Working Like an Ant
|
Working Like an Ant |
 |
AI systems inspired by studies of the natural systems of social insects.
Emergence - the phenomenon embodied in the fact fact that complex patterns and behaviors can result from simple rules and algorithms - recently became one of the key concepts in the field of AI. Numerous research groups are working on various "bottom-up" systems, demonstrating how objects and patterns can arise from simple interactions in ways that are often surprising and counterintuitive. Steven Johnson, cofounder and editor-in-chief of the FEED magazine, recently published a book that explores this concept in more depth.
"Emergence: The Connected Lives of Ants, Brains, Cities and Software" compares the collective intelligence exhibited by ant colonies with other bottom-up systems, including computer games, immune systems and modern cities. Weighting the impact of Web sites like Napster, eBay and Slashdot, Johnson predicts the creation of a new media world in which self-organizing clusters of shared interests structure the entertainment industry.
Emergent phenomena are present in many AI-related disciplines. Some of the most fascinating examples are inspired by studies of the natural systems of social insects such as ants. While the behavior of individual members is unpredictable, the insect colony as a whole is self-organized and displays order. This approach is used in robotics (for creating structured robotic communities) and Artificial Life (Chris Langton's Vants), to name just a few. Some conferences and Web sites are entirely devoted to ant algorithms and ant colony optimization.
Transcom Software, Inc. now starts using this "antified" approach to solve the biggest problem of the networked community today - how to get maximum use of the huge and ever expanding resource of data and information. Their Colony is a self-learning solution for collaboration between members of a work group, between groups, departments, business units, customers and suppliers. It overlays an organization's applications and work patterns with a system of autonomous, self-organizing intelligent software agents based on Transcom's Automated Intelligent Decision Making (AIDM) technology. Based on its context, a document is automatically downloaded, analyzed and categorized by Transcom's Contextraction technology which groups similar documents into categories and provides a brief context-based description. Expertise can be managed in several ways: expert declaration (whereby a user fills out a personal profile indicating their expertise), inferential (whereby the system makes assumptions about user expertise via database footprints), and analytical (whereby the system monitors user contribution to certain categories). The solutions found within Colony can be customized for enterprise, vertical, and meta portals. PlanetColony.com is a good example of a typical meta search engine based on Transcom's solutions. As for the technical requirements, Colony supports any ODBC compliant database and is accessed by a standard Web browser. It is implemented through ISAPI (Internet Server Application Programming Interface) on top of Microsoft's IIS (Internet Information Server) Web server. A white paper titled "Colony 3.0: An Environment for managing with intelligence and knowledge" offers more background details on Transcom's philosophy.
Although this is a quite original approach, it is not the first one in its category. Back in 1998 I described programs modeled after ant chemical systems for communicating information. Paul Kantor, the Rutgers University professor of information sciences, noticed that like ants looking for food, each day millions of people search for information on the Internet. However, unlike the ants, which leave a chemical trail to lead the rest of their colony to lunch, human searchers are on their own, even though others may have already looked for the same information. Each person's decisions made in the course of getting to the right information are essentially lost to the next researcher trying to find the same information. AntWorld, the result of this research, is a software system designed to allow users better manage their Web searching, and to share their findings with other people. The main concept of AntWorld is that of Quest: an information-searching process which is determined by its goals, itinerary (the list of pages you visited) and the user's judgments (your opinions about usefulness of the visited pages to your quest goals). AntWorld tries to match the goals and findings of your current quest and those of other AntWorld quests - both run by you and by other AntWorld users. It creates a suggestion list for you, consisting of the pages that were judged as useful in your current quest and other quests similar to it.
Another interesting resource for exploring emergent phenomena, along with fractals, chaos, complex systems and adaptation, is Gary Flake's The Computational Beauty of Nature. Its basic premise is that the most interesting computational topics today are deeply interrelated. It can serve as a great starting point for serious research in this area, and offers a great fun at the same time.