Back in the 80s, the scientist in the field of AI, Hans Moravec, announced a paradox: what is easy for humans is difficult for machines.
He talked about visual and auditory understanding, and it's easy to imagine how right he was considering the machines of the time.
Things have changed a lot since then. Today, AI systems are much more capable of understanding everything they see or hear. And these are just two of the many parameters on which artificial intelligence is gaining ground on humans.
When we think about AI we are always led to think that it is pure automation. It is not true at all. Just for one thing: do you think AI can't replace photographers and portrait artists? Mistake. He no longer even needs a real-life model to photograph.
Today artificial intelligence is able to "imagine", that is, it can depict things that did not previously exist in reality.
This video shows the results obtained by an AI that has learned to generate photos of people who don't exist. In terms of quality, only a few artists in the world are capable of inventing faces with photographic precision.
GAN, the secret of AI
The ability to "imagine" is one of the characteristics provided by Opposing Generative Network (GAN in English), one of the most explored methods in the AI field. GANs are partly inspired by neuroscientific research.
In practice, a GAN puts two entities in "competition" that learn from each other: one learns to generate fakes, the other learns to recognize them.
The more the ability to recognize a fake improves, the better the fake generated (which corrects the defects of the previous ones) improves. It is a formidable learning mechanism discovered by neuroscientists and inherent in the human brain that is called actor-critic model.
Imagination will not be a privilege of the human mind alone for a long time: how can we exploit this ability acquired by artificial intelligence?
Here is an essay of what is happening thanks to GAN in laboratories all over the world.
Turn night into day
Practical implications of imagination? Being able to represent a subject in a different way or translate one representation into another. For example this AI imagine what the drawing of a photo might look like, or the color version of a black and white photo.
An application of this ability can help us see the world differently, or see beyond what appears visible to us. Take a photo taken at night and turn it into a daytime photo.
A formidable quality that will be able to assist self-driving cars, allowing them to move precisely even in the dark, in the fog or in other adverse conditions.
These are developments that see (alas, as often happens) the military sector to act as a driving force, with first solutions already developed of AI-assisted night vision devices.
Establish the shape of a person by looking at them dressed
When something is not even visible, GANs “imagine” (and visually represent) intelligent and increasingly precise reconstructions. Let's take the case of artificial intelligence called BodyNet, able to build a person's build based on a dressed photo of him.
A very useful ability, for example, to design tailored suits without performing manual measurements or body scanners.
See through the walls
Another AI can literally locate and “see” someone even when they are behind a wall. Using a method similar to that of bats, this artificial intelligence interprets a signal (in this case WiFi) emitted by a device that bounces off objects.
Discover viruses and antivirus, or disease and medicine
AI's imagination ability is not limited to creating images or transforming them. Imagination is the tool that leads us to discover new things, 3d even artificial intelligence is no exception in fields such as cybersecurity or drug development.
Modern cyber security tools include AI that can detect threats in a thousand ways. Researchers have designed a GAN that learns to generate pieces of malicious code capable of breaking antiviruses. It may seem scary, but the good news is that GAN is simultaneously improving its ability to recognize increasingly sophisticated viruses. This is even more valuable when the "virus" is not computer-based, and learning to fight it means developing or finding increasingly effective drugs.
What will all this bring us?
The fourth industrial revolution is not about automation, but about collaboration and symbiosis between humans and machines. GANs are a turning point in the development of artificial intelligence and will help us give real superpowers to our mental abilities.
Even if imagination is not always creativity, it is still a tool that allows us to discover new things: will there ever be something among these that will allow machines to surpass us in creativity?
