further details


The predominant question that Pindar Van Arman, a clasically trained painter, seeks to explore through artistic experiments with his creative AI systems is: Can AI can paint like a human artist? Influenced by Minsky’s Society of Mind concept, Van Arman approaches creativity as something that is not a singular process but multiple ‘agents’ working simultaneously to discover emergent ways of seeing. Van Arman’s Reflective AI uses multiple algorithms that evaluate the structural, dynamic and aesthetic elements of the artwork as it is being created. These algorithms respond to the changes in real time, taking constant input from the entropic variation between the algorithmic intent and the unpredictable physical outcomes of each brushstroke. The ultimate goal of this approach is to create a synthetic embodiment of creativity, a system that replicates the reflective processes of an artist engaging with the canvas.

Central to his explorations is the concept of feedback. Early on, Van Arman faced a critical issue which he summarizes as, “My algorithms were painting blind… Instructions were sent to the robotic head but it painted with no feedback from how it was executing the painting…. To put this into perspective, try closing your eyes and drawing. Now open your eyes and do the same drawing. The difference in quality should be obvious”. To solve this problem, he introduced cameras and responsive algorithms to the robotic system, which enabled it to not just ‘see’ the work but also to respond to the changes instead of simply iterating. From this early basis, he has gone on to fine-tune his feedback loops around increasingly nuanced and reflective algorithmic decision-making processes.

Van Arman’s unique conception of synthetic creative feedback loops is deeply influenced by Paul Klee’s  writing on the painting process as a continuity. In his 1925 Pedagogical Sketchbook, Klee describes a creative process as a feedback loop : “Already at the very beginning of the productive act, shortly after the initial motion to create, occurs the first counter motion, the initial movement of receptivity. This means: the creator controls whether what he has produced so far is good. The work as human action (genesis) is productive as well as receptive. It is continuity.” Van Arman’s work attempts to translate this perspective to AI and machine creativity in order to discover new ways of seeing that are unique to computational creativity.

"The Thinking Eye"
Paul Klee

Another major influence on Reflective AI is Harold Cohen, with whom Van Arman was in correspondence toward the end of his life. One of the limitations to computational creativity that Cohen highlights was the deterministic nature of computing processes. In his essay “Parallel to Perception” he writes:  

By definition processing is a deterministic affair, and for any single run its functions are predetermined and invariant. Feedback from the result to the functions themselves has no part in this process. On the other hand feedback is clearly a part of the human art-making process, or indeed of any intelligent process, and if the only feedback possible within the computer environment is via the human user, then the computer is a tool in no essential way different from any other tool…. 

Based on this limitation, Cohen posed the question of whether computer-based creative systems could, without the feedback from a human user, be able to provide their own (feedback) material. In grappling with this question, Van Arman arrived at a unique way of integrating physical and digital input to both transcend determinism and to find a kind of emergence that is specific to the medium. For these synthetic creative feedback loops to emerge, Reflective AI leverages the unpredictable nature of high-viscous acrylic paints, paintbrushes and markers on a textured canvas. As the paint reacts to its application, it drips, blobs, blends and interacts with the canvas unpredictably. These serendipitous physical changes on the canvas then become novel digital inputs, which in turn influence the next step in the physical process, building a series of feedback loops that evince the machine system's own emergent way of seeing.

Harold Cohen coloring a design made by AARON
photograph from the Computer History Museum


As early as 2016, Pindar Van Arman began experimenting with blockchains, specifically Bitcoin, as a way of providing stakes for participation in his crowd-sourced painting project 'Bitpainter.' Subsequently, his dedicated foray into the world of cryptoart began as one of the first artists on SuperRare in 2018. It is from this time that Van Arman often recalls a dilemma that he faced. In his words: 

In the early years, we were all exchanging ideas, collecting each other’s works, and exploring what cryptoart was all about. While I was in the middle of all this, I had the problem that my work didn’t quite fit in. Everyone else was natively digital and taking advantage of the new medium with all of its advantages for digital artists. For me, however, it didn’t completely make sense. My finished canvases were still physical paintings created by robots, and my NFTs were simply photographs of them. There was nothing natively crypto about my cryptoart, and I often wondered if there was even a reason to make my paintings NFTs…

Faced with this conundrum of working with creative robots on physical canvases in a digitally native world, Van Arman searched for solutions to better adapt to the core medium of cryptoart, i.e., digital works on blockchain. The first solution he found were his bitGANs, which are strictly digital artworks. With Reflection, he re-unifies this digital approach with its physical source. The creative process of the Reflective AI system is such that the physical and the digital works are not separate but form a part and parcel of each other. The physical provides the serendipitous inputs for the creation of the digital work; while the digital is composed in a uniquely blockchain-native manner. This interplay, where the physical is not the end but a means to an end, blurs the lines between the physical and digital environments in Van Arman’s artistic practice. This, perhaps, is also a reflection on our own lives, where too the boundaries of the physical and the digital worlds are blurring and slowly transforming.


Reflection presents a unique perspective on long-form generative composition. Each artwork is generated on mint with a process that integrates both the transaction hash and a set of physical inputs from the robotic system. These are then composed using Solidity (Ethereum's programming language), and both the image and the code are stored in the smart contract, resulting in 100% onchain art with no external dependencies.


Further Reading


"Inside the Studio With an AI-Guided Painting Robot" Time Magazine
D.W.  Pine

"Can AI Create True Art" Scientific American
Ken Weiner

"You Can Give A Robot A Paintbrush, But Does It Create Art?" NPR



"Parallel to Perception"  link to PDF

Harold Cohen

"Colouring Without Seeing: a Problem in Machine Creativity" link to PDF

Harold Cohen

"A Self-Defining Game for One Player" link to PDF

Harold Cohen

"Where Does A.I. End and We Begin?" New York Times

Sougwen Chung

"Readyweights" Mirror

Holly Herndon & Mat Dryhurst

Conversation with Ira Greenberg and Ivona Tau  Outland

"Art’s Intelligence: AI and Human Systems" The Brooklyn Rail

Charlotte Kent

Pedagogical Sketchbook link to PDF

Paul Klee

"Creativity and Artificial Intelligence" link to PDF

Margaret A. Boden

"Bricolage Programming in the Creative Arts" link to PDF

Alex Mclean & Geraint Anthony Wiggins

"The Phenomenal Concept Strategy"  link to PDF

Peter Carruthers & Bénédicte Veillet

"Minds, Brains, and Programs" link to PDF

John R. Searle

"Creativity in the age of generative AI" Nature Human Behaviour

Janet Rafner, Roger E. Beaty, James C. Kaufman, Todd Lubart & Jacob Sherson

Categories  https://classics.mit.edu/Aristotle/categories.1.1.html