We are yet again reminded of the trivial and trifling mentality of certain sections of society , although to some this particular event may signify something far darker and deeper .
- AI artwork sells for $432,500 — nearly 45 times its high estimate — as Christie’s becomes the first auction house to offer a work of art created by an algorithm
The artwork was produced by feeding 15,000 actual art portraits through a machine learning algorithm known as a ‘Generative Adversarial Network‘ .
As such , the AI generated piece sold at Christies has even give birth to a new art genre , known as GAN-ism . I guess the same algorithm that generated the piece also generated the new moniker ‘GAN-ism’ .
Generative Adversarial Machine Learning
Machine learning is concerned with the development of artificial intelligence systems , primarily ‘teaching’ the AI system . Deep learning is a branch of machine learning which aims to create a machine learning system based on a known output , for example if you want an AI system to recreate speech you will already know the required output (spoken words) .
Deep learning systems ultimately aim to ‘build a brain’ which is based on the neuronal network of the human brain . Otherwise known as an artificial neural network (ANN) .
The idea underpinning GAN’s is that they are able to learn for themselves . In fact a GAN comprises two competing (hence adversarial) neural networks , the two neural networks are able to learn from each other . One neural network is called the generator and the other is called the discriminator .
Real world images are fed to the discriminator , which can then classify the images according to a specific rule set , for example ‘is this image a cat’ .
Alongside this process the generator also feeds a random sample image to the discriminator , which could be anything , for example a human face . Because the generator is largely unaware of the discriminators rule set , the images will at this stage not be classified at all and will be discarded (fake) , i.e. a human face is not a ‘cat’.
Meanwhile a feedback loop from the discriminator to the generator allows the generator to ‘learn’ from the applied rule set (cat) . In time this will enable the generator to create an image which will eventually morph from a human face into cat , thereby passing the rule set (real) .
As an added bonus the new algorithms created by the generator which now allow it to produce a cat image , are also fed back to the discriminator in a feedback loop . Thereby also allowing the discriminator to ‘learn’ from the generators algorithm .
It sounds more complicated than it actually is .
For a more simple explanation I can quote from a leading researcher :
Cop vs. Counterfeiter: GANs Slash Data Needed for Deep Learning
“The sparring networks learn from each other. As one works hard to find fake images, for example, the other gets better at creating fakes that are indistinguishable from the originals.”
“After training, what you end up with is a network that is able to paint like Picasso, and you have another network that is able to recognize images and paintings at an unheard-of level of discrimination,”
GAN’s are particularly useful in creating virtual visual images , or even virtual worlds .
GAN’s are widely used by deep state organizations in machine learning applications , especially for simulations of real world scenarios involving real world people and also speech synthesis . A small piece of the puzzle in the race to create a truly intelligent AI system .
The world of high art is a controlled racket , much like many other aspects of society where large amounts of cash change hands . It works like this :
(i) A high profile collector or wealthy individual promotes a particular artist or theme.
(ii) A friend or collaborator of aforementioned artist or individual sells the ‘artwork’ to another wealthy ‘partner’.
(iii) This carousel of cash snowballs until the ‘artwork’ reaches a ridiculous and fanciful valuation , whereupon it is offloaded to an unaware and unsuspecting individual or corporate entity .
(iv) The original partners in the racket bank a large sum of money .
The general theme that continues to spread regarding AI is that it can better and surpass humans themselves , even purely human endeavours such as art are now being slowly assimilated .
A new , new Art Genre
Regarding all of the above , particularly considering the sum of $432,000 which has been pocketed by the collective , I have now produced my own AI generated artwork , along with a new art genre .
I have called the new genre ‘DAB-ism‘ , short for ‘Degenerative Anal Bollocks‘ .
Two masterpieces have been produced so far , they are in the post to Christies as I write , I am expecting a fat cheque (or two) by return . I produced the source material for the masterpieces only this morning .
I emailed copies of both masterpieces to an eminent art critic who had this to say :
“With regard to the issue of content, the disjunctive perturbation of the spatial relationships brings within the realm of discourse the distinctive formal juxtapositions.”
“It should be added that the optical suggestions of the spatial relationships endangers the devious simplicity of the remarkable handling of light.”
My thoughts exactly .
Don’t worry , you too can cash in on the new craze , my advice , get in there quick before it’s too late .
- The Onion’s Top 12 Art World Parodies
- The Onion – Art
- How three French students used borrowed code to put the first AI portrait in Christie’s
- The Painter Behind These Artworks Is an AI Program. Do They Still Count as Art?
- Obvious is a collective of artists, friends an AI reasearchers
- Nightmare Machine
- Deep Dream Generator
Further reading :
- AI Researchers Fight Over Four Letters: NIPS
- WaveNet: A Generative Model for Raw Audio
- TEXT-TO-SPEECH SYNTHESIS USING STFT SPECTRA BASED ON LOW-/MULTI-RESOLUTION GENERATIVE ADVERSARIAL NETWORKS (PDF)
- SPEECH WAVEFORM SYNTHESIS FROM MFCC SEQUENCES WITH GENERATIVE ADVERSARIAL NETWORKS (PDF)
- GENERATIVE ADVERSARIAL NETWORK-BASED POSTFILTER FOR STATISTICAL PARAMETRIC SPEECH SYNTHESIS (PDF)
- EEG Data Augmentation for Emotion Recognition Using a Conditional Wasserstein GAN (PDF)
- Generative Adversarial Networks Conditioned by Brain Signals (PDF)
- Learning from Simulated and Unsupervised Images through Adversarial Training (PDF)
- introduction to spiking neural networks (PDF)
GAN’s further reading :
- A Beginner’s Guide to Generative Adversarial Networks (GANs)
- GAN Deep Learning Architectures – review
- An introduction to Generative Adversarial Networks (with code in TensorFlow)
- Generative Adversarial Networks
- Generative adversarial networks (slide show)
- Generative Adversarial Networks — A Deep Learning Architecture
- Progressive Growing of GAN’s (PDF)