- Towards Epistemically Autonomous Robots:Exploiting the Potential of Physical Systems
The authors outline one path towards constructing interactive artworks with the potential for displaying novel behavior. They use Peter Cariani's taxonomy of adaptive robotic systems as a framework for comparing the capabilities of systems that interact with their environments. The authors then describe two examples of structurally autonomous systems that are able to construct their own sensors independently of a human designer. The first device, the evolved radio, is the result of a recent hardware evolution (HE) experiment conducted by the authors. The second device, the electrochemical ear, was constructed almost 50 years ago by the British cybernetician Gordon Pask. The emergent behavior in both systems is only possible because many conventional engineering constraints were relaxed during their construction. Using existing technology, artists have the opportunity to explore the potential of structurally autonomous systems as interactive artworks.
The sculptor and theorist Ken Rinaldo has surveyed the use of artificial life (ALife) techniques in the arts and argues that the greatest potential for their application lies in developing interactive artworks that go "beyond the hackneyed replicable paths of interactivity that have thus far been presented by the arts community" . Rinaldo's vision is of a "cybernetic ballet of experience, with the machine and human involved in a grand dance of each sensing and responding to the other," which will result in "sculptural and virtual algorithmic manifestations that will far surpass our wildest imaginations." Motivating this vision is what Mitchell Whitelaw has identified as an essential driving force in contemporary ALife art: the "will to escape, the desire for emergence" . A shared interest in emergent phenomena provides common ground between scientists and artists in the ALife community. However, "emergence" is a notoriously slippery concept, with many subtle nuances and definitions. Consequently, the term is used in different ways by ALife practitioners, which can potentially lead to misunderstandings. Although fully aware of these difficulties, we focus here on the following open question: What sort of interactive mechanisms could display emergent behavior that does surpass our wildest imaginations? This paper is primarily written from a scientific perspective, and our aim is not to give a definitive answer, but rather to contribute to an ongoing debate in ALife. Given the scope of this paper, we concentrate on the state of the art in autonomous robotics and can only briefly consider the aesthetic implications of applying this technology in an artistic context.
Initially we outline Cariani's taxonomy of adaptive robotic systems , which provides a useful theoretical framework for comparing the capabilities of machines that interact with their environments. This taxonomy has been used by both scientists and artists  and provides one principled approach to exploring emergent phenomena. The most powerful generative mechanisms in this hierarchy are what Cariani calls epistemically autonomous devices: systems that construct their own sensors and/or effectors and thereby determine their own relations with, and knowledge of, the world. In order to make this theoretical concept more tangible, we describe two concrete examples of such systems. We highlight the key properties that these devices share and that lead to their epistemic autonomy. The first device, the evolved radio, is one of the results of a recent hardware evolution (HE) experiment. The second device, the electrochemical ear, was constructed almost 50 years ago by the British cybernetician Gordon Pask.
Cariani's Taxonomy of Adaptive Robotic Systems
One way to describe the organization of both organisms and robots is in terms of three abstract parts: sensors, effectors and a control mechanism that coordinates the two. A key distinction here between a robot and other devices with sensors is that a robot has effectors that enable it to move. Peter Cariani has developed a taxonomy of adaptive robotic systems using this framework, which enables different organisms and robots to be compared in terms of the capabilities of these three basic parts. Each of them can be more or less flexible, thereby determining how adaptive the device is. The sensors constrain the perceptual categories of the device, and the effectors constrain the ways in which the...