| dc.contributor.author | MacInnes, W. Joseph. | en_US |
| dc.date.accessioned | 2014-10-21T12:37:20Z | |
| dc.date.available | 2004 | |
| dc.date.issued | 2004 | en_US |
| dc.identifier.other | AAINQ89810 | en_US |
| dc.identifier.uri | http://hdl.handle.net/10222/54617 | |
| dc.description | This research looks at novel combinations of machine learning techniques and models of human performance to create software agents that were both efficient and believable. It is the proposal of this dissertation that by using machine learning techniques that mimic cognitive theories of game playing, opponent modeling and deception, that improvements can be made with the performance of software agents. In addition, human traits will be examined as between-subject variables to explore individual differences in this spatial environment. | en_US |
| dc.description | Experiments were run pitting various combinations of human and software agents against each other in duels within the virtual arena. Three different intelligent software agents, each using three levels of recursive modeling were tested against human participants with three dependent variables. Each algorithm was tested for base accuracy in predicting its opponent's position, its effectiveness in fighting a human opponent and, its believability in portraying a human opponent (this last measure can be seen as a limited scope 'Turing Test'). | en_US |
| dc.description | It is interesting to note that the algorithm that performed the best on the accuracy measure was not able to translate that result into effectiveness. The algorithm that modeled human stereotyping and deception, however, not only performed far better on the measure of believability, but on effectiveness as well. Due to the lack of a direct link between the three measures, researchers and programmers should consider their objectives carefully before choosing an algorithm for their agents. In addition to these software results, it was shown that of the individual differences observed in this study, experience in spatial computer games played a larger role in performance than the gender of the participant. | en_US |
| dc.description | Thesis (Ph.D.)--Dalhousie University (Canada), 2004. | en_US |
| dc.language | eng | en_US |
| dc.publisher | Dalhousie University | en_US |
| dc.publisher | | en_US |
| dc.subject | Computer Science. | en_US |
| dc.title | Learning dynamic stereotypes for effective autonomous agents. | en_US |
| dc.type | text | en_US |
| dc.contributor.degree | Ph.D. | en_US |
