I Think, Therefore I Am… Machine?
By Judy Davies, Vice President, Global Marketing Communications, Advantest
The ability to think has been a central, defining aspect of humanity since our beginning. Today, technologists are using artificial intelligence to instill that capability into machines. Through statistical models and algorithms, machine learning enables computers to perform specific tasks without receiving explicit instructions from a human. This means that the computer reaches conclusions by accessing available data, identifying patterns and using logical deduction. This does NOT mean AI systems can generate original ideas (at least, not yet). Rather, their intellect stems from their near-instant ability to crunch large volumes of data and then employ their massive memory capacity to compare and search for linkages that yield logical answers.
An emerging area of machine learning is generative adversarial networks (GANs): deep neural network architectures comprising two nets, in which one is pitted against the other in an unsupervised learning environment. For example, one computer might generate a realistic image, and another is then tasked with determining whether or not the image is authentic. By having these two neural nets engage in game-playing to repeatedly fabricate and then detect realistic likenesses, GANs can be used to produce images that a human observer would assess as genuine.
It should come as no surprise that training GANs is challenging. To use an analogy easily understood by the human mind: It’s easier to recognize an M.C. Escher drawing than it is to replicate one. Nevertheless, GANs hold extraordinary potential. Working from motion patterns captured on video, they can create 3D models of a wide range of objects, from industrial product designs to online avatars. They can also be used to digitally age a person’s image, showing how he or she may look a decade or more in the future – which may be useful in helping to identify teenagers or adults who went missing as children. Going a step further, GANs can sort through many terabytes of images culled from security monitors and traffic cameras to perform facial recognition. This can help to actually identify and track the whereabouts of missing kids or wandering Alzheimer’s patients – not mention wanted criminals.
As with most technology, there is a cautionary aspect to GANs. For example, they could potentially be used to generate artificial images for nefarious purposes, such as creating fake photographs or video clips that unsavory types might use to make innocent people appear guilty for political or financial gain. They may also 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. How to build in safeguards that prevent these types of illicit deployment of GANs is an important consideration.
GANs can be applied to synthesize or fine-tune everything from voice-activated smart electronics to robotic medical procedures. As the technology is further developed and applied, machine learning and GANs are becoming reality. Self-improving AI is increasingly being used to affect the authenticity of what we perceive and think – it’s a literal (human) brain-teaser.
Did you enjoy this article? Subscribe to GOSEMI AND BEYOND