There are many settings where people set long-range goals and make plans to achieve them. Such long-range planning is becoming an integral of the experience in many on-line contexts, where for example people work toward reputational milestones in question-answering sites, build up to administrative roles in open-source authoring domains, and reach educational goals in on-line learning communities.
We study this process using an underlying set of network models for the process of planning over multiple stages. These models allow us to flexibly incorporate some of the human behavioral biases that come into play when people attempt to reach long-range goals --- particularly, the tendency toward present bias, in which costs and benefits incurred in the present receive particular weight. We show how our network models directly produce a wide range of qualitative phenomena observed in the literature on present-biased behavior, including procrastination, abandonment of long-range tasks, and the benefits of reduced sets of choices. We then explore a set of analyses that quantify over the set of all planning problems in our formalism; among other results, we provide bounds on the cost incurred by a present-biased agent relative to the optimal plan, and we show that any instance in which this cost significantly exceeds the optimum must contain -- in a precise graph-theoretic sense -- a large ``procrastination'' structure.
This talk is based on joint work with Sigal Oren and Manish Raghavan.