Chinese news agency Xinhua recently added two new staff members that are a little bit different than your regular news anchors. They may look fairly normal at first glance, but the two members are actually fully digital versions of real humans. According to the agency, the AI news anchors will be able to report the news 24 hours a day, all year round without taking a break. The AI news anchors were developed in collaboration with Chinese search engine Sogou. They are advertised as being the future of broadcasting.
The Very Definition of Uncanny Valley
While many would argue that this is not actual artificial intelligence, machine learning technology was indeed used to create these digital news anchors. Interestingly enough, the virtual personas are actually based on two real anchors working at Xinhua. The developers used genuine footage of their news reports and animated their mouths to make them as lifelike as possible. Thanks to machine learning technology, the AI news anchors look fairly convincing as the mouth animations always seem to fit what they are saying. That said, they are not entirely believable.
Xinhua unveiled both English and Chinese speaking versions to demonstrate that these types of broadcasts can work in any language. Unfortunately, the AI news anchors don’t sound quite right in either language. The anchors can transform text inputs into speech but the problem is that their voices are fully synthesized. Synthesized voices are definitely better nowadays than they used to be they’re still not likely to fool anyone just yet. In other words, the AI news anchors sound as lifeless and emotionless as one might expect. Similarly, the digital anchors don’t look 100% realistic from a visual standpoint either. The mouth animations are not bad at all but other parts of the face are almost entirely still.
Having said all that, this is certainly a very impressive first effort. It’s not difficult to see how this type of technology could be used to create entirely realistic AI news anchors at some point in the near future. Of course, broadcasting is just one example. This technology could easily translate into a wide variety of different areas to reduce production costs. At least, that’s what Sogou and other companies involved with machine learning are trying to pitch.
This isn’t the First Time We’ve Seen this Technology in Action
As demonstrated by Jordan Peele a couple of months back, it’s pretty easy to make a realistic version of someone like ex-president Barack Obama nowadays and make him say whatever you want. Peele does a great impersonation of Obama’s voice so his digital doppelganger is even more impressive than the Chinese AI news anchors. For now, you’ll be able to spot the fake if you’re paying attention but it seems like it’s only a matter of time until this technology will be able to fool even the most keen-eyed viewers. When that happens, this tech will probably be used for more than mere entertainment.