Pushing AI in Art

A view of the process from MIT Computer Science & Artificial Intelligence Lab

Another paper has come out about how researchers have taught an AI system to paint in the manner of a number of artists. From a machine learning perspective, the project is pretty darn impressive and I personally find it quite interesting to read through the training process. It’s particularly unique in that it appears to not start from finished works like most so far, rather by watching how people paint and emulating the human process. 

I want nothing to take away from the triumph that these researchers have achieved, but I guess I’m waiting for the era when AI emulation gives way to the creation of new expressions. I am hoping soon enough that the more clever of the artists out there can take this technology and use it to create something spectacular and well beyond mimicking other artists. 

Some examples from Ms. Shane’s excellently named blog

A great example of what I’m hoping for is what Janelle Shane has experimented with by combining a machine learning system (Runway AI) trained on the Great British Bake-off and apparently random squirrel images to come out with something decidedly different and fresh – in a pretty odd way. The process has a feel to me not unlike the process of glitching out video and the effects look like they have the same presence.

To me, this is when AI art will really come into its own and I’m excited for the first project when someone can take that and bend it to make their own statements. 

Machine Learning as a component of Lorem’s Adversarial Feelings

There’s a lot of hype about machine learning entering the art world. I’ve seen a lot of projects as well. The Adversarial Feelings project by Lorem seems to be one of the best examples of really integrating ML technologies into human artistic endeavors. I especially like what is described as an interactive process that moves back and forth between human and machine in terms of building the work, as I’m not really on board with the methodology that works in a one-way fashion where a data set is learnt and permutations are belched out in so often a project.

The integration of three disciplines into one project is also quite exciting. Putting video, sound design and ML specialists together makes for pushing all three further than a project of just one specialty and I think could inform the breadth of possibility when ML gets integrated even more so into the arts – far better than data scientists working in a vacuum on projects or artists wading into the shallows of the technology.

I definitely suggest checking out the interview and learning more about the project and how the interactions came together between the specialties.