Category: Art

Generating Art using Robotics and AI: What is out there? (2012)

*** Access the full report here.

     Acess the introduction post here***

One of the first uses of randomization in arts is commonly attributed to Wolfgang Amadeus Mozart, in the field of music. He wrote the measures and instructions for a musical composition dice game, where pre-written measures of music were chosen according to a mapping between these and the results of throwing die; these were pasted together afterwards to create a Minuet. [1]

However, it was only in the 20th century that randomization in arts became more popular. John Cage’s work named Music of Changes in 1951 is often considered the first piece to be largely produced through random methods, where he used I Ching, a Chinese classic text commonly used as a divination system. He would “ask” the book questions about various aspects of the composition at hand and use the answers to compose.

In visual arts, it was Ellsworth Kelly, in 1957, who first featured generative methods in his pieces by using chance operations to assign colors in a grid, or colourful pictures and cheap materials cut into strips or squares to produce collages.


Ellsworth Kelly, Spectrum Colors Arranged by Chance V, 1951

Other artists, namely, Alfred Jensen and Carl Andre, traditionally abstract artists also contributed to the generative art movement during the 20th century.

However, it was with Frieder Nake, a mathematician, that computer art became more well defined. He began to write computer programs to generate drawings. He used Zuze Z64 Graphomat, a digital plotter from 1961 developed by computer pioneer Konrad Zuse, to do drawings as of 1963. [2]

In 1973, the software AARON was born in University of Stanford developed by artist Harold Cohen and it is now a project that has been ongoing for many decades. AARON was one of the first machines to be made with the intention to programme a creative agent that produces art and collaborates with its creator. However, despite being an autonomous generator of drawings, it cannot learn new styles nor objects. Each new functionality and style has to be hand-coded. [3] For instance, when generating the drawing of a person, it goes back to memory and finds the components needed to generate a person (e.g. arm, torso, etc.), however, the randomness is introduced to decide the position of each component, for instance, at what angle the elbow is going to be at. \cite{aaron}


AARON, In Zana’s Room, 2013

In 2006, in the Computational Creativity Research Group at Imperial College London the Painting Fool was born. The objective is to create a software that will be taken seriously as a creative agent.

Shortly after, in 2009, eDavid was started. This is a project from University of Konstanz, where researchers are creating a robot that physically paints.

eDavid project page

In 2012, Vangobot was built as a collaboration between an artist and a programmer. It’s a robot that given an image and instructions paints on a canvas, being able to accomplish different styles. [4] BNJMN was also created in 2012, as a project from the Basel Academy of Art and Design, and it is a “mobile sensory image production mechanism” [5]. It roams in search of paper and produces abstract art.




[2] R. Bowlin (2010), Z64 Graphomat and Frieder Nake. Available:

[3] P. McCorduck, Aarons Code, W.H.Freeman & Co Ltd, 1991

[4] L. Kelly, D. Marx, (2012), The vangobot project, Available:


Generating Art using Robotics and AI – Introduction & Context

*** Access the full report here. ***

When talking about art generated by robots, there are several interpretations ranging from a very basic system that prints art to a fully automated creative agent. What’s more, between these two extremes, there have been a lot of different contributions and projects involving both the art and the scientific community.

Within the arts community, generative art refers to any art practices where the artist uses a system (this could be a set of rules, a computer program, a machine, or any other procedural invention) that has a degree of autonomy that contributes to or results in art.[1]

A common misconception is to think generative art is only using computer programmming in art. In reality, the key concept of this type of art is the procedural invention. This means that generative art includes the usage of chemical reactions, living plants, condensation/crystallization processes, melting substances, or any other physical process that would take place autonomously. The keypoint here is that the artist chooses to give up some degree of control to an external system and thus, the artwork will then result from more than the artists’ input and intuitive decisions.

For Philip Galenter the motivation for randomization in art, and with this, the usage of randomization and generative techniques derives from complexity science and complexity theory. The idea is that complex systems have generally a certain degree of randomness, thus, to create truly complex art, it makes sense to include some randomness in the process.[2]

However, not all types of randomness imply complexity in the context that would be used in art, where complexity is often associated to being interesting and communicating higher order information. Thus, there is a need to find a way to quantify complexity with a measure that is appropriate in the context of art.

The usage of Shannon’s entrophy was quickly abandoned as it did not fit the artistic definition of complex, as highly random processes do not necessarily generate the most complex or interesting pieces.[2]

From an intuitive point of view, working artists understand that an audience will quickly be tired of both a highly ordered and a highly disordered aesthetic experience because both lack structural complexity worthy of their continued attention. The intuition that structure and complexity increase somewhere between the extremes of order and disorder fits in with the idea of effective complexity, introduced below.

What seemed to be more relevant in the context of art was Murray Gell-Mann’s idea of effective complexity. In effective complex systems, the ones which are highly ordered or disordered are given a low score, indicating simplicity, and systems that are somewhere in between are given a high score, indicating complexity.

To measure effective complexity Gell-Mann proposes to split a given system into two algorithmic terms: the first algorithm capturing structure and the second algorithm capturing random deviation. The effective complexity would then be proportional to the size of the optimally compressed program for the first algorithm that captures structure. The random aspects are forgotten and the aspects that exhibit structure are compressed (abstracted and generalized). Structural aspects that resist compression are then experienced as being complex. Thus, highly ordered systems from nature such as crystals, or highly disordered systems such as atmospheric gases, yield low measures of effective complexity, however, the adaptive systems found in nature, the living things biology have high effective complexity.[3]

One of the objectives of generative artists, as mentioned above, is to introduce new elements to their art and use techniques that are originated in science, but another one that is mentioned in [2] is that generative art is seen as a way to bring artists and scientists to collaborate more.

It is noted that, despite the existence of a trend towards technology-based art, the fundamental philosophies of arts, humanities and science in the 20th are very distinct. It is hoped that generative art, due in part to its technology based nature, to be an interesting area where both artists and scientists can exhibit and work in tandem.[2]

What I would say is the corresponding to generative art within the scientific community is the sub-field of Artificial Intelligence called Computational Creativity. In Computational Creativity, one of the main interests is the answer to the following question: “under what circumstances (if any) is it appropriate to describe the behaviour of a computational system as creative?

There are three main motivations for the study of Computational Creativity[4]:

  1. To provide a computational perspective on human creativity, in order to help us to understand it (cognitive science);
  2. To enable machines to be creative, in order to enhance our lives in some way (engineering);
  3. To produce tools which enhance human creativity (aids for creative individuals).

The theory Computational creativity theory (CCT) is seen as the analogue in Computational Creativity research to Computational Learning Theory (CLT) in the field of Machine Learning. In Computational Creativity research, the question that is asked is “What does it mean to say a computer has created something?”, whereas in Machine Learning, what is largely studied is how to induce a concept definition (or set of rules) to fit data which has been given in advance, and to use this learning to classify new examples with respect to the learned concept. Machine learning enabled the formalisation of acts of learning within Computational Learning Theory. Dr. Simon Colton formalizes the idea of CCT: the aim of this theory is to provide a rigorous, computationally detailed and plausible description of how creation can be achieved.[5]

Some more technical problems are also part of this sub-field of research, such as how to teach a physical robot to paint, how to formalize a creative process or how to represent various aspects of art (let this be video, images, colours) in a computer. By treating these problems, this research field becomes multidisciplinary, involving concepts from Mathematics, Computer Science and Mechanical Engineering, to name a few.

[1] P. Galanter (2003), What is Generative Art?, Available:

[2] P. Galanter (2003), Complexism and evolutionary art in The Art of Artificial Evolution, Springer, New York, 2008, pp. 311 – 330

[3] M. Gell-Mann, What is complexity? in Complexity, John Whiley and Sons, Vol 1 No 1, 1995

[4] S. Colton, On Impact and Evaluation in Computational Creativity: A Discussion of the Turing Test and an Alternative Proposal, In Proceedings of the AISB symposium on AI and Philosophy, 2011

[5] S. Colton, J. Charnley and A. Pease, Computational Creativity Theory: The FACE and IDEA models, In Proceedings of the International Conference on Computational Creativity, 2011

Creative process

Today I tried to draw. When I was younger, I used to draw a lot – just things I saw or from my imagination. But quickly I was told a good artist is not the one who draws well but the one who says something through their drawings. Nowadays I find myself trying to draw something with a meaning. Of course, I still think they should somewhat aesthetically pleasing, but there is more to it than nice artwork – rather, the message matters more. I try to think of the ones I admire – Magritte, Kandinsky, Dali, Beksinki, Munch, etc…

One of the last paintings that stuck on my memory was “Dead Child” by Stancic. ( To me there is something painful and disturbing about this painting, the contrast between the soft warm pallete and the concept portraited but you know, I hate talking about paintings because I think it’s a personal thing and I inevitably end of listening to my own words spat out in some attempt of showing understanding and verbalising things that I can’t verbalise that well. Another painting I recall causing a tremendous impression on me was by Paul Cezanne, named “Old woman with a rosary” (

I attribute the quality of a “good painting” to the feelings they cause in me. Of course, for instance, Leonardo Da Vinci’s works on anatomy ( are absolutely brilliant but it’s a different kind of admiration – towards the technical side and scientific genius. Same for a few of the artists I mentioned above – I admire them for their technical ability and composition but not necessarily for the in between lines message.

I find it extremely hard to paint something that has some extraordinary meaning and when I consciously try to do it, it feels rather fake. I have the often feeling that these concepts I try approach are overly exploited and overused. Love, death, pain, moral, what else? I suppose good concepts don’t just appear without any attempt. Having seen Miroslav Sutej’s exhibition opened my eyes to the hard work of artists who take art seriously, by seeing his attempts after attempts to capture a good enough piece. When I was younger I decided to do science because I thought art was too linear and easy – perhaps now I can see the challenges here and moreover, the similarities in the process of coming up with something new, between art and science.