CHAPTER III — ORGANISM
Alexander Mordvintsev
Can computation become conscious?
Conversation II of III
An organism is more than an intricate arrangement of parts. It persists, adapts, repairs itself and responds to the world around it. The question is no longer whether complexity can emerge from simple interactions, but when those interactions begin to resemble life itself.
Alexander Mordvintsev's practice explores precisely this threshold. From his early work on DeepDream to recent research into Neural Cellular Automata and Artificial Life, he investigates how local interactions can produce systems that grow, regenerate and learn without central control. Rather than designing finished forms, his work establishes the conditions from which living behaviours can emerge.
In this conversation, Mordvintsev reflects on why emergence fascinates him, how Neural Cellular Automata extend Conway's original Game of Life, why resilience may distinguish life from computation, and whether consciousness might eventually arise from sufficiently adaptive systems.
CHAPTER III — ORGANISM — CONVERSATION II OF III
"I think AIs will 'become' conscious once people start recognising them as such."
Alexander Mordvintsev, interviewed by Kate Vass — Zürich, Switzerland
Kate Vass: Many people first encountered your work through DeepDream, but your recent practice has shifted toward Artificial Life, and self-organising systems. What drew you from interpreting neural networks to exploring how life itself might emerge from computation?
Alexander Mordvintsev: For me this feels more like returning to my roots. I grew up with books about cellular automata and computer graphics, and loved how much visual complexity the tiny printed programs could produce. The discovery that you could model and simulate aspects of our universe inside a computer stayed with me. DeepDream came much later, as part of my efforts to understand deep neural networks and differentiable programming, and that experience gave me a lot of useful tools to move further.
KV: Artificial Substrate presents life-like behaviour emerging from simple local interactions rather than explicit design. What fascinates you most about systems that organise themselves, and what do they reveal about the nature of life?
AM: I guess the most fascinating thing for me is the fact that we all exist as parts of such a system. This is a framing that provides a universal sense of meaning and belonging to a process which began on this (or another?) planet billions of years ago. This process of combinatorial exploration and integration of information puts everything in perspective. It establishes links between the creation of life, the emergence of spoken and written languages, religions and ideologies, book-printing, computers and internet, and eventually modern AI.
KV: EMERGENCE begins with Conway's Game of Life and follows a trajectory from simple rules to learning systems. How do you see the relationship between Conway's cellular automata and today's learning-based artificial life systems?
AM: Conway's Game of Life was a gateway into the world of complex systems and related computational curiosities for generations of researchers and artists. Funny enough, Conway wasn't particularly proud of this work and considered it a distraction. Yet the idea that the system rules must be universal in space and time, and that all interactions are local, makes this model resemble a toy version of "physics". The fact that even such a primitive system is Turing complete and supports a tremendous diversity of patterns and behaviours inspired many people. I guess the biggest drawback of the Game of Life is its fragility: flip a single pixel and a big beautiful structure falls apart. This is the problem I tried to overcome in my work on learning-based neural cellular automata. Our modern digital technology is also built around the idea of overcoming noise. Harnessing noise and putting it to use the way nature does could be the next big frontier in computing.
KV: The exhibition asks whether emergence itself can become conscious. From your perspective, what distinguishes a complex adaptive system from something we might one day recognise as genuinely alive or even conscious?
AM: I used to avoid going deep into philosophical debates around such hard-to-define and hard-to-quantify things as consciousness, preferring to stay in more practical, down-to-earth realms. Yet recently th ese practical realms started throwing these questions in my face every day. My thinking here is largely second-hand, mostly following people like Jürgen Schmidhuber, Michael Levin and Blaise Agüera y Arcas, and it centres on a system's capacity to model and predict the world around it. A sufficiently capable world model has to contain a model of itself (and of the fact that it is itself being modelled by the world) and so on, ad infinitum. From a practical perspective, I think AIs will "become" conscious once people start recognising them as such, and I think there's a real chance for me to see this in my lifetime.
KV: Throughout history, artists have imagined new forms of life long before science could explain them. As someone working between research and artistic practice, what role do you think artists play in shaping our understanding of artificial life?
AM: I've been asked about the relation between art and science quite a few times over my career, so my default tongue-in-cheek reaction is: it's art if you don't have to explain the trick; it's science if you can show that someone did something similar before. But seriously speaking, whatever I work on, there are countless decisions to make. Which results look interesting, which paper to read, which experiment to try next? Answering these questions lies in the realm of intuition and is guided by "taste", which makes it feel like an artistic practice. As Picasso once said: “I never do a painting as a work of art. All of them are researches.” A lot of work coming from the ALife community feels more like art, but I also often see artists doing things which quite certainly bear scientific significance (e.g. Memo Akten's recent thread on ant colony object transport).
— 20th August