^
2
^
Ken Goldberg 3/24/2016
Permalink|Reply
Private. Collaborators only.
Selection made on Version 7
A beautiful example of an Entanglement process is the use of simulated biologically-inspired algorithms to design artificial objects through evolution and morphogenesis. Multiple designs are mutated, bred and selected over many generations in a process analogous to Darwinian selection. The artifacts created by such processes look very different from those produced by engineering [10]. An evolved motorcycle chassis will look more like a pelvic bone than a bicycle frame[11]. A computer program produced by a process of evolutionary design may be as difficult to understand as a neural circuit in the brain. Thus, the artifacts that are designed by these biologically-inspired processes take on both the beneficial and the problematic characteristics of biological organisms [12]. Their beauty is in their functional adaption. This is the elegance of the Entanglement: a new expression of beauty emerging from process. In an Entangled design process, the humans will often have input without control; for example, they may influence aesthetic choices by participating in the selection process or by tuning parameters. Such processes lend themselves to collaboration among multiple machines and multiple humans because the interfaces between the parts are fluid and adaptive. The final product is very much a collaborative effort of humans and machines, often with a superior result. It may exhibit behaviors that are surprising to the humans. Some of these behaviors may be adaptive. For example, early walking machines evolved on the Connection Machine took advantage of an obscure round-off error in the floating-point unit that the human programmers did not even know existed[13]. In this sense, artifacts created by the entangled processes may have some of the robustness of a biological organism, as well as some of the surprise and delight.
Such processes lend themselves to collaboration among multiple machines and multiple humans because the interfaces between the parts are fluid and adaptive. The final product is very much a collaborative effort of humans and machines, often with a superior result.
What is needed is a science of collaboration that combines diverse sets of machines (eg ensemble theory, random forests), with diverse sets of humans. In contrast with the rhetoric of Singularity, one might call this Multiplicity.