Machines Do Dream
"Machines Do Dream" is an interactive art installation which allows the viewing public to generate and own an AI artwork.
This is the video played when the installation detects movement near it.
Below are the extracts from the exhibition which will go alongside the installation for Version 2 which I am working on.
Machines Do Dream
This installation is an AI (Artificial Intelligence) machine capable of Generative art.
It has been trained to “understand” the structure, form and style of thousands of digital artworks across many art genres.
Along the way, it has created a vast virtual space, within which lives millions of synthetic art images. The art which the machine creates depends on how and where this space is explored.
Like all good machines it lets humans have some control over it’s behaviour.
<Diagram will be inserted here - mockup below>
The more unusual or experimental work will live on less trodden paths and in it’s darkest corners. You can control this with the first dial - (1 in the picture above).
Additionally, the machine works with something called “semantic scoring”. This allows it to gauge how much it’s artwork resembles real world things. It can then use this score as a filter. The second dial (2 in the picture) allows you to adjust this filter. It controls the balance between how figurative or abstract you wish the work to be.
Now it’s over to you to experiment and explore. Bear in mind that the artwork it creates is entirely uncurated. It is also entirely unique. You will literally be the first to experience what it has created.
Take the time to adjust the dials and find something you like. Press the “Keep” button when you do - (labeled 3 above). You may keep as many as you wish. Otherwise, if you don’t find something, remember that you are exploring a great space. Maybe tune the dials and then press the “Try Again” button.
Digital rights have been waived by the artist for work which you keep.
Due to budget limitations the display used in this installation is old. The artwork on view may lack some of its actual vibrancy and clarity.
All work you keep is archived and is available to order as a print. These are professionally printed on museum grade archival paper at 1200 DPI. See bridie-hinton.com or contact firstname.lastname@example.org for more details.
“Machines do Dream” is a cutting edge technical art project using software that has only been available since Spring 2021. What has been referred to as the “machine” is a type of AI model called a Generative Adversarial Neural Network or GAN. It is capable of producing Generative artwork to a fine art standard.
It’s using the very latest NVidia’s Stylegan AI libraries and OpenAI CLIP software.
The GAN models have taken several months to fully train using NVidia GPUs - mainly class V100 and P100s. These are quite capable of billions of calculations per second.
When it comes to code development and tooling in the world of data science, the Python programming language is peerless. Barely 100 lines of code have been written in this entire project.
Art datasets which are in the public domain were used to initiate training. Subsequently, art derived from earlier generations of the model have been used. Alongside the model these datasets have been heavily curated and developed. It has taken many weeks using the very latest human beings.
The GAN model is deployed on Google’s Cloud Platform and uses Google Colab+. The local on-board computer, which coordinates work with Colab is a Raspberry Pi. It has a number of infrared senses which react to movement.
Again, due to budget constraints GPU access is limited. Google Colab Pro’s RATE and timeout limits may disrupt the generation of artwork at certain times.