Held over five days in early April, each day of the Workshop consisted of a keynote presentation followed by an open-ended discussion whose goal was to uncover common threads across interdisciplinary bounds. From generative systems in virtual worlds, and open-ended technology and society, to evolutionary innovations and artiﬁcial life, and roadmapping the future the workshop introduced new perspectives that may be useful in identifying areas of common interest across domains – partiularly those in AI and cognitive sciences.
Here, we present three sessions in full. First, Dr. Sebastian Risi discussed self-assembling designs and their potential implications for AI. Second, Dr. Lana Sinapayen spoke of failure states as information-dense sources of insight within complex systems, and third, Dr. Tim Taylor discussed innovations in real and virtual worlds, and his approach to open-ended evolution.
In sharing these reﬂections we hope to identify opportunities for future research and to open the door for new collaborations on mutual topics of interest. A detailed summary of events and open-questions can be found here.
A very big thank you to everyone who attended, and our speakers for their time and allowing us to make their presentations public.
Dr. Sebastian joined us to talk about self-assembling designs. In biology, a single cell has all the instructions to grow into a full-fledged animal. What if AI-based methods could do the same? In this discussion, Sebastian discusses recent work in Neural Cellular Automata: simulated systems where many cells learn to cooperate to form complex structures, and neural CAs that have shown promising results in growing images, self-repairing soft robots, and regenerating 3D Minecraft structures.
Dr. Lana Sinapayen introduces us to the idea of "Failure as a Research Tool". In understanding complex systems, failure states can be an information-dense source of insight. What happens if a module is removed? What causes this behavior to fail? Even in successful models, failure patterns tell us a lot about how systems represent themselves. Understanding why things fail is a key step in recreating interesting systems in artificial ways.
Dr. Taylor has been researching real and virtual worlds since the mid-90s. While earning his PhD in Tierra-like systems, Dr. Taylor noticed these systems are marked by bursts of initial evolutionary activity that settle to a stable state over time. In this presentation, Dr. Taylor revisits his approach from an open-ended evolution perspective, covering different types of open-endedness, and how we might achieve open-endedness through novelty and innovation.