An Architecture for Supporting Quasi-Agent Entities in the WWW

Charles Brooks, W. Scott Meeks, Murray Mazer, OSF Research Institute

One-line summary: Circumvent Web usage limitations by providing HTTP stream transducers for annotation, semi-autonomous WWW-based (really HTTP-based) "browsing associates", and front-ends for encapsulating non-HTTP content (mail, news, Zephyr-grams, etc).

Overview/Main Points

• Quasi-agent goals: autonomy, temporal continuity (runs for a long time rather than doing one transaction and quitting), ability to communicate with user and other entities.
• Long-lived stateful stream transducers: browsing assistance that lives in the HTTP path. Can be used to filter content, characterize (summary digest) content, index surfed pages, "document correlation", etc.
• Transducers can be "stacked" via thin interface (HTTP proxy) to compose functionality.
• Browsing associates: silently watch your browsing and do some of their own in the background. Examples: modify HTML when it detects that a previously-visited link has been updated; build up graphical maps as user surfs that can be clicked to control Netscape remotely; etc. These agents have their own separate UI based on Tcl/Tk.
• Encapsulation/annotation: web mail, news, etc. Group servers capture and compose annotation info, metainformation, etc. from various people and dynamically generate annotated document in HTML.
• Investigating using Web protocols for interagent communication, as well as having agents be conversant in various browser API's, Zephyr notification, Tcl/Tk send function, etc.

Relevance

Stateful proxy-like entities that help you out during Web browsing. It's Web-centric in terms of implementation but the architecture seems more general, even though authors don't directly make that claim.

Flaws

• Authors would like to generalize the "transducer" notion, but seems like they are running into the obstacles of stateless HTTP, HTML as the only medium, etc.
• Plans to build a "shared bulletin board" (basically a newsgroup) via HTTP gateway, or putting a Zephyr client front-end on it; how is this different from putting that on top of NNTP news?
• In general, this and other work sometimes blurs the difference between a useful hack that provides new functionality per se (like browsing associates, transducers) and semi-useful hacks that just reinvent the wheel using Web protocols and clients (web mail, bulletin board).