In late May 2019, Facebook users who follow topics of interest related to politics in the United States were shocked to find a peculiar video in their newsfeeds. The video showed Nancy Pelosi, Speaker of the House of Representatives, speaking to an audience about her recent participation in an event held at the Rose Garden. What was strange about the video is that Representative Pelosi looked and sounded as if she were inebriated. The video, which received more than two million views in just a couple of days, was not real; it was part of a new and worrisome trend of digital alteration known as “deepfake” reality, and it was likely created with an advanced artificial intelligence platform such as TensorFlow, a project managed by Google.
An Accessible AI Platform
TensorFlow is a second generation AI platform as well as a computer processing unit upon which advanced circuit logic boards will be built in the future and for the purpose of creating truly smart supercomputers. Another part of the TensorFlow project is known as the Google Compute Engine, which is a cloud computing platform for virtualized AI projects. As an open-source project that is constantly learning on its own, TensorFlow can be accessed by just about anyone who is interested in developing AI projects, which is probably how the deepfake Pelosi video came about.
AI’s Fast-Paced Development
Although the field of artificial intelligence has been undergoing development since the days of World War II, some of the most significant milestones have taken place in recent years thanks to the work of technology giants such as IBM and Google. Ever since the IBM Deep Blue computer made headlines by winning a chess tournament against grandmaster Garry Kimovich Kasparov in 1996, AI development has grown by leaps and bounds. Deep Blue has retired and is now part of the Computer History Museum in Mountain View, not far from Google headquarters. IBM would later develop the Watson supercomputer, which in 2011 easily beat two Jeopardy! masters on national television; Google wanted to take AI further than Jeopardy! and chess, thus setting off to develop a construct that could beat a human greatly skilled in the ancient game of Go. When the Google AlphaGo program beat Korean Go master Lee Sedol in 2016, the field of AI turned a new page because this is a game that requires some level of irrational thought process, which means that computers are learning to pick up on some of the nuances of human reasoning.
The Promise of DeepDream
Of all the applications and projects currently associated with TensorFlow, the DeepDream program is one of the most interesting. DeepDream involves computer vision, which goes further than capturing images with camera lenses; the goal is to train DeepDream to see and interpret the world like humans do, and this could one day give visually impaired people the gift of artificial vision.