Biologically Inspired Computing

Josh Bongard, University of Vermont
Evolutionary algorithms and robotics hold great promise as integrated design and modeling tools.

A column on intelligent systems and AI research appears passé, when there are exclusive journals and other outlets devoted to their study. But there is an opportunity to present ideas and developments in a way that appeals to experts and nonpractitioners alike. This column is intended to fulfill that need by serving as a “what’s happening in the AI disciplines” commentary. Every other month, it will feature articles written by prominent AI researchers on the latest work in their areas. We cast a broad definition of AI that embraces classical topics, interfaces with other disciplines, and applications in novel domains.

Despite the relentless, breathtaking advances in comput ing and related technologies, we continue to be humbled by the variety, adaptability, and sophistication of the natural world around us. From the beginning, computing has been inspired by nature: Alan Turing asked whether computers could think like us, while John von Neumann, armed only with pencil and paper, sketched out an automaton that could self-replicate.

Since then, a divide has grown between computational scientists on whether to continue creating faster, more efficient algorithms and hardware that exhibit centralized control or to place less emphasis on speed and efficiency than on robustness, adaptability, and emergent organization from the interaction of many loosely coupled processes. These latter approaches have come to be known as biologically inspired computing, which is not so much a field as a philosophy that links various disciplines such as artificial intelligence, evolutionary computation, biorobotics, artificial life, and agent-based systems.

One argument against bio-inspired computing is that large-scale systems cannot emerge from blind, bottom-up processes. However, one study found that Wikipedia, with little centralized editorial control, is nearly as accurate as Encyclopedia Britannica (G. Giles, “Internet Encyclopedias Go Head to
Head,” Nature, 15 Dec. 2005, pp. 900- 901). Social-networking platforms, peer-to-peer networks, and usergenerated content sites—although not strictly bio-inspired—have likewise demonstrated that such systems can grow, organize, and improve themselves with little direction from above.

Bio-inspired computing has passed in and out of vogue during the past few decades, mostly due to the claim that it does not embody true computer science in the sense of delivering guaranteed performance in clearly defined domains. But bio-inspired algorithms can exhibit strength through flexibility, or strength in numbers: They often work well even when the desired task is poorly defined, adapt to unforeseen changes in the task environment, or achieve global behavior through interaction among many, simply programmed agents.

Read full text