The Reality Gap

2018 - in progress (documentation assistance by Lauren Glegg )

A population of small machines collect solar energy, some through the window and others from a fluorescent light fixture. Each individual captures just enough energy to wake up for 10-15 seconds every few minutes. Their movement evolves through a collective simulation of the Darwinian principle of natural selection. Selective pressure favours machines whose behaviour is novel relative to the rest of the population. Rather than “survival of the fittest”, adaptation in this system can be characterized as “survival of the most unique” [1]. The process unfolds very slowly, and this installation is partly an experiment to see how perceptible their evolutionary novelty search will be over the course of the month-long exhibit. Does simulated evolution happening below the surface produce increasingly unique sounds and movements in the gallery space?

This installation is part of an ongoing project titled Open Ended Ensemble. The work lies within the field of Artificial Life, which combines the disciplines of computer science, biochemistry, and art to examine social aspects of living systems and “contribute to theoretical biology by locating life-as-we-know-it within the larger picture of life-as-it-could-be” [2].

Biologically-inspired computing and machine learning have received significant attention in recent years. This is primarily due to the success of deep learning, which uses artificial neural networks to model the hierarchical sensory systems of living organisms, e.g. [3]. While effective, the approach is computationally demanding and typically requires costly specialized hardware (Graphics Processing Units). As such, research in these fields is in danger of being limited to the realm of ‘big science’, where large tech companies such as NVIDIA, Google, and Facebook have a clear advantage over publicly funded research institutions. Worse yet, deep learning incurs a staggering carbon footprint . By contrast, each machine in this exhibit uses one bare-bones, low-power, and inexpensive computer chip. Through radio communication, they implement a distributed evolutionary algorithm that adapts its complexity to the requirements of the problem. While this particular system’s only purpose is to exhibit behavioural diversity, evolutionary systems of this nature can be used to solve real-world problems at a fraction of the cost of deep learning .

[1] J. Lehman and K. O. Stanley, “Abandoning Objectives: Evolution Through the Search for Novelty Alone,” Evolutionary Computation, vol. 19, no. 2, pp. 189–223, Jun. 2011.
[2] C. G. Langton, Artificial Life: An Overview. Cambridge, MA, USA: MIT Press, 1995.
[3] V. Mnih et al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, Feb. 2015.