30-10-2009, 04:22 PM
Swarm Intelligence Swarm intelligence has captured considerable attention (e.g. Bonabeau, Kennedy). With swarm intelligence, small units or agents no smarter than an ant, can solve network problems, such as the traveling salesman problem. The problem of finding needed information in the world wide web is substantially more complicated than the problem of finding the shortest path through a network. Yet it has similarities to the problem that real ants face when they seek food. While some ants navigate by reference to the sun (Gallistel; Bisch-Knaden), the familiar domestic ant navigates on a swarm basis by leaving pheromonic tracks, which indicate that he is returning to the nest with food. Other ants who find those tracks can follow them in the direction of diminishing strength to find the food themselves. People who search the web face a problem much more complicated than that faced by the ants. Each individual search or quest has its own context and goals, which include the prior knowledge of the searcher, limitations of time, and details of the problem to be solved. (Kantor et al. 1999, 2000) Fortunately, the humans who search the web are also much more intelligent than the ants in the kitchen. They make complex and subtle judgments about the validity and usefulness of the items that they find in relation to their quest. Thus, if the nodes of the web could have, attached to their outward links, pheromonic markers indicating which quests have followed those links with benefit, the web itself can become a distributed, decentralized resource to aid in searching. With such a self aware web, the searcher need only stumble across a node that has been previously traversed by someone with a similar quest in order to begin following a useful path. This approach stands in sharp contrast to the centralized approach typified by Google and other major search engines. The idea of digital pheromones and of quest matching was put forward in earlier work by Kantor and others (Kantor, 2000). They built a centralized instantiation, an unscalable bottleneck that the proposed research intends to remove. The goals of this research are to develop a distributed extension of the earlier AntWorld, called AntWeb. It will include digital pheromonic encoding schemes for quests with their context, distributed means for storing and matching quests, means for marking outgoing links to aid in quests, and most significantly means for relaying judgements back to the node in question so that the links can be marked appropriately. The vision of AntWeb is that the Web will be increasingly populated by "AntAware" (AA) sites which each maintain a database of links that have been followed from that site. With each page link is associated a set of pairs consisting of a representation of the quest (a pheromone) and a judgment, on that quest, of the page to which the page link points. Subsequent visitors to the site, if they make their quest pheromones available to the site, will receive "enhanced" pages. Enhancement is achieved in a two step process. The local AA site finds, in its base, pheromones most closely matching the present one. It combines the judgments associated with those pheromones, and marks its page in a way that is specific to the present quest. This marking may be done by highlighting, or by appropriate icons. This increases the value to the present visitor. This visitor, in turn, it is hoped, will provide further judgments of the pages, which will add to the completeness and precision of the database at the AA site. We propose to advance this vision by building two prototype systems, and we will use them to research a set of interrelated scientific issues. That research will be the principle scientific product of the proposed project. [attachment=381]
