How Do I Know? A Machine Told Me So

More than 375 years ago, René Descartes wrote “I think, therefore I am.” And “Think” has been a slogan used by no less a technology giant than IBM for more than a century. The thought process has been a defining aspect of humanity since our beginning. But now technologists are working to imbue that capability into machines through artificial intelligence.

Programming computers is nothing new, but creating computers that can do more than execute external commands from people and actually “learn” on their own opens a new world of possibilities. Machine learning uses statistical models and algorithms to enable computers to carry out specific tasks without receiving explicit instructions. In effect, such a computer relies on accessing available data, identifying patterns and using logical deduction to reach conclusions. It’s important to note that AI systems do not generate original ideas. Their intellect derives from applying their massive memory capacity and the ability to nearly instantaneously crunch large volumes of data, comparing and searching for linkages that allow them to arrive at answers.

An emerging area of machine learning is generative adversarial networks (GANs). To give an example, one computer might generate a realistic image and another system then tries to determine if the image is authentic or not. By having two neural networks essentially game-playing with each other – which can be called unsupervised learning – to repeatedly fabricate and then detect realistic likenesses, GANs can be used to produce images that look genuine to human observers.

As you might expect, training GANs can be challenging. Think of it this way: It’s easier to recognize a visually puzzling M.C. Escher drawing than to replicate one. But the potential is extraordinary.  Working from motion patterns captured on video, GANs can create 3D models of objects ranging from industrial product designs to online avatars. Or they can be used to digitally age an image of a person to show how he or she may look a decade or more in the future. And GANs can be applied in culling through terabytes of images from in-store security monitors and traffic cameras to perform facial recognition and track the whereabouts of wanted persons, from fleeing criminals to runaway kids.

As with most technology, GANs also can be used in illicit ways. For example, they might be applied in creating artificial images for nefarious reasons, such as fake photographs or video clips that can make otherwise innocent people look guilty for political or financially gain. Or it could be used to circumvent the CAPTCHA security feature of wavy letters and numbers that many websites use to deter bots from accessing the sites in the guise of human viewers.

In addition to visual manipulation, GANs can be applied in synthesizing and “fine tuning” everything from voice-activated smart electronics to robotic medical procedures. As the technology is further developed and applied, GAN and machine learning is becoming reality. How ironic that self-improving artificial intelligence is being used in affecting the authenticity of what we perceive and think.