Behavioural Computation: Analysis, Models, and Algorithms for Supporting Human Improvement on the Web

点击翻译按钮获取中文摘要

The goal of this project is to develop new knowledge, models, and algorithms to support human learning and improvement in online social systems. A wide array of online social systems support people wanting to make progress. People want to learn new skills, improve their performance in personal and professional endeavours, and make better decisions---and do so in increasingly massive numbers online. For example, Wikipedia is an ever-growing global collaborative effort to summarise the world's knowledge, which others can learn from; tools like Duolingo and Coursera are learning platforms that support enthusiastic learners; and the trend towards quantifying and sharing our progress in activities like exercising is supported by sites like Facebook and Strava. In online systems such as these, people set goals and have external targets to aim for, and are motivated to achieve these milestones. The recent availability of these massive-scale datasets of years of human behaviour makes possible a new kind of computational science of motivation, improvement, and learning. Such a computational science is increasingly necessary, as the scale of these systems raises new algorithmic and computational challenges. By observing behavioural traces of millions of people trying to make progress, learn, and improve, we will observe large-scale patterns of human improvement that were previously invisible. We will use this knowledge to model paths of progress at an unprecedented resolution, heterogeneity, and accuracy. Using these models, we will then develop design principles and algorithms that support people trying to improve in a scientifically rigorous and informed way.This research is important for the millions of Canadians who use online social systems like Wikipedia, Coursera, and Facebook to learn and improve. These systems have been designed with our current understanding of human improvement, which has not yet fully exploited the new availability of massive datasets of people making progress online. Computational challenges like predicting exactly who is likely to be motivated by what kind of goal and when will be addressed and solved. When implemented on a large scale, algorithms using this much more detailed understanding will be able to personalize goals to best motivate people on an individual level. A little bit of extra motivation, multiplied over many goals and potentially thousands or even millions of people, would result in a significant boost for productivity and achievement across many domains.