Identifying perceptually indistinguishable objects
Santore, John F
MetadataShow full item record
This dissertation reports on an investigation into how Perceptually Indistinguishable Objects (PIOs) can be identified. An experiment with 68 human participants was performed to investigate how and how well people identify PIOs. The experiment was designed as a protocol analysis experiment. Participants performed a video-game like task of counting or following, both of which entail identifying objects. The analysis of this experiment shows that the human participants had a marked preference for certain situations that they believed helped them identify the PIOs more readily. Participants would try, as much as possible, to keep themselves in these situations. A cognitively plausible computational model of identifying PIOs is developed from the results of the human subjects experiment. The cues and strategies that participants in the experiment went out of their way to use are examined and treated separately. Some participant-preferred strategies always lead to the correct answer/identification when the participant's background beliefs are correct. These strategies are generally perceptually based and are called base cases. The other set of strategies that the participants tried to use are not quite as perceptually based and are called intermediate cases. These strategies, when correctly used, lead to the right answer a great deal of the time, but are slightly more prone to failure than the base cases. The knowledge needed for the general case of identifying PIOs is also discussed and an algorithm for the model is included. Finally, a new simulated cognitive robot is described that includes an implementation of the computational model of identifying PIOs. The robot was tested in the same environment that the human participants used for the experiments and on the same tasks. The mistakes that the robot made fell into a subset of the mistakes that the human participants made in the experiment.