HCC-Small: Web Games to Advance Interactive Learning Agents

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Machine Learning (ML) promises a way to build adaptive systems while avoiding tedious and tenuous pre-programming; however, ML techniques are not designed for input from naive users, remaining by and large a tool built by experts for experts. This project reframes the ML research question by considering systems that are meant to learn from everyday people, asking: "how can machines take better advantage of the input that an everyday person is able to provide?" To this end, project activities include the following: (1) Building a suite of short computer games that involve training interactive characters, spanning a wide variety of ML algorithms and a wide variety of interaction domains and (2) deploying these games on a website, and collecting data on the success of various techniques in learning from the average person on the web. The goal is for this interactive learning agents website to yield a broad and principled understanding of human interaction with ML systems. The success of this research agenda will have a significant impact, leading to the development of interactive machines that assist us with the large and small tasks of our daily lives. Results have the potential to provide a framework for any designer who wishes to incorporate learning and adaptation into an end-user application. Further, the research promises to bridge the Machine Learning and Human-Comptuer Interaction research communities, creating a mutually beneficial partnership.