Software Agents

Software agents are computer programs that perform customized, automated tasks, such as searching the Internet for the lowest price for a particular product, prioritizing incoming email, or coordinating production schedules.

As the term implies, software agents allow users to delegate some of the labor of navigating the surging sea of information that flows through computer networks. They function as what computer scientist Alan Kay describes as “soft” robots, carrying out the bidding of their clients in cyberspace, and thereby taking on the work of information filtering and prioritizing. If the Internet promises to lower transaction costs by making prodigious amounts of information readily available, software agents contribute by reducing the time and energy spent in sifting through this information.

Given the scope of activities envisioned for software agents, the term is a flexible one, characterized, as writer Neil Fawcett notes, by the attributes of [Page 414]customization, autonomy, and the ability to “learn” and adapt to new situations. Other definitions add the attributes of longevity (that such agents endure long enough to “remember” their clients' preferences and directives), communicability (that they are able to communicate with other agents), and proactivity (that they anticipate client needs). For example, a basic online search engine exhibits some of the attributes of a software agent, but not all. Search engines allow users to delegate the task of searching the Internet based on customized preferences, whereas software agents incorporate the additional ability to learn from, and eventually anticipate, user preferences.

Learning is critical to the performance of software agents, and therefore their development is closely linked to research in artificial intelligence (AI). Fawcett traces their genesis to the notion of “roving software,” developed by the AI community at the Massachusetts Institute of Technology (MIT). Kay credits MIT's John McCarthy with coming up with the idea of software agents in the mid-1950s, and Oliver G. Selfridge with coining the term a few years later. The goal was to develop software that could not only follow instructions, but also seek assistance and advice along the way.

If AI researchers envisioned the software agent and its capabilities, the development of the Internet and the World Wide Web helped to create its habitat. A networked environment provides roaming room for software agents, allowing them to interact with each other and with computer users. MIT Media Laboratory founder Nicholas Negroponte suggests, for example, that software agents—perhaps with voice-recognition capability and programmed “personalities” of their own—will increasingly mediate the interface between humans and the network, allowing clients to spend less time online.

As is the case with new technology in general, many of the current and developing applications of software agents are commercial. For example, the MIT Media Laboratory's Software Agents Group is exploring the use of software agents to facilitate realestate purchases and airline bookings. Emerging corporate applications include the use of networked agents to decentralize and streamline industrial production. Julian Perkin, for example, describes the use of networked software agents by DaimlerChrysler to increase flexibility in the automated production of automobile engines. A system of specialized agents responsible, in turn, for monitoring the progress of a specific part, plotting its path through the production process, and updating the availability of individual machines all work together to make the manufacturing process as efficient as possible. The results of this decentralized, collaborative approach, according to Perkin, are often superior to those of top-down automated systems.

...