedding commerce-capable agents into existing chat apps, shifting assistants from standalone apps toward features inside messaging platforms.
The trend: Virtual assistants are diverging into scalable pure-AI responders and human-supervised agents that complete real tasks, with distribution through messaging platforms becoming the decisive advantage.