A unique collaboration between human artist and artificial intelligence
Can art be created by anything other than human creativity? Can artificial intelligence working hand-in-hand with a human artist learn artistic concepts and produce its own original art? Our unique ground-breaking collaboration have answered these questions with a resounding YES.
In a three month-long collaboration between Maltese artist Mark Mallia and Umnai’s AI system Uma (for Universal Machine Artist), we have created a new form of artistic expression: the emergence of human-directed AI art. The human artist acts as a muse and mentor to the AI algorithms and the learning process, resulting in Uma’s creation of own original creative work.
Mark is currently teaching Uma Umnai the basics of painting and artistic expression, by going back to his style as it was two decades ago when he was working as an artist in Miami and gradually painting examples that reflect his evolution up to the present day. The AI is learning how painting style evolves, and having an artist in residence at hand, gets much more insight into the creative process itself. The first result of this collaboration is the set of paintings — the first one by Mark, “Agnes and Ingrid” and Uma’s interpretation of it — “Agnes and Ingrid Re-Imagined”.
Mark says that he consciously tried to produce a work that does not easily fit with his current artistic style, going into cycles of loathing at his own painting. Interestingly, while “Agnes and Ingrid” itself does not sum up Mark’s style, Uma, being trained on Mark’s evolving style, has produced a painting that is, in Mark’s own words, ‘more true to Mark’ than his own “Agnes and Ingrid”.
The origins of Uma lie in work done by Umnai on anti-fraud systems, where we learnt about a method called Generative Adversarial Networks that uses two networks that work against each other — one that creates an example image and one that “critiques” it. During our work, we realised that if we apply a hierarchy of GANs we can add structure and achieve better results, leading to our patent pending anti-fraud technology.
For Uma, we modified the technology to generate images under the supervision of a human artist. In Uma, the artist acts as a third adversary and mentor to the two GAN adversaries — leading to possibly some of the first ever truly creative art created by AI.
We have also made a lot of improvements behind the scenes to increase the resolution of the AI art, enabling us to print out the artwork on a larger scale format than was previously possible. Mark’s oil painting, in an 80 x 120 cm format presented a challenge in itself to be input to Uma’s network — we had to invent a few techniques to handle such large amounts of image data and make efficient use of the GPUs that were needed to process the image over the course of 3 days. The resulting AI artwork was printed on an 80 x 120 cm canvas which was then signed with gold paint as UMA and the Umnai logo next to it to signify that the artwork was created by an AI entity.
The collaboration between Mark and Uma was conceived by Angelo Dalli, the CEO of Umnai and serial entrepreneur who strongly believes that creativity and innovation go hand-in-hand. Behind the scenes, Uma has been trained and perfected by Mauro Pirrone, Umnai’s Chief Software Architect. Uma itself exists on a couple of hundred servers on the cloud.
AI is still making infant steps in creativity. The collaboration between Mark and Uma will continue — who knows what lies ahead?
The two artworks, “Agnes and Ingrid” and “Agnes and Ingrid Re-Imagined” will be auctioned during the Art vs Artificial auction during the Malta Blockchain Awards on the 31st of October 2018, with bidding in both Euros and Bitcoin. In line with our vision to create ethical AI that impacts society positively, all proceeds from the auction will be donated to the Malta Community Chest Fund.
Mark Mallia and Angelo Dalli are grateful to 111 Art Gallery for Mark Mallia’s financial support and artistic representation; Eman Pulis, for the enthusiastic support from the start; Aleksey Leonov, for photography; Michael De Giovanni, for the final processing of the physical canvas output and the rest of the Umnai Team and Olga Finkel, for legal and financial support.