Agents are bits of software that help computer users by performing routine tasks, typically in the background, while the users continue to work on other matters. Information gathering, filtering, and presentation are some well-defined tasks prescribed to agents . Traditional software such as word processors, spreadsheets, and calculators only respond to human input in a fixed and predictable manner. Intelligent agents are capable of "thinking" and producing intelligent feedback.
SAFER (Zhu et al., 2000) is an infrastructure designed to serve agents in e-/m-commerce and it also aims to establish necessary mechanisms to manipulate them. It focuses on three fundamental activities of agents, namely, fabrication, evolution, and roaming.
Agents are self-contained software entities that could travel over a network (in this case, the Internet) to do specified tasks (Yang & Guan, 1999). SAFER agents are envisaged to act on the behalf of customers to carry on e-commerce/ m-commerce activities in a simple and independent manner. Common tasks that could be entrusted to agents include remote product search, negotiation over interested items, payment for purchased item, etc.
The agent community (see Figure 1) is the basic unit in SAFER. A network client can join a SAFER community by applying to a local community administration center (CAC). The CAC will issue a digital certificate to the applicant if it accepts the application. This certificate can be used to identify the client's agents by trusted remote hosts that the agents roam to. Under these organized communities, agents are fabricated by the agent factory per member's request. After localization, they can be controlled by the individual owner through a coordinating entity called agent butler . The security of the agents can also be enhanced under a tightly coupled structure.
A key element in any e-commerce system is the method of payment. However, existing monetary and fund-transfer arrangements are difficult to be transplanted directly into the e-commerce marketplace . The global and crossborder reaches of the Internet further complicate the considerations for any online payment system.
Currently, a common e-payment method involves the client transmitting to the merchant details of a payment card such as a VISA credit card. The merchant receives the information and proceeds to carry out a payment request with the card issuer via traditional payment card procedures. This system is simple and does not require the development of a new commercial infrastructure. However, the system is highly susceptible to frauds from either transacting party. The card information transmitted over the Internet could also be stolen by malicious parties. Many electronic payment schemes have been proposed to offer solutions to these problems.
Digital cash (CAF ƒ °) (Mjolsnes & Michelsen, 1997) uses electronic tokens (a unique coded string) to represent monetary value. The issuing financial institute of the tokens has a record of all the tokens. Merchants that receive the tokens will transfer them to a clearinghouse to process them. When the tokens are verified by the issuing financial institute, the actual transfer of funds takes place and the token cannot be used anymore. Digital cash can be used anonymously. Transactions can also be carried out offline and the token cleared at a later time. However, the need to police potentially infinite amounts of digital cash tokens means the accounting and clearing resources would have to be enormous .
The secure electronic transaction (SET) protocol is an evolution of the existing credit-card based payment system. It provides enhanced security for information transfer as well as authentication of transaction participant identities by registration and certification. SET is intuitive as it is based on the current payment model. It is supported by major corporations such as VISA, Inc. and MasterCard. SET is also an international standard with published protocol specifications.
It has also been proposed to utilize smart cards for e-payments (Poh & Guan, 2000). These could include cash cards. A special card-reader would read the prepaid value in the card and carry out deduction with corresponding transfer of payment over the Internet. The scheme utilizes existing infrastructure but has additional hardware requirements (card reader). The other major disadvantage is that it may not be compatible globally.
According to the nomenclature of Maes in the MIT Media Lab (Guttman & Maes, 1998), the consumer buying behaviour model (see Figure 2) is separated into six stages, namely:
Payment and delivery
Product service and evaluation
The definition helps to categorize existing agent-mediated e-/m-commerce systems. These stages may overlap and migrate from one step to another in a nonlinear and iterative way. Though it is only a simplification of the more complex behaviours, it provides a solution to identify the roles of agents as mediator in e-/m-commerce. However, currently, there is no automated system today with all these stages. Some pilot research projects assist various stages of the buying process.
For example, an agent marketplace system called Kashbah (Chavez & Maes, 1996) was implemented by the MIT Media Lab using multiple agents that are intended to bring about fundamental changes in the way buying and selling is conducted and doing much of the work on the user 's behalf. Buyers who need to procure particular goods would create an agent, give it basic strategic direction, and send it off into the electronic marketplace. The Kashbah agents would then proactively seek out potential sellers and negotiate with them on the buyer's behalf, making the best possible deal based on a set of constraints specified by the buyer, including a highest acceptable price and a transaction completion date. However, it is clear that it only covers specific aspects of the buying process, e.g., from stage two to four as listed above. They do not support the payment stage in their systems. Another research project called BABSy (Rockinger & Baumeister, 2000), proposed by M ƒ ¼nchen University, Germany, is also based on the conceptual structure shown in Figure 2. An accounting agent is mentioned in BABSy, which is considered to represent the bank within the system. This agent keeps track of payments made towards merchants. It will also notify merchants when a transaction in their favor has been booked. The design of this system is supposed to bring together all the main steps of the buying behaviour, including purchase itself. However the functionality of the detailed payment module is not very clear. Moreover, the role of the user agent is quite simple so that the security concern is seriously brought to the surface.
To fully automate the whole buying behaviour module, it is clear that an agent-based payment system should be defined with clear function modules and high security guarantee. At the same time, the role of different types of agents should be clearly presented as well.
We propose a modularized electronic payment system for agent-based e-/m-commerce, especially for the SAFER architecture. It combines the agent technology with current payment schemes, which will be described in the following sections. The SAFER payment system does not limit itself to a fixed method for electronic payment. The payment functionality of agents or the agent butler is extensible and will be able to handle different forms of payment such as payment card or digital cash, etc. For the current system implementation, SET was chosen as the default payment scheme.