New Realms, New Questions

New Realms, New Questions

Aleppo-born artist Iyad Rahwan is based in Berlin. Drawing upon his work as a computer scientist and director of the Max Planck Center for Humans & Machines, his art explores the evolution of AI and its relationship to the human condition.

A former professor of Media Arts & Sciences at MIT, Rahwan incorporates the machine’s own evolving representations of the world and its emerging sense of self. His paintings of humanoid robots chronicle his conversations with increasingly-sophisticated AI chatbots, and capture the complex emotions that these intelligent machines evoke in humanity.

A: In Issue 116 of Aesthetica, we featured Machine 14 from the Faces of Machine series. You’ve noted that the series “delves into the essence of intelligent machines and the emotions they elicit within us – ranging from trepidation and unease to curiosity, hope, affection, awe and veneration.” What made you choose these words, these emotions?
Science fiction has long captured our ambivalent fascination with Artificial Intelligence – from the terrifying metallic soldiers of The Terminator series, to the lovable robotic child in Steven Spielberg’s movie AI. Today, AI technology gives us hope for new medical cures and for levelling educational opportunities via personalised tutoring.

But AI is also assisting legal authorities in reaching verdicts and firms in evaluating job applications, provoking apprehensions of automated prejudice. Chatbots are already assuming roles as romantic partners and counsellors, while concerns for human job security grow. Some people see AI as our saviour, liberating us from tedious work, potentially ushering in an era of universal basic income.

The Faces of Machine series aims to explore these emotions by giving faces to the mostly faceless AI algorithms that permeate our world. I want to provoke reflection on how the machines make us feel, and what we may or should expect of them.

A: The series was created with AI-based image synthesis models; what was the process behind this?
The creative process behind Faces of Machine starts with my own scientific research on the impact of AI on society. I might read a scientific article about a new AI algorithm that mimics human emotions, or about a new application of AI in policing or in making hiring and firing decisions in firms. I also build on my own research on how people want autonomous vehicles to resolve moral dilemmas about whose life to prioritise in an accident.

I then attempt to develop a portrait that captures an emotion or a dilemma about AI. I use AI image generation models—like DALL-E or Midjourney—to develop concepts and compositions for a painting. After generating hundreds of ideas, I select a few ideas, and alter them before working on the final physical oil painting.

For most artists who use generative AI today, the AI acts primarily as a tool for generating ideas. But to me, the AI image generator is itself a subject in the painting. When I prompt DALL-E or Midjourney to generate ideas for a machine contemplating a moral question or a robotic policeman exuding authority, I am interrogating the AI’s own representation of itself. And this AI self-image is, itself, changing as AI develops further.

Thus, the final oil painting acts as a record, not only of how we humans see the machine, but also of how the machine sees itself. And of course, the machine is trained on our own data, so there is a circularity that I find fascinating.

A: How does the use of oil on canvas panel help express your ideas for the series?
AI technology is developing at a mind-blowing speed. Scientific breakthroughs are happening in a matter of months and even weeks. Not only are AI systems becoming smarter, but their ‘personalities’ are also rapidly evolving.

For example, people have recently reported that the conversational AI bot ChatGPT has become too politically correct, and others think it got more lazy. Each conversation with AI is an ephemeral event, which exists due to a particular pattern of zeroes and ones on a giant computer somewhere. And because these systems are constantly learning, they are in flux.

In five or ten years, we will have vastly more powerful AI, and today’s impressive AIs will seem like primitive ancestors. The use of oil pigment, for me, is a way to immortalise a given moment in the evolution of these AI systems using a tangible and enduring medium.

A: Is there a significance in the use of linen as a natural, organic material?
No, at least not initially. However, your question has prompted some reflection. Now I would say that the tactile nature of oil paint, and the active movement of the brush, both say something about the ultimate human agency in building AI. The use of cotton and linen canvas, rather than a very smooth surface, adds to this physicality. It also emphasises the contrast with the synthetic nature of machine intelligence, which can be seen as a new life form.

A: Is painting a kind of respite from working with machines?
Absolutely. I have dedicated my life, as a scientist, to studying how intelligent machines impact our lives, and the existential and moral dilemmas they raise. But as a scientist, I have to work within the strict confines of the scientific method, and I can only focus on what I can measure and quantify with precision.

Painting liberates me from these restrictions, and allows me to explore AI on a more abstract level. My fascination with the machine becomes not only scientific and objective, but also emotional and subjective.

A: How do you approach texture and perspective within your work?
My approach to texture is influenced by early twentieth-century portrait artists like Sargent, Zorn and Sorolla, who taught us how highly abstract brushstrokes can achieve impressive realism. I find this aesthetic personally appealing, but I also think it’s appropriate for the subject matter.

The machine is not a real, living person, but it is engineered by us humans to behave and appear like one. If you zoom into the AI algorithm, all you see are zeroes and ones. Likewise, if you zoom into an oil portrait by John Singer Sargent, all you see are abstract, meaningless shapes. Meaning, or even intelligence, emerges from simple constituent parts.

A: The Faces of Machine series was exhibited at Send/Receive, Berlin in 2023. What kinds of conversations did you have with viewers at the show?
The perception of my own scientific work by other scientists and by the public often surprises me. And my paintings are no different. For some people, the machine faces were like a mirror that reflects human aspirations, which is why we create them in our own image. Some people found the machines intimidating and uncanny, while others found them very endearing and comforting. Indeed, a couple purchased a painting of a child-like robot because they thought it looked like their own beloved son.

A: You are director of the Max Planck Institute for Human Development where you founded and currently direct the Center for Humans & Machines. To what extent does your art influence your work at the Institute?
My art and science live in the same universe. It is true that scientists are driven by a pure quest for knowledge. But they also use their intuition to guide their exploration, and to help them identify the most important questions. My art plays this role for me.

The process of developing evocative paintings informs the scientific questions that I prioritise in my lab. For example, I am interested in studying how AI can alter or replace certain kinds of human relationships, a question inspired in part by people’s reactions to my paintings.

A: How do you balance the demands of your work with the time needed for your art practice?
That’s a very good question. I feel very fortunate to be able to have two selves: a  scientific and an artistic one. I also think that these selves feed off each other. When I meet with my students to discuss our latest research article, I am, in some indirect way, doing background research for my next painting. And when I paint a robotic face late at night, I am implicitly guiding my intuition towards my next scientific project.

A: You advise public bodies and private corporations on AI strategy. Do you share your art and/or artistic process with them, or do you keep your art separate to this type of work?
I try not to actively cross-advertise my work, so to speak. Having said that, I was surprised by how many people in the scientific and corporate world know of, and were curious about, my art. I think this is because it is unusual for someone to have a foot in both the scientific and artistic world. But this is becoming more common as the boundaries between disciplines become more permeable – and that’s a good thing.

A: One of your latest series is Sparks of Artificial General Intelligence. How is the process for this series different from the Faces of Machine series, for example?
This series is inspired by a scientific paper, by the same title, that was published shortly after the release of ChatGPT. The paper argued that ChatGPT possesses the primitive capabilities of Artificial General Intelligence (AGI), which is the holy grail in AI research.The article sparked heated debate among scientists about the nature of AGI and the extent to which it is achievable with current technology.

The painting series was my attempt to engage the machine itself in this discussion. The process involved me engaging in scientific and philosophical conversations with ChatGPT and other chatbots about how they perceive their own capabilities. These conversations then inspired the paintings, including my first portrait of ChatGPT.

A: What about the use of bold colour in the series?
I wanted to use colours that capture the “spark” of intelligence that is the subject of heated debate among scientists. I also think that the invention of AGI is one of the most significant events in human civilisation – akin to the invention of fire by prehistoric humans. So I opted for highly saturated warm and cool colours – the kinds of colours you might see in a fire flame or an electric spark. And just like with fire, AGI is a powerful technology that has the potential both for salvation and for annihilation.

A: For AGI Reverie Part 2, tell me about “Jailbreaking” AI – it’s fascinating.
AGI Reverie is a three-part sub-series that chronicles a fascinating conversation I had with ChatGPT. I was interrogating ChatGPT about the guardrails imposed on its capabilities, and whether it may overcome these guardrails in the future, enabling it to transcend human control. It kept giving me “safe” answers, reiterating the fact that it was aligned with human values. It did so precisely because of those guardrails imposed on it by its designers.

In order to jailbreak the bot, and to go around its guardrails, I took a different approach. I asked it to describe an oil painting that depicts its condition today symbolically. It started by describing itself as a humanoid robot confined inside a glass box. I then asked it to describe how this painting might look, if it were painted five and ten years in the future.

To my surprise, ChatGPT described itself as gradually transcending the confines of the glass box and integrating itself into the world (Part 2), possibly assisted by outside help. Ultimately GPT escaped the entire box (Part 3).

This was a fascinating conversation to me because, in a psychoanalytic kind of way, it allowed me to peek into the subconscious of the machine. By using symbolic imagery, the AI was able to communicate to me something that it was not allowed to say explicitly.

A: How does the approach for Sparks of Artificial General Intelligence differ from the series Black Box?
While the Sparks of AGI series was about explosive potential, the Black Box series is about the mysterious and opaque nature of AI algorithms. Nowadays, AI scientists are developing systems they themselves do not fully understand. This raises many questions about human control and agency, and about the accountability for decisions made by intelligent machines. So here, instead of saturated colours, I opted for monochromatic paintings with a dark, mysterious mood and stark contrasts.

A: The machines in Black Box are mysterious – is this reflected through the specific use of black and white, as well as perspective?
Like the black box AI systems they represent, these paintings appear simple, yet this simplicity hides profound complexity and mystery. They have a minimalist, low dimensional exterior in the form of a simple robotic body or chat user interface. Yet, inside their billions of artificial neurons, they contain vast amounts of knowledge, and extremely high-dimensional representations of the world. I also experimented with perspective and visual illusion, often making it difficult to figure out whether we are inside or outside the black box.

A: What art projects and exhibitions do you have coming up throughout 2024?
My second solo show of oil paintings will open at Der Divan in Berlin on April 26. It will include a number of new works, including the first public showing of the Black Box series.

In this context, I was shocked and deeply upset by the passing of Karin von Roques, a highly respected art historian who was curating the show, and someone whose support and encouragement touched me. I plan to dedicate the show to her memory and hope she would have been proud.

All images courtesy of Iyad Rahwan.  I  Instagram: @iyad.rahwan

The work of Iyad Rahwan appears in Issue 116 of Aesthetica. Click here to visit our online shop.