Jun 22, 2016 | By Benedict
Computers scientists from the IT University of Copenhagen and the University of Wyoming have developed an AI ‘artist’, capable of creating 3D printed artworks. The program uses deep learning and innovation engines to create 3D artifacts without human input.
All across the world, 3D printing is being used to make art. From unique design innovations by Neri Oxman and Stratasys all the way to provocative 3D printed models by Megumi Igarashi, 3D design tools are all set to be the paintbrushes of the future. Behind the large majority of additively manufactured artworks, however, is a human artist. But does there need to be? That is a question being asked by Joel Lehman and Sebastian Risi, both of the IT University of Copenhagen, and Jeff Clune, of the University of Wyoming. This trio of computer science experts has collaborated on an usual AI project, in which a computer has been taught to make 3D (and 3D printed) art. The results? Intriguing, to say the least.
In their effort to create a virtual artist, the computer scientists turned to the image recognition capabilities of deep learning, an algorithm-based area of machine learning that can be used to model high-level data abstractions. The team believed that a combination of deep neural networks (DNN)—a kind of deep learning architecture—and an evolutionary algorithm could be used to create virtual 3D artwork without human input. Their creation, which they have called “creative object generation,” might not be creating masterpieces just yet, but its formative efforts are certainly eyebrow-raising.
The AI artist creates its unique pieces in the following way: its evolutionary algorithm generates a random blueprint, which it turns into a formless 3D image. Pictures of this 3D image are then sent to the deep neural network, which decides whether the 3D image resembles anything familiar: landscapes, people, or just about anything that exists. The deep neural network then gives the algorithm a resemblance report, i.e. “this looks 0.1% like a horse,” and from then on it’s a process of trial and error: every time the deep neural network returns the 3D image, the algorithm makes changes to it to it and sends it back. The deep neural network then says if the 3D image is becoming more or less recognizable, and the process repeats millions of times until a satisfactory “artwork” is created.
The computer scientists behind the process have likened this back-and-forth creation process to that of natural evolution—both exhibiting pathways from simple organisms to complex ones. “It’s incredibly fascinating to me that [evolution], with no volitional thinking, was able to create things of such enormous complexity,” Lehman told Popular Science. “Things that are still beyond our ability to engineer.”
The program was left running for around two weeks, with the back-and-forth process occurring over 2.5 million times in that time period. By the end of the fortnight, the deep neural network was giving the algorithm 95% accuracy reports. Many of the creations are odd, surreal, yet ultimately recognizable—just like much human-authored art. Lehman modestly judged the 3D creations to be “kind of pretty.” A selection of the 3D creations, those that “(1) were possible to print, (2) were colorful, and (3) highlighted interesting features used by DNNs to classify images,” were sent to Shapeways for 3D printing in color sandstone material. Pictures of the 3D printed objects (shown below) were then fed back to the deep neural network, which generally correctly identified the artworks as their intended objects.
The researchers’ paper, titled “Creative Generation of 3D Objects with Deep Learning and Innovation Engines,” can be downloaded here, and will also be presented at the International Conference on Computational Creativity, which takes place June 27 - July 1 in Paris.
Posted in 3D Printing Technology
Maybe you also like:
- New Harvard laser 3D printer prints metal structures in mid-air by lasering nanoparticles
- 3D printing listed as a top at-risk emerging technology in CERT report
- On-The-Fly Print by Cornell team 3D prints high speed wireframe prototypes while you design
- 3D printing industry to reach $21 billion by 2020, UPS and CTA report shows
- Researchers use titanium-tantalum alloy to improve stress absorption of 3D printed implants
- Eat Cambridge food festival to serve five-course 3D printed meal with Nufood 3D food printer
- New A$3.7M 3D bioprinting center to open doors in Australia
- 3D printing breakthrough makes low-cost flexible carbon nanotube sensors for wearables possible
- 3D Food Printing: VTT Finland to explore 3D printed snacks packed with nutrients
- Full color 3D models made faster and cheaper with computational thermoforming
- How to address safety concerns with metal 3D printing
- Dresden names itself the 3D printing capital of Europe
The Computer knows it will never be able to truly think until it's ingested a few mushrooms, or have the experience of the alternate reality contained within.
Just a human wrote at 6/23/2016 4:36:21 AM:
That is less attempts than I would need, and a lot quicker too!
3Ders.org provides the latest news about 3D printing technology and 3D printers. We are now seven years old and have around 1.5 million unique visitors per month.
