Biopoiesis is a series of experiments exploring the relationships between structure, matter, and self-organization, with a goal toward the development of autonomous computation and control systems in a computational “primordial soup”. It is based on the work of cyberneticist Gordon Pask and his research into electrochemical control systems that had the ability to adaptively construct their own sensors. This series of experiments demonstrates the computational possibilities of natural processes that might serve as an alternative to the more commonplace digital forms of computation.

The Biopoiesis project features the construction of many simple computational devices that each harness an electrochemical reaction. Information (an electrical signal) is passed through electrodes to a tank filled with a metallic salt solution (e.g., stannous chloride). The resultant electrochemical reaction grows into dendritic metallic threads – ultimately leading to the formation of a continuously shifting signal network (not unlike a neuronal cell assembly). These dendrites are fluid and unstable: they bifurcate and dissipate in unpredictable ways, leading to resistance changes that modify the flow of information through the network.

If a subset of electrodes in the electrochemical solution receives input from an environmental sensor, and the electrochemical output can affect that sensor, then the network will move towards a dynamic equilibrium with its environment. Moreover, the dendritic network carries a decremental memory trace of its previous activities: when the environment changes, the system is perturbed but not immediately reset. Thus, the prior activity and configuration of the system affects how it handles a change in its environment. Furthermore, by “rewarding” conductance changes associated with a particular kind of environmental “perturbation” (e.g. changes in sound, light, humidity, etc) it may be possible to steer or “train” the network to associate particular inputs with particular outputs. The “reward” in this case entails feeding the system more current.

The computational approach taken in this research can be employed in countless ways and with many different media. For example, dendritic growth and dissipation patterns might be applied to the transformation of vertex data in computer graphics or gesture recognition in video tracking. The re-patchable nature of the project also allows one to explore the computational possibilities of natural processes that might serve as an alternative to the more commonplace digital forms of computation, while the malleable nature of the medium allows for the exploration of virtually unbounded search spaces and implementation of opened-ended evolutionary designs.

More info: biopoiesis.dprime.org

GitHub