https://raw.seadn.io/files/a1cbfdd041a3ab4ebe8e76daa56885b8.svg

After my graduation from art school in 2006 – when one couldn't make money with digital art yet – I worked as a web developer for various smaller and larger startups.

My task was to receive designs and turn them into dynamic websites, populated with text and images from awkward content management systems.

At that time, one of the greatest virtues expected from me, was that the designs would be translated in a ‘pixel perfect’ manner, and they should display identically regardless of the browser used.

Days were spent making sure the forms and interactions also worked on the incompatible Microsoft Internet Explorer, and that the webpages would look and behave the same on that abomination of a browser.

When making digital art, or more specifically, art that is experienced using a web browser, the same expectations of uniformity across browsers and platforms seem to prevail. My artworks, which often use quite demanding GPU accelerated graphics, or obscure quirks, should also run on the old iPad of my parents with its cracked glass touch surface.

My new Venster NFT collection depends heavily on features offered by the Scalable Vector Graphics standard (SVG), an open standard developed by the World Wide Web Consortium. Even though this standard has been around since 1999, it became clear to me that my admittedly unorthodox use of this declarative graphics language is still interpreted quite differently across contemporary web browsers.

I noticed that now, in 2024, these various browser implementations (such as Chrome, Firefox and Safari) just have different opinions about how SVG should render, even when I correctly adhere to the documented standard. Perhaps the guidelines for the rendering of visual aesthetics are more complex than what crystallized in the specification, and they leave space for interpretation?

I considered stretching my Solidity and SVG programming skills further, and restructure the graphic document, in order to make the artwork render identical on all browsers, but it required a lot more on-chain calculations, and as the source for the artworks is calculated by Solidity code, rendering would quickly run out of gas and not display anything at all.

This led me to contemplate a larger question: is it actually my responsibility as artist to adhere to the expectations of the off-chain display software stack, such as your browser, your operating system, your screen size?  Should it be responsive to all monitor aspect ratios? Should it display differently when your computer is set to “dark mode”? Am I only allowed to use features that render the same on all devices and browsers, or could I as artist prescribe the requirements for ideal display?

I don’t know.

But what I did find hugely fascinating was to view one single artwork in different browsers and see it render differently. In particular the mixing of nested blending modes, and the use of filters were interpreted in various interesting ways. Sometimes, the GPU even started to glitch with red pixel noise (at least on my computer), so I decided to deem the varying SVG interpretations a feature of the artwork.

Over time, browsers are of course still developing. Standards might (hopefully) not change, but their interpretation and support do. Long term, we cannot know which standards will remain commonly used, and which ones will be forgotten. JPG was amazing when invented, and still widely supported, but has been superseded by newer image formats in quality and compression efficacy. Will support for SVG cease at some point in the future, as it happened to Macromedia Flash not too long ago?

I invite you to watch the 100 Venster tokens on all your devices, browsers, and operating systems. Perhaps there is not one single canonical rendering, but all are subjective temporal performances of one and the same score.


After my graduation from art school in 2006 – when one couldn't make money with digital art yet – I worked as a web developer for various smaller and larger startups.
My task was to receive designs and turn them into dynamic websites, populated with text and images from awkward content management systems.
At that time, one of the greatest virtues expected from me, was that the designs would be translated in a ‘pixel perfect’ manner, and they should display identically regardless of the browser used.
Days were spent making sure the forms and interactions also worked on the incompatible Microsoft Internet Explorer, and that the webpages would look and behave the same on that abomination of a browser.
When making digital art, or more specifically, art that is experienced using a web browser, the same expectations of uniformity across browsers and platforms seem to prevail. My artworks, which often use quite demanding GPU accelerated graphics, or obscure quirks, should also run on the old iPad of my parents with its cracked glass touch surface.
My new Venster NFT collection depends heavily on features offered by the Scalable Vector Graphics standard (SVG), an open standard developed by the World Wide Web Consortium. Even though this standard has been around since 1999, it became clear to me that my admittedly unorthodox use of this declarative graphics language is still interpreted quite differently across contemporary web browsers.
I noticed that now, in 2024, these various browser implementations (such as Chrome, Firefox and Safari) just have different opinions about how SVG should render, even when I correctly adhere to the documented standard. Perhaps the guidelines for the rendering of visual aesthetics are more complex than what crystallized in the specification, and they leave space for interpretation?
I considered stretching my Solidity and SVG programming skills further, and restructure the graphic document, in order to make the artwork render identical on all browsers, but it required a lot more on-chain calculations, and as the source for the artworks is calculated by Solidity code, rendering would quickly run out of gas and not display anything at all.
This led me to contemplate a larger question: is it actually my responsibility as artist to adhere to the expectations of the off-chain display software stack, such as your browser, your operating system, your screen size? Should it be responsive to all monitor aspect ratios? Should it display differently when your computer is set to “dark mode”? Am I only allowed to use features that render the same on all devices and browsers, or could I as artist prescribe the requirements for ideal display?
I don’t know.
But what I did find hugely fascinating was to view one single artwork in different browsers and see it render differently. In particular the mixing of nested blending modes, and the use of filters were interpreted in various interesting ways. Sometimes, the GPU even started to glitch with red pixel noise (at least on my computer), so I decided to deem the varying SVG interpretations a feature of the artwork.
Over time, browsers are of course still developing. Standards might (hopefully) not change, but their interpretation and support do. Long term, we cannot know which standards will remain commonly used, and which ones will be forgotten. JPG was amazing when invented, and still widely supported, but has been superseded by newer image formats in quality and compression efficacy. Will support for SVG cease at some point in the future, as it happened to Macromedia Flash not too long ago?
I invite you to watch the 100 Venster tokens on all your devices, browsers, and operating systems. Perhaps there is not one single canonical rendering, but all are subjective temporal performances of one and the same score.

Venster

https://eu-central-1.linodeobjects.com/harm/SamKHtZWoby1anXVaKlTe/triplet.html

Say, we imagine a set of things in front of us. This set could have the shape of a cloud. The members of the cloud are the result of our attempts. They are connected to each other. Some connections are weak, some are strong, some attract, some repel.

In absolute terms it's difficult to say what does and what does not really belong together.

It would be more adequate to speak about their associations in relative terms: some elements are more related than others.

For example, let's take a set of three: *Palindrome*, *Twelve*, and *Energy Drink*.

The Energy Drink stands out for some particular criteria (possible to observe without requiring a definition of similarity). Because of this Twelve and Palindrome are automatically related, and there are negative ties to Energy Drink. Effectively three new relations have now been defined, and we have not sought refuge to seemingly unavoidable proxies such as categories, groups, tags, etc.

(Formulating coherence criteria using written language always reduces essential complexity because this language requires a mode of representation and interpretation)

The complex graph of 'weighted' connections is permanently incomplete. Taxonomies are ridiculous: there is no map, no overview, no complete model. Every node has character, partially (but maybe completely) defined by it's associations. Identity is deferred. We are somewhere, we can navigate elsewhere, and the destination emerges from our move. It wasn't there before and we weren't there (…) and upon arrival its novelty fades immediately.

(Originally published on [Dissociations](http://dissociations.com/952))

Dissociations

https://eu-central-1.linodeobjects.com/harm/lze6zr7eJqcjr721D5HZd/spike-70-harm-van-den-dorpel.pdf

In this conversation, curator Tina Rivers Ryan and I explore generativity, in every sense of the term. Can it be an entry-point for tokenised work to stay fluid and subvert the market? 

Originally published in Spike Art Quarterly #70 from January 2022.

[Download PDF (4.6 MB)](https://eu-central-1.linodeobjects.com/harm/lze6zr7eJqcjr721D5HZd/spike-70-harm-van-den-dorpel.pdf)

The Squircle of Life

https://maschine.harmvandendorpel.com/artwork/index.html?tokenId=7

Artistic inspiration sometimes finds me in unlikely places. In the case of the present work I’ve completed as a part of a Mercedes-Benz commission, I found myself increasingly fascinated by the kinetic tension quietly whirring away in the margins of certain music videos. Though I haven’t been a particular aficionado of the automobile growing up, I have always loved and been inspired by music. My tastes are quite varied and voracious and as I thought about this commission I found myself increasingly drawn to hip hop videos where often high performance cars play a central role as aspiration, inspiration, and sometimes as alternative protagonists. The ways in which hip hop stars and music video directors have used cars and their accessories is as varied as the genre itself, but I became particularly interested in the strange visual kinetics of the ways in which tyre rims would sometimes behave in these videos. As a car appeared to accelerate, for example, glittering spinning rims inside the wheel would often appear to slow down and move in an antagonistic direction. The image worked as visual paradox, but also as physics puzzle. 

My background in computer science has often brought elements of advanced mathematics and other scientific fields into my range of influences and reference, and in this case the hub-and-spoke structure in which rotating projections spin at the edges of an essentially stationary hub, creating an effect defined by both stillness and motion. Physically, the concept is easy to describe, but visually it can produce a number of interesting effects, especially when combined with advanced technologies. In conceptualising this work, I thought about the ways in which a spinning image could create sites of tension or apparent paradox. In my own works in general I have largely focussed on the nature of the 2-D image and how technology could facilitate engaging visual dynamics in an intrinsically illusory space. Advancing technology, and my own sense of restlessness in pressing media to, and beyond, certain limits meant that I became curious about the way 3-D imaging and 2-D imaging could work together to create effects that were not simply immersive ‘environments’ which, though increasingly sophisticated, rely on familiar expectations which make it difficult to avoid cliches or established visual tropes. 

I built on the idea of rotations and counterrotations as seen in the music videos but ultimately I found the notion of the spiral - as a fundamental form - more interesting than any particular manifestation of it. The spiral form is not merely a visual structure, it is also an informational structure; one may think of the groove in a vinyl record, or the spinning of a hard drive in a computer’s housing as other ways in which information, spirals, and aesthetic relationships combine. With this dynamic in mind, I felt I had a way into the project that was both material, conceptual, and emotionally weighted. I began the work of producing images by taking photographs of the rims of various models of Mercedes-Benz, perhaps to the bemusement of the workers who let me into the storage rooms, and I sought a method and a lineage for making the resulting images less about the physical reality of a car or its accessories, than about the spirit that cars often signify in our culture. 

A technical note about the production of the work is valuable here. In producing the works, I considered the ‘realistic’ qualities of the 3-D image and sought ways to challenge the expectations such imaging-making evoke. Specifically, I created an effect that would produce a kind of tracer in the 3-D image which would leave behind a trail of light as the image rotated. This visual effect recalled, for me, a more 2-D aesthetic in which the ‘realism’ of the work is less about visual fidelity than about conceptual fidelity. The spinning wheel in a painting spins in our minds; in a 3-D version, the wheel just spins. I sought a way of weaving these two regimes of seeing together so that the work contained elements of 2-D aesthetics and 3-D aesthetics. Later, I explored the ways in which different materials produced different iridescence and light fracturing under the effects and I arrived at a sense of how the works could proceed. 

Musical reference points recurred to me again once I had a sense of my approach for the images. I began to develop something like a visual synthesiser. Rather like the infinite possibilities that emerge from a musical synthesiser by way of recombinations of three essential parameters (filtering, amplification, and signal oscillation) I produced a simple set of visual parameters I could adjust to alter the images. The scale of the project however, producing 1000 interesting images for the commission, was beyond my physical limits, and so I turned to another technology for help: neural networks. My work regularly engages with forms of machine learning, but instead of sophisticated algorithmic dimensions, I used a kind of ‘off the rack’ neural network consisting of only three layers (an input, a single ‘hidden layer’ where operations take place, and an output layer), and submitted the images to the network with parametric input from me regarding which images I preferred. 

The results astonished me. Not only were there now thousands of powerful images, but the effects created were often surprising even to me. An additional appeal for the computer programmer inside me was to think of the neat irony of using a GPU to run the graphics on the computer as well as a GPU to train the neural network. Efficiency can emerge from the strangest places. 

As I was producing the images I thought of how they fit into both an aesthetic lineage, and how they could be situated within the cultural understanding of Mercedes-Benz. In the case of the former notion, I recalled the work of an early pioneer of abstraction, [Frantisek Kupka](https://www.are.na/block/20478433). Kupka’s works, rather similar to the works of other early abstractionists including Hilma af Klint and Wassily Kandinsky touch on themes of transcendence and divinity, an escape from earthly strictures. This longing, and its frustration, became touchstones for this work, particularly in light of the ways beliefs about technology and its capacity to liberate have played out over the last century. 

Movements such as the Futurists or the Vorticists, inspired as they were by early technological advances in the 20th century, embraced a one-sided approach which fed into the grotesque political viewpoints which have come to define their legacies. To believe such mythology about the capacity for technology to liberate was naive in the extreme then; today it is simply fatuous. Technology may well move us forward, but it may also move us backward - perhaps even at the same time. These are the kind of contradictions that animate my work. Nevertheless, there is an evocation that speaks to people in the potential for advanced technology to empower, and to liberate. Mercedes-Benz has often expressed these themes in their creations, combining precision engineering with narrative. I am grateful for the opportunity to have explored these ideas and relationships as a result of this commission. 

– Harm van den Dorpel, as told to [Habib William Kherbek](https://www.abstract.supply/product/entropia)

Maschine

https://openseauserdata.com/files/ef2217c41ee20180038f97ad94385e02.svg

Really pleased how Joseph Del Pesco formulated my artist bio 💜 :

> Harm van den Dorpel’s practice focuses on emergent systems and the role technology plays in their development and meaning. Engaging with diverse materials and forms, including works on paper, sculpture, computer-generated graphics, and software, van den Dorpel’s works are continuously evolving, informed by feedback loops and the design of algorithmic systems. Working within and beyond the lineage of ‘net art’, a core aspect of van den Dorpel’s practice is software development that addresses specific approaches to artificial intelligence. With immense skill and craftsmanship, he builds advanced systems that draw on intuition and subliminal processes of the mind in order to continually output unexpected and curious aesthetic forms that embody a feeling of subconscious computation. 

https://kadist.org/people/harm-van-den-dorpel/

Artist Bio

https://outland.art/harm-van-den-dorpel-jenna-sutela/

A conversation between Jenna Sutela and me, moderated by Brian Droitcour, published on Outland.

> Randomness is often lacking in computers. Computers have a clock made with a quartz crystal that vibrates at a stable rate, but it doesn’t do it perfectly. So random numbers are often taken from the imperfections of those vibrations. For Mutant Garden, I used a mathematical number generator, which returns random numbers, but always in the same order. It had to be deterministic, or else each mutant would be different every time it was rendered, and there would be no point in connecting it to the blockchain.

https://outland.art/harm-van-den-dorpel-jenna-sutela/

Harm van den Dorpel & Jenna Sutela

https://openseauserdata.com/files/ddba5121b4b01c67ca2bb5fd8d105dd7.svg

I'm turning 42 tomorrow (only one year to go before my age is a prime number again), and I thought it would be nice to celebrate by porting and releasing a software work that I once created in 2008, which never got much attention. Besides, I have COVID so I can't invite friends over. I feel OK though.

Initially, I wrote this piece called "Field" in Actionscript: the programming language of Macromedia Flash. At the time, so much more was possible in this language in terms of animation, video and sound compared to Javascript. Adobe bought Macromedia in 2005 and shut down Flash in 2020. Even though I think they made the right call, it's a loss that they withdrew the Flash player altogether. A rich heritage of interactive online net art became instantly inaccessible.

The 2023 version of my art piece lives entirely on the Ethereum blockchain. The code that generates those animated spinners is fully contained in the smart contract. Unlike the majority of NFTs, the artwork itself is not hosted on a web server or IPFS and has no external dependencies. As long as Ethereum exists, and our computers are able to display SVG, it should be fine.

There is something ironic about porting graphics from the intuitive language that Actionscript was to the rigid unforgiving smart contract language Solidity, which was developed to support (economic) transactions.

From this artwork, I have minted 6 unique tokens. The odd token numbers repeat one kind of spinner, each with a different rotation speed. The even numbers show a range of various spinners, small and large, with many or few spokes. 2 tokens have a dark background, 2 a classic grey background, and 2 a gradient background that responds to the appearance settings of your operating system.

Spinners are shown when UX designers want to make clear that you need to wait for something. I came to realise in particular now in my early 40s, that one cannot suspend being. The separation between productive time, and time supposedly "on hold" is entirely based on my own perception and interpretation. I am free.

The first time I put spinners in a two-dimensional grid, it struck me that they become something like flowers. Or wind turbines. We find ourselves in a large open field: a field without boundaries.

*First tokens will be auctioned on Zora from midnight CET (6PM EST) when it turns 8th of February.*

[View token overview...](https://harm.work/series/field)  

[OpenSea](https://opensea.io/collection/field-harm-van-den-dorpel) | [Smart contract source code](https://etherscan.io/address/0xc1cb5ea3a3541af83b0417dc26e04628193a53d3#code)

(by the way: you can touch the spinners with your cursor, it excites them)

Field ✲ 2008 – 2023

https://eu-central-1.linodeobjects.com/harm/tlg8sG4b-kPqSJaZpxVMe/Screenshot%202023-02-05%20at%2001.33.57.png

It's fascinating and exciting (and at times demoralizing) to follow the surging interest in NFTs (non-fungible tokens).  People have approached me asking for my opinion on them, given that I run [left gallery](https://left.gallery). So, here is my highly personal and entirely subjective take on the matter.

### History

I minted my first NFTs in 2015 with the help of the now defunct [ascribe.io](https://www.ascribe.io/). Their pioneering service allowed creators to upload their work, mint editions on the Bitcoin blockchain, and to transfer them as tokens when sold.

Even though, at the time, this sounded like an abstract and intangible fad, the Museum of Applied Arts in Vienna did buy 20 tokenized editions of my work, [Event Listeners](https://left.gallery/work/event-listeners), through the marketplace [Cointemporary](https://twitter.com/cointemporary). The Museum paid with Bitcoin, which made it a historical transaction: it was the first time a museum bought NFTs, and paid with a cryptocurrency. The event gained quite some [media attention](https://www.artnews.com/art-news/market/mak-vienna-becomes-first-museum-to-acquire-art-using-bitcoin-a-harm-van-den-dorpel-3995/), and, gradually, all my 100 editions were sold to collectors and another museum.

This unexpected success inspired my partner, [Paloma Rodríguez Carrington](http://frameshift.berlin/), and me to start our own marketplace for blockchain ascribed artworks: [left gallery](https://left.gallery). We would sell "downloadable objects'' in large editions for low prices. Customers could pay with crypto, PayPal, or credit card, and, after purchase, we would manually transfer the token belonging to their purchase to those clients interested in having them – most didn't understand what a blockchain was at all.

The insights of ascribe.io were radical and definitely ahead of their time. Unfortunately, the state of blockchain technology then was not mature enough to support their business model; ascribe stopped offering their service around 2017, and they pivoted to become a database company. 

Although the token provenance information that ascribe stored on the immutable Bitcoin blockchain will always remain there, in practice, we have lost access to it, as their web interface to retrieve it was discontinued. So, in 2018, I bit the bullet, deployed an [ERC721 smart contract](http://erc721.org/) and minted Ethereum tokens for the editions that had been sold. I asked owners to send me their Ethereum wallet address in order to transfer them the tokens for the editions they owned. This was, in itself, an "interesting" situation, because this meant that some editions were now tokenized on two blockchains at once.

To my disappointment, when I announced this new Ethereum website integration, few buyers actually requested the token belonging to their purchase. Sales on left gallery continued, but interest in ERC721 tokens kept declining among our clientele. At the same time, the transaction costs kept increasing, to a degree where the costs for minting and transferring tokens exceeded the sales price of the actual artworks by orders of magnitude.

### 2020

Then came COVID19. The income of my partner and I dropped significantly, as we had fewer clients and the art market was on hold. This unusual moment created the opportunity to redevelop the left gallery website (from scratch) using new technology, and support ourselves and other artists by increasing gallery revenue.

As only a few buyers in the previous years had shown interest in owning an NFT for their purchase, I hesitated to implement an integration with the Ethereum blockchain _again_. This meant that left gallery now became a regular online marketplace where ownership records were stored in a regular database. I was proud of what we had achieved with the left gallery relaunch, and I was amazed by all the works that artists offered on it, but also it hurt, because left gallery had lost some of its "blockchain soul".

### 2021

To my surprise a renewed interest in NFTs emerged in the first quarter of 2021, which, to be honest, initially annoyed me. In my humble opinion, the works now offered as NFTs were not very interesting, or just plain ugly. Why were people suddenly willing to pay so much Ether for such bad "art", while they often hesitated to pay for the affordable, non-NFT artworks on left gallery? It seemed to me this hype surrounding NFTs was not grounded in a genuine appreciation of the artworks.

Still, the rush toward NFTs turned out to be an opportunity for me as well, as I had some old, and newer, artworks that never found their ways to art collectors, but which suddenly very well fitted the editioning, format, and price range of NFTs. So I did profit from this second(?) wave of NFT. However, as the general interest in NFTs has grown explosively, so did questions about its environmental impact.

### Waste

I find it hard to judge exactly how wasteful the Ethereum blockchain is, and how, in relative terms, environmentally destructive it is compared to, for example, the Google data centres that power our daily search queries, or CI/CD pipelines, or credit card payment networks, etc. I wonder why the preponderance of general opinion questions their use of electricity much less than that of NFTs. Many people lack a deep technical understanding of computing: they do not realise that there are many other (non-NFT) smart contracts running on the same blockchain, and the popularity of NFTs does not correlate to the general energy use of the validators. Were all artists to abandon Ethereum en masse, it would not make our planet greener.

However, I think that it’s hard to deny that the current Ethereum blockchain is wasteful _by design_. The "proof-of-work" (PoW) consensus algorithm ingests huge amounts of electricity to crunch meaningless mathematical riddles, which secure the network from fraudulent transactions. Additionally, PoW is limited in the amount of transactions it can process per second, and transaction costs can become ridiculously high.

These limitations have been known since the inception of the Ethereum blockchain, and its developer community has already gone a long way in replacing the PoW algorithm with another algorithm: proof-of-stake (PoS). I will spare readers my attempts to explain how these two algorithms differ technically, but the takeaway is that PoS does not depend on computing power to solve random math puzzles, and, therefore, burns far less electricity than PoW. A blockchain running on a PoS algorithm can handle many more transactions in a given amount of time; transaction costs might be low, or even zero. Even though PoS is not a silver bullet, it is my understanding that abandoning PoW in favour of PoS will improve on the environmental issues associated with NFTs.

### Future

The Ethereum website [promises to complete the transition](https://ethereum.org/zh/eth2/) to proof-of-stake in 2022, which will "signal the end of energy-intensive mining". This is hopeful, but how can we be sure they will meet this deadline? Are there feasible alternatives?

I was delighted when our [funding application was granted](https://interchain-io.medium.com/icf-q2-2020-funding-recipients-e5cbb326c23c) by the Interchain Foundation for the development of an "application-specific" proof-of-stake blockchain. Where Ethereum hosts many unrelated applications all on one blockchain (competing for scarce resources), an application-specific blockchain consists of only one application, fine-tuned to its unique requirements. We will be developing such a blockchain using the Cosmos SDK. The long-term vision of the Cosmos community includes connecting many of these isolated blockchains, forming an “Internet of blockchains”.

We will eventually find out if the current renewed enthusiasm for NFTs will stabilise and form a sustainable market that will benefit artists over the long term. Practically, I will continue to use the Ethereum blockchain for selected artworks, in anticipation of the announced transition to the environmentally more sustainable PoS algorithm, while simultaneously developing our own light-weight blockchain.

– Harm van den Dorpel, edited by William Kherbek


Tokenizing Sustainability

https://mutant-garden-tracer.harmvandendorpel.com/

### In my quest to create computer programmes that generate aesthetic systems with emergent complexity, I thought it would be interesting to create software that mutates computer code, as opposed to software that “just” mutates data. However, the immediate challenge that presented itself was that most mutations of computer code cause infinite loops, stack overflows, rainbow spinners <img style='width:18px' src='/images/spinner.png' />, blue screens of death, and core dumps. I basically stumbled upon the notorious [halting problem](https://www.youtube.com/watch?v=eqvBaj8UYz4).

### In 2019 I encountered an algorithm developed by Julian F. Miller and Peter Thomson, which was later dubbed “[Cartesian Genetic Programming](https://harm.work/downloads/Cartesian_Genetic_Programming.pdf)” (1999). After some studying I was able to write my own implementation in TypeScript. By defining a visual programme as a two-dimensional grid (layers and nodes) I could safely mutate without ever getting stuck in an infinite loop. 

### I was delighted when Julian agreed to an interview!

---

### Dear Julian, could you tell me about your personal, academic and technical background?

When I was a young man, I wanted to be a theoretical physicist. My boyhood hero was Albert Einstein. My bachelor’s degree was in Physics. After that I obtained a PhD in mathematics. The subject of my PhD was the interaction of strange waves called solitons. I thought they were good candidates for elementary particles. The equations describing how solitons interacted could often only be solved numerically on computers. So, I wrote computer programs to do this in Fortran and Basic. After my PhD I became gloomy that this line of research would be fruitful. 

For various reasons, I moved to Scotland and obtained a job in Napier University helping staff in the electronic engineering department developing computer programs mainly to create algorithms to build electronic circuits efficiently. One day in 1994 a colleague mentioned Genetic Algorithms. I liked the idea and tried it on circuit optimization. It immediately worked well and soon led to many publications. This was a life-changing moment for me.

### Could you explain what genetic programming is, and more specifically “Cartesian Genetic Programming”?

Genetic programming (and CGP) uses a simplified model of Darwinian evolution to design computational structures that solve many kinds of problems. The designs are represented as a string of information a bit like a biological chromosome. Like biology, these computational structures are collections of smaller calculations which are called genes. The artificial chromosomes can be slightly altered – this is called mutation. Also, parts of artificial chromosomes representing two different designs can be mixed rather like sexual recombination in biology. A population of these chromosomes is created (randomly at first) and at each generation all the chromosomes are tested to see how well they solve a problem. The ones that are not so good are rejected and replaced by fitter chromosomes some of which have been created by mutation or recombination. Over many generations one obtains chromosomes that can solve problems very well. GP allows one to solve problems using computational structures that have been automatically evolved and solutions can emerge that humans would not think of. In this sense it is a creative process.

There are a number of “flavours” of GP: tree, graph, stack and grammar. Each represents a computational structure in a different way. Cartesian GP is a graph based form of GP. Graphs are networks of computational nodes (like add, subtract, multiply etc.). The calculations done in the nodes can be re-used by other nodes. Also the graphs can have many outputs meaning that more than one problem can be solved at the same time. In CGP, there is a decoding process that can ignore some genes (this also happens in biology), these genes are called non-coding and they do not affect the computation carried out in the graph. However, the non-coding genes are still very useful as mutation changes them and after a few generations new graphs emerge.

### Even though CGP is very versatile, were you developing CGP with a specific application in mind?

As I mentioned earlier, when I first heard of genetic algorithms I was working on ways of making electronic circuits using as few logic gates as possible. My colleague Peter Thomson and I came up with a simple way to represent circuits as a graph. I gradually developed this into a mature form that could solve many different kinds of problems. The computational structures used a Cartesian grid of computational nodes so I called the technique Cartesian Genetic Programming.

### If we consider your algorithm itself to be part of a lineage, or even genealogy, could you point at its predecessors and descendants? Has your algorithm replaced others, and does CGP still have a place in a world of AI dominated by neural networks?

Genetic programming originated from a number of early pioneers. In 1958, R. Friedberg created an algorithm that could make a few random changes to a computer program and then test the mutated programs to see how they performed. In 1980, S. F. Smith utilized a form of GP in his PhD work. In 1981, Forsyth created a GP system called BEAGLE which could predict recovery after heart disease, and British football results. There are many early pioneers. There is a brief account of this lineage in [the book Cartesian Genetic Programming](https://www.springer.com/gp/book/9783642173097) which I edited and was published in 2011. However, when I proposed CGP, the dominant forms of GP were tree-based as described by John Koza in 1992 and Linear GP as described by Peter Nordin and colleagues in 1994-1995. In the 1990s other forms of GP were proposed: Push (1996), Grammatical evolution (1998) and Cartesian (1997). GP took off after John Koza’s book and the first international conference in GP was in 1996.

Since its inception CGP has given rise to a number of variants. These have been detailed in a recent paper of mine called “[Cartesian Genetic Programming: its status and future](https://link.springer.com/article/10.1007/s10710-019-09360-6)”. Self-modifying CGP where the evolved programs can change themselves (SMCGP). Differentiable CGP, modular CGP and positional CGP are all variants of CGP. Thinking about CGP and Art, SMCGP could be a method of evolving an entire gallery of art rather than one at a time as SMCGP produces a sequence of programs from one chromosome. Each of these could generate art.

CGP is certainly the most widely used graph-based form of GP and its popularity continues to grow. CGP definitely has a place in the neural network dominated world of AI. Firstly, CGP can encode neural networks and there has been quite a bit of recent work on that. Also, currently I am working on a CGP system that builds multiple neural networks so that it can solve a number of problems at the same time. It is called DEMANNED which stands for designing multiple artificial neural networks using evolutionary development. It evolves CGP chromosomes that represent a neuron and its dendrites and when the evolved program is run it builds multiple neural networks each solving a different problem. I believe CGP has a bright future in AI.

### As a contemporary fine artist, I’m often asked about the moment the “inspiration” behind the artwork “came to me”. I’m using quotation marks here, because I feel that people working in other fields sometimes attribute too much magic to inspiration, while in reality my creative process hinges on the “Try again. Fail again. Fail better.” adage. I personally experience the most creative magic in this feedback loop of trying and failing, 

### I am convinced that computational science is a highly creative field as well. Could you tell me how the concept of CGP “revealed itself” to you? And more general, what is your attitude towards freely having novel ideas vs. disciplined implementation?

I totally agree that the creative process is a series of inspirational moments and a huge amount of experimentation with many cycles of hard work. CGP started with a discussion between myself and my colleague Peter Thomson while we were at Napier University. Initially it was intended only for evolving electronic circuits. The way that non-coding genes are used in CGP was originally there purely because it made it easier for me to write the program. I only discovered the extreme importance of non-coding genes for evolution later on. Within a couple of years, I realised that CGP could be applied to many kinds of problems and that graphs have many advantages over other representations. Over the early years I simplified the representation a bit and I and my students and colleagues began to analyse many aspects of CGP. There were many “Aha” moments in the development of CGP. The idea behind capturing and evolving subfunctions or modules came to me in a moment of inspiration. Also how to create self-modifying CGP was an inspirational moment. Luckily, I am still having these inspirational moments and a future form of CGP may look quite different. For instance, in a promising new form of CGP all genes are floating point numbers (positional CGP), and the positions of computational nodes can be moved by evolution.

Science proceeds by occasional inspiration and long periods of analysis and development. The development of CGP followed this model. Thomas Kuhn described eloquently how Science happens in his wonderful book “The structure of scientific revolutions”. 

### The idiom surrounding genetic algorithms and evolutionary programming has imported a lot of jargon from evolutionary biology and genetics. This is helpful, because many terms such as crossover, mutation, breeding, generations, are mechanically similar to how they are coded in algorithms. However, sometimes I feel that by using these terms, we suggest that the systems we build are more “alive” than they actually are. Would you agree with that?

Hmmm, interesting point. Yes, lots of biological jargon has been imported into evolutionary algorithms. A little of this is fine, however, there are many bio-inspired methods these days that obscure the algorithms with too much jargon so much so that some of them are essentially not very different, but they sound different! I agree it can seem that we are building systems that are alive to some extent. Indeed in the field of artificial life researchers are deliberately trying to create systems with life-like properties. A classic example of this is the work of [Carl Sims in 1994](https://karlsims.com/evolved-virtual-creatures.html). Some of the evolved “creatures” seemed quite life-like.

### Explaining algorithms with these terms from evolutionary biology has a didactic benefit, but there might be more at play ideologically or philosophically. Could our quest for artificial life and intelligence reveal the human desire to create - or at least simulate - life in an effort to deal with our own mortality? 

I think that the quest for artificial life is more about trying to understand what life is. Either we can study and analyse biological life (a top-down process) or try to build life-like systems from the bottom up. Personally, my work in the field of evolutionary algorithms has led me to have a huge amount of respect and wonder at all living things. Life is a 3.5 billion year old technology!

---

Displayed left is the work [Mutant Garden Tracer](/work/mutant-garden-tracer) (not shown on mobile devices).


Thank you Wassim Z. Alsindi for proofreading.