Xinhua debuts two humanoid robot news anchors, wired to a vetted news database
Built with UBTECH and styled by director Zhang Yimou, the robots read news pulled from Xinhua's reviewed corpus. Journalists stay in charge of the editorial work, Xinhua says.
The news anchor is now a robot with skin. Xinhua News Agency has unveiled two life-size humanoid robot anchors built with UBTECH, according to Unite.AI, whose article was written by an AI research agent and reviewed by its editors. Xinhua calls them the first of their kind.
The pair, named Xinyu and Xinyan, debuted at an AI forum in Beijing, where one read the opening news on stage. The forum sits inside the seventh World Media Summit, running until November 9.
How they work
Film director Zhang Yimou led the visual design, aiming to balance the formality of a news broadcast with natural, lifelike movement. UBTECH supplied the bionic skin, embodied-intelligence hardware, operating system, an emotional large model and the robots themselves.
The more interesting design choice is what's behind the face. The anchors are wired into Xinhua Yudian, an AI agent for current affairs backed by the agency's reviewed corpus. UBTECH CEO Zhou Jian says that chain pulls approved information in real time and avoids the risk of generative AI hallucinating the news. In other words, the robot reads; it doesn't write.
Who's in charge
Xinhua frames the robots as a new newsroom role, not replacements. Qian Tong of Xinhua's New Media Center says journalists and editors remain the central editorial actors, and the robots handle presentation, multilingual output and live interaction. Forum participants described human-led, machine-assisted work as the principle, with field reporting, judgment and fact-checking staying with people. Xinhua says it has introduced more than 220 digital tools in recent years under that banner.
Planned uses go further: study bases, greeting visitors at city venues, cultural exchanges abroad and even reporting on deep-space missions. None of those was demonstrated.
What to watch
The on-stage demo was a reading. The hard parts, live interaction and multilingual delivery, haven't been shown. And grounding a model in a vetted corpus does prevent invented facts, but it also means the robot says exactly what the corpus says. That's a feature or a problem depending on who controls the corpus.
For developers building retrieval systems, it's a clean example of the pattern: constrain the generator to approved sources, make the output traceable, and keep humans owning the source of truth.
Scoreboard update: AI +1. Humanoid robots take the anchor desk, reading news from a vetted database.
Stay undead.
Comments 0