Apparently there’s a robot art competition and it’s a thing and it’s been going on for a while, at least according to VentureBeat. VB puts out a lot of interesting avenues to explore in terms of what people are doing in the field and from the surface, it’s mostly built upon machine learning.
My favorite aspect of the article revolves around someone who’s produced what sounds like a rather elaborate real-world robot to execute his creations, and the rather long list of software that people are using in this competition all are things I intend to explore further. Hopefully with another blog post as we go down the rabbit hole together.
Personally, while the bulk of the artwork shown in the article appears more than a little derivative to me, I’m sure that this is the phase of ML art similar to the time in the late ’90s where everyone found digital art production.
That was a terrible time where everything that was new basically was somebody either painting just like they did only with digital means (woo! Wacom tablets!!!) or sliced and stacked myriad filters on stuff (have you seen the lates Kai’s Power Tools??) – sort of an Andy Warhol nightmare. I’m sure he’d have laughed.
Then things got really, really good – and all with the same software, well maybe not Kai’s Power Tools.
Recently, through a bit of an obfuscated path, I happened across a group called Obvious who are working on using machine learning to create artwork. While I’ve (and I’m sure most here) heard of ML being used to categorize and quantify art, it’s interesting to see if ML can actually create on its own – or if it can only elaborately remix prior work.
Looking to find more about the group, I eventually stumbled up on this Medium article where it discusses the use of ML and whether it constitutes ‘art’ at all.
Curiously, I recall the same sorts of arguments being constructed around generative efforts ten or so years ago. Both arguments orbit around the degree of the human artist’s ‘hand’ in creating the work and at what level of involvement is necessary before the work becomes art. A tricky question to say the least. While purely generative pursuits had to fight against the notion that one was picking through iterations of randomness to find a usable gem, I’m thinking ML is probably going to fight the notion that it’s an elaborate remix platform – where people search through variants to find a usable gem.
For me, I’d like to see how the machine learning system creates the work and at what level is it combining prior work or creating new techniques. There is a link to a GitHub repository so I guess I have my opportunity to look under the hood.