Can AI be creative?
Generally, when we think of Artificial Intelligence, what crosses our mind is stuff like prediction and recognition.
But when we are trying to make something that can surpass human intelligence, we can not ignore the fact that human brains are “Creative”.
Although AI is not capable of creating something new on its own, it can take reference from the previous data and create something different in a recognizable way. This takes us to the concept of GAN(Generative Adversarial Network), creating data that has never been seen before.
Generative Adversarial Network
GAN is a way to do generative modeling where we have training data and we like to produce examples of data that seem to have been drawn from the same data.
An AI’s creativity is somewhat different from human creativity. Where humans can take inspiration from many different things and create something that is unique in its own way, what GAN creates resembles the training data in some sense.
Before moving further let us get some basic idea about the architecture of a GAN.
Architecture of GAN
GAN architecture consists of two neural networks, a Generative Model to generate new data and a Discriminative Model to discriminate among generated data and original data.
The discriminator is first trained to recognize real and generated data then the generator is trained over the discriminator data to make samples of data that can fool the discriminator.
Both the neural networks learn together and compete with each other in the training phase. When the discriminator makes a mistake in distinguishing between the real data and the generated data the training is considered complete.
AI being creative
When it comes to AI that does not predict or recognize, mostly all of them are related to GANs. GANs can do really interesting stuff like creating portraits of imaginary people, animating 2D pictures, creating Deepfakes, generating songs, and much more. Below are some examples of AI doing creative work.
This person does not exist
The website uses styleGAN to show portraits of people who do not exist in the real world. Every time you refresh this website a new picture appears on your screen.
What about an image coming to life?
With an AI-powered technology called Deep Nostalgia, it is possible to create an animated video of 15–20 sec from a single photograph. Making the person in the photograph smile, move their face, and blink their eyes. Below is a video of deep nostalgia in action.
Does predicting someone’s face from their voice seem interesting?
The paper Speech2Face: Learning the Face Behind a Voice features the study of the task of reconstructing the face of a person from short audio of that person speaking using GAN.
Deepfakes
GANs can also generate Deepfakes, where computer-generated images are superimposed on existing pictures and videos to create fake narratives.
What if our favorite songs could continue?
Jukebox from OpenAI, if provided with a small part of a song, can create new songs of different genres and even add parts to continue existing songs.
Conclusion
Creativity has many different definitions describing different ideologies and mentalities. Due to the flexibility of this term the question “ Can analyzing the available resources and rearranging the data in order to get something different be considered as creativity?” surrounds the minds of many.
All the advancements in AI that we can see are in the field of Artificial Narrow Intelligence implying that an AI is only capable of doing some specific tasks and to attain Artificial General Intelligence that can surpass human intelligence there is a long way to go.
