Software agents are artificial secretaries which are the electronic
counterpart of their real-world namesakes. This is not to say that
anyone looks forward to the prompt delivery of simulated coffee each
morning! Rather, these personal assistants are designed for tasks
such as: filtering e-mail, scanning NetNews, providing appointment
reminders, etc.
Given the complexity of on-line environments and
huge volume of information flowing across the Internet, this type of
agent looks quite attractive.
Interface agents ([Sheth, 1994][Maes, 1994]) are a special class of on-line assistants which are designed to simply user-interaction with particular pre-existing applications. These agents are designed to learn and predict users' behaviors and preferences. SodaBots have little in common with interface agents, because interface agents are each highly elaborate, custom-crafted programs designed for very specific applications. Also, SodaBot does not have built-in capabilities for learning user behavior and preferences. Providing these would be an interesting direction for future development of the SodaBot system.
For example, Sheth provides an interactive information retrieval system for UseNet NetNews articles. It is designed to autonomously select those articles whose content would interest the user according to some complex metric involving the user's previous selections. Sheth's work is similar in spirit to much of the activity regarding knowbots ([Waldrop, 1990][Kahn and Cerf, 1988]), which has generally fallen outside of the realm of mainstream AI research, although there are some exceptions, e.g. [Knoblock and Arens, 1994]. Knowbots (knowledge robots) are intended roughly as librarians for enormous digital data-libraries. They are not general purpose but are highly elaborate, specific creations. Thus, these too have little if anything in common with SodaBots.