ABSTRACT

For many problems, there are multiple solution paths that lead from the initial problem

state to the goal state. Different paths may require different amount of time and effort.

With experience, the problem-solver learns to choose solution paths that requires less

time and effort, and as a result, performance improves. For example, if the problem is to

drive to a particular destination in a city, numerous solution paths may exist. If the person

is new to the city, very little knowledge is available to decide which paths to take. In this

case, decisions on which paths to take may solely rely on simple heuristics, such as

hill-climbing. Unless the city is extremely complex, simple heuristics are usually

sufficient to lead the person to the destination. However, although simple heuristics are

usually sufficient to provide a solution to the problem, there is no guarantee that the

solution is good (or fast, in the current example). Fortunately, with experience, the person

may be able to acquire information about the speeds for various routes in the city. With

this kind of information, the person may be able to choose faster paths that lead to the

destination. Although many cognitive mechanisms have been proposed to account for this

kind of learning in problem solving (e.g. Anzai and Simon, 1979, Agre & Shrager, 1994,

Lovett and Anderson, 1996), not many studies have directly addressed the effects of

experienced effort in problem-solving, and how people learn to choose less effortful

solution paths with experience. In my dissertation, I am going to design several

experiments to understand how people acquire problem-specific information and how

they use the information to improve their performance. Specifically, I will focus on how

people learn the amount of effort involved in different solution paths, and how they

improve performance by choosing the less effortful paths. I am planning to build

cognitive models using ACT-R (Anderson & Lebiere, 1998) to understand the

mechanisms behind this kind of learning. In this extended abstract, I will focus on

describing the task and the model that I am planning to build.