China’s Xinhua News Agency, alongside the Chinese search engine Sogou.com, have revealed what they call the “world’s first artificial intelligence news anchor.”
Its appearance, facial expressions, and voice were modeled after the real news anchor Zhang Zhao, using machine learning algorithms that essentially “taught” it to recreate his likeness. The final product, what you see in the above introduction video, is meant to read the news just like a human. Mostly.
Here’s another example of the English-speaking virtual anchor at work. They’ve also created another anchor that speaks Chinese.
Many have highlighted the uncanny valley-effect given off by Xinhua News’ virtual reporter. Professor Michael Wooldridge of the University of Oxford told BBC News that, to him, it appears “very flat, very single-paced,” and lacks the human quality of real news presenters.
He also brings up the issue of how human audiences may have a difficult time forming connections with an uncanny animation and a robotic voice. The virtual anchor, he says, will therefore lack the authority of, as BBC News puts it, today’s “highly trusted public figures.”
Can a virtual news anchor be trusted in a similar way? Can it be trusted more? Do these questions even make sense in this context?
In the video at the top of this post (a 30-second preview uploaded to YouTube by New China TV), the virtual anchor states that “texts will be typed into [its] system uninterrupted.” News is meant to be input and read as it happens, and this, according to Xinhua, is one of the virtual anchor’s many benefits: It “can work 24 hours a day…reducing news production costs and improving efficiency.” They won’t need an actual person sitting in front of a camera to get the news out quickly.
And yet, the virtual anchor is not human, and cannot think.
In fact, there’s much debate and frustration over the use of the term “artificial intelligence” in this context, even in the comments of this very article (see below).
As mentioned above, Xinhua’s virtual anchor was created using machine learning algorithms to replicate the appearance and speech of real news anchors. However, as far as the actual reporting and writing of the news goes, that’s all still done by humans. The virtual anchor is not, itself, intelligent, but it was created using methods related to artificial intelligence. Because of this, as Will Knight of MIT Technology Review puts it, “You can call it an ‘AI anchor,’ but that’s a little confusing.”
So, virtual or not, humans are still at the wheel, just as they are with human anchors. The words they read have to come from someone, somewhere.
In the case of simply reading news, however, this means the virtual reporter isn’t so different from ordinary humans reading teleprompters, as far as the end result is concerned. Words are simply being read aloud. The larger issue, I suppose, is what happens to the news industry when everyone gets replaced by artificial intelligence and fancy computer models.
At the very least, I hope they work on that digitized voice, so it doesn’t sound so much like all those lame, knock-off YouTube list videos. I imagine that, eventually, we’ll have realistic speech synthesis, human-like CGI, and AIs smart enough to gather news on their own. Then what?
Just imagine news being reported by someone (something?) like Siren.