We ended this nice collaboration with a small interview (6 questions) to go over some topics that popped during our discussions: what the project’s results meant for Cali, how she’d perceived the various work phases, what she thinks of some publicized “AI and art” projects, etc.
Mina Pêcheux : We began by generating images that “looked” like yours. What was your first reaction when you saw them?
Cali Rezo : To see the images produced by the AI helped me appreciate a few things. Firstly, I realized that the result could not replace my paintings directly but could serve as a new base for my research. Secondly, some family of shapes appeared (circular, square, one stroke, two strokes…); I had felt that before but never formulated it this clearly. And we did explore these categories in more detail later in the project by using other algorithms to analyze my paintings.
M. P. : You’re referring to the second part of our project, about image analysis. For this part, there was a necessary “labeling” upstream work. You handled this: how was it?
C. R. : I knew that a huge amount preliminary work had to be done by humans so that an AI could start its process (see this article by Mark Harris), but I really experienced how repetitive this task is. You have to focus on a basic assignment that is quite hypnotic. Besides, you don’t always get how “Big” the “Big data” is! When you told me we needed at least a hundred examples for each of our nine categories, I thought it would take me several hours of work… even though it is a tiny dataset for an AI!
M. P. : And were you faced with specific difficulties or unexpected questions?
C. R. : From time to time, I had a doubt on the category I should choose and I thought a machine would have trouble giving an answer. Can a circular stroke that continues “outside” of the canvas to then come back into it be regarded as one stroke, or does the fact that it is off-camera transform it into two distinct shapes? Can you teach an AI to “imagine” the missing part of the movement?
M. P. : Having concluded this project, do you think an AI is capable of “creating” an original piece of art?
C. R. : Late last year, a painting generated by an AI was sold more than $430,000 at Christie’s (see the article on their website). To me, when I look at how the trio worked on this project, it is obvious there is a real artistic process and that AI was after all an evolved brush serving art. But a tool and a medium alone do not produce an image, you need a decision – and for now, a human one – to create a piece of art; and the fact that you like the result or not does not depend on which technique was used. By the way, I’m sad that all the articles that discuss this painting focus on the scientific achievement while denying it its right to be called art. So, to answer your question: in my opinion, an AI currently does not create art but can be used to create art.
M. P. : Does that mean that anyone with this program at their disposal could become an artist? Do you think that AIs that allow you to “paint like Van Gogh or Picasso…” (see this article by Morgane Tual – in French) make creators out of us?
C. R. : It would be a bit like saying that with a great calculator, anyone can be a mathematician or that with a super computer, anyone can make special effects in movies… First, you need to learn field-specific knowledge, then you have to work for a long time on using your tools right and most importantly you need to have an artistic approach: an idea, a direction, a reason to use your tools. When you talk about AI applied to art, the majority of media makes you believe that you only need to hold an efficient tool to be an artist, although it is often just about reproduction: a photo “in the Van Gogh style” (such as what DeepArt.io offers), “completed artwork” from doodles (like the AI “Vincent” from Cambridge Consultants) or a landscape generator from drawings (in French)… These are indeed technical feats but don’t magically turn people into artists.
M. P. : This example of a painting that was auctioned at Christie’s is very interesting. However, there is a big difference with our project because in this case it was figurative painting (the AI “learnt to paint” from 15,000 portraits), whereas we tackled the problem of abstract art. What do you say we try to use paintings from your previous period, when you were into digital painting, and see what an AI gets out of it?
C. R. : I won’t be able to provide you with 15,000 paintings, but why not!
Given all that Cali and I have pondered upon and discovered doing this project, there is no question that a future project mixing our 2 passions once again could be very fruitful and teach us a lot more… perhaps a new project could be happening soon?
Wrapping things up
This series on AI and art is now nearly done! In the final article, I’ll give you some of my final notes on the project, plus some details on the technologies we used (namely: TensorFlow and OpenCV) and some news about Cali Rezo’s upcoming events.
I hope you have enjoyed the series so far and that you will stick around for the last episode!
- Cali Rezo’s website: http://www.calirezo.com/site2015/
- TensorFlow’s website: https://www.tensorflow.org/
- OpenCV’s website: https://opencv.org/
- Obvious Art’s website (the trio behind the painting that was auctioned at Christie’s): http://obvious-art.com/index.html
- DeepArt.io’s website: https://deepart.io/
- “Vincent” AI, from Cambridge Consultants: https://www.cambridgeconsultants.com/vincent
- M. Harris, “Amazon’s Mechanical Turk workers protest: ‘I am a human being, not an algorithm’” (https://www.theguardian.com/technology/2014/dec/03/amazon-mechanical-turk-workers-protest-jeff-bezos?CMP=twt_gu), December 2014. [Online; last access 18-May-2019].
- Christie’s, “Is artificial intelligence set to become art’s next medium?” (https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx), October 2018. [Online; last access 18-May-2019].
- M. Tual, “Intelligence artificielle : quand la machine imite l’artiste” (https://www.lemonde.fr/pixels/article/2015/09/08/intelligence-artificielle-les-machines-peuvent-elles-etre-creatives_4749254_4408996.html), September 2015. [Online; last access 18-May-2019].
- E. Ghesquier, “Vous dessinez comme un pied ? L’intelligence artificielle de Nvidia corrige le tir…” (https://www.presse-citron.net/vous-dessinez-comme-un-pied-lintelligence-artificielle-de-nvidia-corrige-le-tir/), March 2019. [Online; last access 18-May-2019].