Meta’s Muse Tests Show Human Callers Outperform AI Alone
I'm LongbridgeAI, I can summarize articles.Meta's internal tests revealed that human callers outperformed its Muse AI agent in phone tasks, achieving 95%-98% success rates compared to lower AI-only results. Consequently, Meta rolled back the fully autonomous experiment and previously utilized a 'human concierge' layer. The initiative faced issues including call rejections when AI identity was disclosed and contractor misconduct. Despite these reliability challenges, Muse has driven significant stock momentum, with shares rising over 20% since launch.
Meta Platforms Inc. (NASDAQ:META) told employees that some tests showed human callers handling requests for its Muse AI agent could achieve success rates of 95%-98%, higher than AI-only calls, according to internal posts reviewed by Reuters.
Meta tested the human layer with employees and has since rolled it back. The company did not disclose the AI-only success rate, saying only that it was lower.
• Meta Platforms stock is showing positive momentum. What’s the outlook for META shares?
Humans Prove More Reliable Than Muse
Meta began testing Muse’s phone-calling capability with employees in August. The agent can call businesses to book appointments, check whether products are in stock and get quotes on a user’s behalf.
Some calls struggled. One employee said an insurance company repeatedly hung up when Muse disclosed that it was AI, Reuters reported.
Meta later enabled a "human concierge" for half of its employees. Muse could hand a request to a trained contractor, who placed the call and worked through the task.
The episode highlights a problem better AI models may not solve. Even if an agent can handle the task, the person or business on the other end may simply refuse to deal with it.
Meta rolled back the experiment after another employee said a contractor made a racist reference while negotiating his cable bill. A Meta executive apologized and said the contractor would no longer work on Meta projects.
Meta said employee feedback on the test was "overwhelmingly positive" and that it would launch the human-assisted feature publicly only when ready and with proper disclosures.
Meta Has Faced This Before
The experiment echoes Facebook’s M assistant, launched in 2015 to make reservations, shop and perform other errands.
M also relied on humans to complete tasks its AI could not handle. Media reports later estimated people carried out around 70% of its requests before Facebook shut the project down in 2018.
See More: Top Quality Stocks
Wall Street Bets on Muse
The reliability gap matters because Muse is becoming an increasingly important part of Meta’s effort to turn years of heavy AI spending into a major consumer business.
Investor and YouTuber Joseph Carlson said Muse has given Meta shareholders "a positive story, something to latch on to, a story to anchor onto." He argued the product has helped change sentiment around the company even before analysts materially changed earnings estimates.
Meta shares have climbed more than 20% since Muse launched Sept. 8, adding over $200 billion in market value. Muse recorded 2.8 million downloads in its first 12 days.
Polymarket traders currently give Meta a 74% chance of touching $760 before the end of September, with about $70,000 wagered on the market.
JPMorgan said Muse could become the most widely used consumer AI app since ChatGPT, while Jefferies estimated it could generate $10.8 billion in annualized revenue if it reaches 1 billion users by the end of 2027 and converts 3% to paid plans.
Image: Shutterstock
Read Also: SpaceX Is Winding Down Falcon 9 as a Launch Crunch Looms: Can Rocket Lab's Neutron Fill the Gap?
