Truly, the year 2018 can be seen as the beginning of the creativity heyday made with/by Artificial Intelligence.
Sure, the whole story begun earlier, Google Deep Dream was already a continuation of human attempts to use a machine for art.
With the development of #AI, the production of art by digital entities became a fully new form — an interim state between author and tool.
The attempts for use of Artificial Intelligence in the art world became controversially prominent in the summer of 2018, when collective Obvious sold an AI generated work at Christie’s (with some crucial copyright issues).
— Artnet (@artnet) August 21, 2018
In 2018, another key player from the subset of creative AI tools became popular: BigGAN (GAN = Generative Adversarial Networks). Let’s take a closer look at this auspicious development.
Researches, programmers, and artists are working with Deep Learning using TensorFlow (open-source software library for dataflow programming by Google, practical for Deep Learning) and trying to let the Neural Networks visualize the world’s perception with Artificial Intelligence.
The best thing is: Google provides, with Collaborative Research, the interactive possibility for everyone to try out and experiment with BigGAN.
In my series “AI & Creativity” I want to observe with you the newest tendencies, try out new tools, and present the #AI artists.
I have to admit, there is a backfill I have to catch up on, since the things are developing so rapidly. There is still a lot to show!
Today: BigGAN as #AI image generator
You can install TensorFlow and other required components on your local drive (with some know-how and workarounds). But there is also a Notebook where you can play around with the system, even if you aren’t an #AI specialist. And I tell you, if you start it, you will be eager to learn everything about Python and Deep Learning in order to get more deus ex machina.
Here is the basic BigGAN TensorFlow Notebook to start with.
The usage is easy:
- Run every cell by clicking on the [>] sign.
- Test the default settings.
- Try out what you want (don’t worry, you won’t break the internet).
In the “Category” you will find diverse motifs and objects which can be generated by BigGAN.
Here is what BigGAN understands under “Valley”:
Not bad. Besides some weird artifacts, one could believe it was just another mountain vista being snapped by a photo camera.
The system is trained on huge amounts of labelled visual data — also on a great variety of diverse nature photos. The images above look photorealistic, even if they are entirely generated by Deep Learned neural networks.
So let’s dig deeper.
409) Analog clock
As you see, the system fails on details here. We can literally observe it trying to deliver its concept of the asked object: “roundish”, “with arrows”, “with numbers”. It doesn’t simulate though, because in case of simulation the numbers would be correctly depicted.
There are clear parallels between attempts of AI to interpret the world and the Theory of Forms/Ideas by Plato:
Ideas or Forms were meta-physical essences of the material things. The material things weren’t originals, but just imitations of the Ideas/Forms.
So, in this case, the objects we humans know and the visions of recognized entities by AI are equivalent. They are only endeavors to cover the Ideas. What do you think about it?
The results look rather like a drawing or a postcard than just an envelope. (Some complications during labeling by humans, probably). But here we see AI trying to depict human creativity. Writings. Doodles. Texts we can for a second recognize as a text, but in the next moment, they just slip away like a book title seen in a dream. Our brain is working intensely to comprehend the vision.
And so let’s confuse our brain even more with a generated image.
917) Comic Book
You clearly see the topoi of a comic book: superhero figures (colors resembles Superman), big catchy letters on the cover, a comic shop (a pretty devastated one, though), and a wall with posters.
This is not just a simulation of what comics should look like. It’s a clear attempt of Artificial Intelligence to visualize the understanding of the ‘Idea’ and ‘Form’ of the entity “comic” – based on already familiar ideas and images. Call me esoteric if you want, but I see here way more than just a product of bits and bytes. It’s an urge to understand the world out there.
The Basic BigGAN tool allows other weirdness as well, but let’s look at it in our next post. See you!
My series “AI&Creativity” will be continued!