Steps Toward the Permanent Colonization of Cyberspace

Jeffrey Bradshaw

Cyberspace is a lonely, dangerous, and relatively impoverished place for today's agents. Though promoted as collaborative, agents do not easily sustain rich long-term peer-to-peer relationships, let alone any sort of community involvement. While their features for secure reliable interaction are often touted, there is no social safety net to help agents out when they get stuck, or worse yet to prevent them from setting the network on fire when they go off the deep end. Despite the fact that agent designers want them to communicate at an "almost human" level, agents are cut off from most of the world in which humans operate. Though capable of self-directed mobility, they are hobbled by severe practical restrictions on when and where they can go. Ostensibly endowed with autonomy, an agent's very existence can be terminated unceremoniously by the first passerby who happens to find the power switch. In short, the kinds of agents that we want-full-fledged citizens of the wired world, equipped with their own stamped passports and Berlitz traveler's guides explaining foreign phrases and places that allow them to hail, meet, and greet comrades of any sort in an open networked landscape and, if not able to team up on a project, at least able to ask intelligibly for directions-these kinds of agents, alas, exist today only in our imaginations (and, of course, in the vision sections of our research proposals).

The DARPA control of agent-based systems program, in cooperation with more general efforts being coordinated by NSF, DOE, and NASA, is developing new kinds of network and middleware infrastructure. Designed from the ground up to exploit next-generation Internet capabilities, grid-based approaches aim to provide a universal source of dynamically pluggable, pervasive, and dependable computing power, while guaranteeing levels of security and quality of service that will make new classes of applications possible. On top of DARPA's agent grid, various additional Java-platform features, agent conversation and management policies, and "social institutions" are being developed to support long-lived agent communities.

In this talk, I will describe some of the requirements driving the design and early implementations of the agent grid, and by discussion and demonstration show how its future is already taking shape.