Deep fakes are digitally manipulated videos that have been created using deep learning technology to make the subject of the video (often a famous person) say anything the video maker wants them to say, even incorporating the style and facial expressions of another person.
Example
An example here is a video that demonstrates the technique, and features a fake video of Barack Obama saying things that he would never normally (publicly) say. Example: https://www.youtube.com/watch?v=AmUC4m6w1wo
Improving Fast
The technique, which had its less than auspicious first uses in pornography, where porn actors were made to look and sound like famous people, has much improved and become arguably more convincing as deep learning and AI have led to more seamless and convincing results.
Style Transfer
The development of the technology used in deep fake videos has improved to the point where even a person’s style can be superimposed and incorporated. An example of this can be seen in videos created by researchers at Carnegie Mellon University, who have been able to use artificial intelligence technology to transfer the facial expressions of one person in a video to another.
See this example on YouTube: https://www.youtube.com/watch?v=ehD3C60i6lw where John Oliver is made to reflect the style of Stephen Colbert, a daffodil is made to bloom (time lapse) the same way as a hibiscus, and Barack Obama is given the same facial expressions and style as Dr Martin Luther King and President Donald Trump.
What’s The Danger?
The danger, according to US lawmakers and intelligence organisations, is that videos could be made by adversarial nation states and used as another tool in disinformation campaigns. For example, at key moments, politicians and other influential figures could be made to appear to make false and /or inflammatory statements that could be believed by less politically aware recipients. In short, these videos could be used to influence opinions e.g. at election-time, and could afford a foreign power a way to interfere that relies upon human error – the same thing that many successful cyber attacks have relied upon.
What Does This Mean For Your Business?
With the US Midterm elections on the way, with allegations of Russian interference and possible collusion still hanging over President Trump’s head, and with some evidence that Facebook was used by a foreign power to try an influence the last US election result, it is understandable that the US government is worried about any tools that could be used to interfere in their democratic process. This is one of the reasons why Microsoft has seized 6 phishing domains that allegedly belong to Russian government hackers, and has introduced a pilot AccountGuard secure email service for election candidates.
If the technology behind deep fake videos keeps improving, it is possible to see it being used as another tool in other types of cyber-crime.
There is, of course, an upside and some ways that deep fake technology can be used in a positive way. For example, deep fake could be used to help film-makers to reduce costs and speed up work, make humorous videos and advertisements, and even help in corporate training.
