SYMPOSIUM MINISYMPOSIA

ABSTRACT

Intelligent interfaces hold the promise of freeing computer users from routine drudgery, from the mental maintenance of staggering amounts of arbitrary information. On a typical Unix workstation, such as one in our department, there are approximately 1700 different programs one can run; there are over 250,000 files and directories on our small network; our 22 public workstations are fairly homogeneous, but most of our 70 private machines have either local file systems, special devices attached, or both. When one considers what is accessible over the Internet, these numbers increase astronomically. Consequently, text-based computer interfaces are almost unavoidable because they allow the user to express _in_language_ concepts difficult to express by direct manipulation of icons and menus.

An intelligent interface supports automation of tasks, but automation in this domain is challenging. Our goal is to support the automation of tasks, whether directly by the user (who can incorporate domain knowledge) or through an automated system such as a learning agent. Unfortunately, current shells cannot be adapted to allow a greater level of automation for a number of reasons:

- Their behavior can only be changed in a few predefined ways.

- Their languages lack the full functionality of traditional programming languages.

- The ability to launch and interact with autonomous processes or agents is limited.

- The incorporation and organization of knowledge in shell files is problematic. This is further exasperated when multiple shells run concurrently.

In order to enable automation of user tasks, we find a new command shell architecture is needed. Our approach is to centralize knowledge in a persistent way while parallelizing execution. The working prototype, FISH, is extensible and customizable, with a full programming language. It supports efficient communication between the machine and the user in many ways, most notably by giving the user access to the results of previously executed commands. Indexical references into previous results are possible by using a sophisticated command language which contains a suite of syntactic and semantic analysis functions. Significant strides in learning are possible due to the centralization of knowledge and its persistence.

Of the technologies and techniques that this system enables, the following four are of great importance:

- Maintenance of Persistent, Organized Knowledge: For example, the user defines a new shell function. It is automatically accessible to all other concurrently running sessions and to future sessions.

- Global Learning: E.g., a learning agent watches the commands the user enters in multiple concurrent sessions. Upon detecting a repetitive sequence of commands, the agent automatically writes a function to automate that command sequence.

- Agent Interaction: The aforementioned agent not only writes the function but also enters it directly into the knowledge repository so that the user has immediate use of it.

- Typed, Indexical Reference to Previous Results: For example, arguments such as ``that'' and ``those'' could be given to functions indicating that the previous result should be parsed for the appropriate type of argument(s).

Through the use of prototype, we have working examples of maintenance of persistent, organized knowledge, global learning, and agent interaction. We are currently implementing support for typed, indexical reference to previous results.

Getting More Information

To obtain more information about the meeting send e-mail to: fscc98@pax.st.usm.edu.

Return to PAX
Return to USM

USM
inquire@delphi.st.usm.edu
Updated
http://pax.st.usm.edu