Last month’s blog was about the challenges in determining real scientific research and reports that include falsified or erroneous conclusions. It is difficult enough to understand results that are presented without complete coverage of the underlying premises. There are professionals who have indicated that careful watching of a person’s gestures or mannerism during a presentation can provide evidence of inaccuracies. Professionals are able to evaluate changes in voice tones that can provide
Or at least they used to be. In an April 2018 article “Forging Voices and Faces: The Dangers of Audio and Video Fabrication” [Ref. 1], The author covers mentions a speech that President John F. Kennedy was to give the evening of November 22, 1963. A company re-created the speech that was to be given with a synthetization of audio fragments of Kennedy’s actual voice. There are currently a number of programs that can be employed to synthesize audio.
In the 1960s there were a number of efforts by the Russians to remove people, who were no longer in favor, from pictures. At that time, it took massive computer power costing in excess of $50,000 and much labor to achieve the removal of a person. [See Ref. 2 for a number of pictures showing the removal.]
Today, there are a number of way of programs that can be employed on relatively inexpensive desk computers that can make credible changes to photographs. The Wall Street Journal is working to educate their journalists on identifying what are being called “deepfakes”. [Ref. 3] Somethings like, direction of shadows or changes of resolution within a picture can be obvious. As the Wall Street Journal states: “Seeing isn’t believing anymore. Deep-learning computer applications can now generate fake video and audio recordings that look strikingly real.” [Ref. 5].
One of the most recent articles on the impact of computer-generated capabilities is from the IEEE. [Ref. 5] The focus is on the ability of artificial intelligence software to generate Digital doppelgangers of anyone. Work at the University of Washington cited in the article shows how “fake” images can be created from available images on the internet. In particular, the researchers chose to work with high resolution images of Barack Obama. The researchers had a Neural net analyze millions of video frames to create elements of all facial mannerisms as he talked.
There are still areas that need improvement to do an actual replication of the person speaking because the superposition of facial features including the muscles that move when a person is speaking can not yet be accurately replicated when the person turns slightly. But, that is just a matter of improvements in the techniques.
There are articles, which are intentionally not referenced, that show the techniques of using a Hollywood procedure of capturing the movements of an actor in a general manner. Then taking the capture movement points and coupling those movements to another person. It is a technique that is also used in animation. It’s possible to photoshop the head of a person on to a look-alike body. Any person can be inserted into the actions. His or her voice can be created to enhance the believability of the video. What happens when these techniques are available on personal computers?
Where does this lead to? Reference 6 indicates that there are government officials that are concerned the next US presidential election could witness a number of fake videos and cause a serious disruption of the election. We are losing the ability to have sources of information that can be trusted. How does a civilization survive when viewed, heard, or read information has a strong probability of being monoligated?