Many who hold balance and fact-based news highly are worried about the rise of false news, and the social context where it is often encouraged by some political figures. To counter this organizations like the Pro-Truth Pledge have been formed (see Digital Journal’s feature on this “Journalists and public figures taking Pro-Truth Pledge“). Nevertheless, the number of deliberately misleading and counterfactual stories shows little sign of diminishing.
One reason for the vast number of inaccurate stories is due to computer generated text and this looks set to become worse as artificial intelligence improves. In the most sophisticated form this is with AI-driven “fake” videos. Here, Business Insider reports, computers are able to analyze a person’s voice and facial expressions, then use the digital data to produce a convincing video of that person saying.
This feat has been demonstrated at the University of Washington. To demonstrate this a video has been constructed. For this the researchers took audio of former President Barack Obama and proceeded to synthesize a high quality video of the statesman speaking via an accurate lip sync, which was composited into a target video clip.
Machines were trained using several hours of his weekly address footage. This enabled the recurrent neural network to learn the necessary mapping from raw audio features to mouth shapes. The video outcome is shown below:
The researchers succeeded in synthesizing high quality mouth texture and composite this with three-dimensional pose matching in order to change what Obama appears to be saying in a photorealistic manner.
The implications of this are that people need to be surer as to the source of an article pertaining to be news.
In related artificial intelligence news, sophisticated machines are being used by researchers in a round-about way – to help science understand the basis of human thought processes. See the Digital Journal article “Artificial intelligence starts to examine human intelligence.”
