The new technology will be equally valuable to managers in industry, academia, government and the military.
The project, nicknamed "RADAR" for Reflective Agents with Distributed Adaptive Reasoning, will help its human master with tasks like scheduling meetings, allocating resources, creating coherent reports from snippets of information, and managing email by grouping related messages, flagging high priority requests and automatically proposing answers to routine messages.
The goal is to develop a system that can both save time for its user and improve the quality of decisions. RADAR will handle some routine tasks by itself, will ask for confirmation on others, and will produce suggestions and drafts that its user can modify as needed. Over time, the system must learn when and how often to interrupt its busy user with questions and suggestions. To accomplish all this, the RADAR research team must employ techniques from a variety of fields, including machine learning, human-computer interaction, natural-language processing, optimization, knowledge representation, flexible planning and behavioral studies of human managers.
The RADAR project's principal investigators include SCS professors Daniel P. Siewiorek, director of Carnegie Mellon's Human-Computer Interaction Institute; Jaime Carbonell, director of the Language Technologies Institute; and Principal Research Computer Scientist Scott Fahlman. The project will initially focus on four tasks to illustrate how the system's learning curve increases people's productivity: email, scheduling, webmaster and space planning.
"With each task, we'll run experiments to see how well people do by themselves and make comparisons," Siewiorek said. "We will also look at people plus a human assistant and compare that to the software agent."
In addition to working on these four specific tasks, the project will develop cross-cutting technologies that can be used in all of these tasks and in other personal-assistant tasks as well. These include a shared knowledge base, a module that decides when to interrupt the user with questions, and a module that extracts information such as meeting requests from email messages written in English.
"The key scientific challenge in this work is to endow RADAR with enough flexibility and general knowledge to handle tasks of this nature," said Fahlman. "Like any good assistant, RADAR must understand its human master's activities and preferences and how they change over time. RADAR must respond to specific instructions--i.e. 'Notify me as soon as the new budget numbers arrive by email'--without the need for reprogramming. But the system also must be able to learn by interacting with its master to see how he or she reacts to various events. It must know when to interrupt its master with a question and when to defer."