Quantifying inductive bias: AI learning algorithms and Valiant's learning framework
Artificial Intelligence
Exploiting convergence to improve natural language understanding
Interacting with Computers
Watch what I do: programming by demonstration
Watch what I do: programming by demonstration
Prototyping an intelligent agent through Wizard of Oz
CHI '93 Proceedings of the INTERACT '93 and CHI '93 Conference on Human Factors in Computing Systems
Instructible agents
Beyond Intelligent Machines: Just Do It!
IEEE Software
Machine Learning
Machine Learning
Varying the user interaction within multi-agent systems
AGENTS '00 Proceedings of the fourth international conference on Autonomous agents
A multiagent system for U.S. defense research contracting
Communications of the ACM - The disappearing computer
ELA—A new Approach for Learning Agents
Autonomous Agents and Multi-Agent Systems
Improving Computer Supported Cooperative Design With Personal Assistant Agents
Journal of Integrated Design & Process Science - Computer Supported Cooperative Work In Design
Hi-index | 4.10 |
Graphical user interfaces have helped center computer use on viewing and editing, rather than on programming. Yet the need for end-user programming continues to grow. Software developers have responded to the demand with a barrage of customizable applications and operating systems. But the learning curve associated with a high level of customizability-even in GUI-based operating systems-often prevents users from easily modifying their software. Ironically, the question has become, "What is the easiest way for end users to program?" Perhaps the best way to customize a program, given current interface and software design, is for users to annotate tasks-verbally or via the keyboard-as they are executing them. Experiments have shown that users can "teach" a computer most easily by demonstrating a desired behavior. But the teaching approach raises new questions about how the system, as a learning machine, will correlate, generalize, and disambiguate a user's instructions. To understand how best to create a system that can learn, the authors conducted an experiment in which users attempt to train an intelligent agent to edit a bibliography. Armed with the results of these experiments, the authors implemented an interactive machine learning system, which they call Configurable Instructible Machine Architecture. Designed to acquire behavior concepts from few examples, Cima keeps users informed and allows them to influence the course of learning. Programming by demonstration reduces boring, repetitive work. Perhaps the most important lesson the authors learned is the value of involving users in the design process. By testing and critiquing their design ideas, users keep the designers focused on their objective: agents that make computer-based work more productive and more enjoyable.