He may no longer be president of the White House but former US president Barack Obama is still making some presidential speeches, although all is not as it appears.
Despite looking and sounding exactly like the real deal the new Obama is actually a digital construction carefully crafted by researchers at the University of Washington who fed 14 hours of genuine speeches into a neural net which has binge-watched so much footage it can now generate new clips indistinguishable from reality.
The result is that researchers are now able to put any words of their choosing into the mouth of their photorealistic sock puppet, modeling the movement of his lips to mimic Obama’s actual tics and mannerisms.
Asked by the BBC whether we should be perturbed by the phoney president, professor Ira Kemelmacher-Sclizerman of the University of Washington, said: “Every technology can be used in a negative way and so we all should work towards making sure that doesn’t happen. Once you know how to create something you can reverse engineer it and so you could identify methods for identifying what is an edited video and what is a real video.”
Past attempts to produce lifelike representations of identifiable figures have stumbled in the so-called ‘Uncanny Valley’ where slight discrepancies in the motion of the mouth are picked up subconsciously by viewers, prompting the development of custom algorithms capable of delivering the requisite accuracy.