Picture a newsreader you can walk up to, ask a question, and get an answer from, face to face. That is what Xinhua, China’s state news agency, put on stage on Thursday, 5 November, at the World Media Summit in Beijing: two life-size humanoid robots, one male and one female, presented as “bionic humanoid robot news anchors”. The female anchor is called Xinyan. The real question is less how human they look and more where their answers come from.
What Xinhua presented, and who built it
The robots were developed jointly by Xinhua and UBTech Robotics, one of China’s best-known humanoid makers. UBTech handled the full manufacturing process and supplied the bionic skin, the operating system, an “emotional” large language model and what it calls embodied intelligence hardware, meaning the physical body, sensors and motors an AI system uses to act in the world. Xinhua is reported to hold complete property rights to the robots.
The look came from an unusual place. Film director Zhang Yimou served as chief image designer. His team worked on the faces as carefully as on a film set: “We refine facial expressions, eye movements, lip shapes … to achieve a calm tone, poise and eye contact,” he said. That choice tells you what the project is really about. A news anchor is not only a voice reading text; it is a face the audience is asked to trust.
How the robots answer questions
Large language models, the systems behind today’s chatbots, are known to “hallucinate”: they produce quotes, numbers or events that sound plausible but are invented. For a newsreader, that is the worst possible habit. Xinhua’s answer is to connect both anchors to Xinhua Yudian, an AI news platform and agent it launched in June. Instead of improvising from whatever a model absorbed in training, the robots draw on the agency’s own current-affairs material.
According to Interesting Engineering, UBTech founder and chairman Zhou Jian described a “brain–expression–control” system that retrieves vetted news in real time to reduce hallucinations. Xinhua says the arrangement “makes the content traceable, secure and controllable”. Technically, this is a familiar pattern called retrieval: the model looks up approved documents first and phrases its answer from them.
How human do they really look?
Now watch the movement. According to The Canberra Times (Australian Associated Press), which carried the report from the summit, the heads tilt and turn at regular intervals, the eyes dart around and blink, and up close you can see a camera sitting where a human pupil would be. So the “eye contact” is real in one sense: there is a lens looking at you. The face sells calm and poise; the motion still reads as a machine running a routine.
This is not Xinhua’s first artificial presenter. In November 2018 it released an AI anchor built from digital composites of human presenters with a synthesised voice, a screen-only figure with visible glitches and limited expressions. The new step is the combination: a physical body in the room, linked to a news AI that can respond to questions live.
Will they replace journalists?
Xinhua says no. The robots are meant to help present information, deliver content in several languages and interact with audiences in real time, but not to write news on their own. Qian Tong, director of Xinhua’s New Media Center, said journalists and editors “will remain the central editorial actors in the newsroom”. A Xinhua vice president told the summit that harnessing AI is about “empowering, not replacing” journalists, and Zhang Yimou framed the robots as assistants “adding new experiences”.
Why media experts are sceptical
Jock Cheetham, a senior news and media lecturer at Charles Sturt University in Australia, does not expect robot newsreaders on Australian TV any time soon. Established broadcasters, he argues, would risk eroding trust with presenters that might seem fake, cheap or odd; online commentators and influencers are more likely to try AI newsreaders first. He even expects real human presenters to become a mark of quality, as authenticity grows more valuable. AAP notes that some Australian radio networks already use bot voices for bulletins, so the debate is about where the line sits, not whether AI enters the newsroom.
Our assessment: the library matters more than the face
The facts reported so far are about design, architecture and roles. What is missing is just as telling: neither report includes an error rate, response times or the list of languages. Our reading is that the retrieval design is a sensible fix for one problem and a shift of responsibility for another. Pulling answers from checked material makes invented scandals on live air far less likely, but the robot still has to pick the right passage, summarise it and say it without changing its meaning, and each of those steps can fail. More importantly, retrieval moves the trust question from the machine to whoever stocks the library. Here that is a state news agency, and its own wording, “controllable”, says the content is meant to stay within boundaries it sets. A humanoid body adds presence, not reliability. Judge these anchors by the quality and openness of what they read from, not by how convincingly they blink.
