Assessing Implicit Gender and Ethnic Biases in Selecting Individuals for Software Developer Positions

摘要

This thesis presents a pre-research trial and a preliminary exploration of implicit bias based on the researcher's interest in ethics and the transfer of implicit bias to artificial intelligence (AI) applications. Human knowledge is used at some point in the development of knowledge bases for AI applications. As a result, the researcher was interested in determining whether humans do have implicit ethnic and/or gender biases. This pre-research trial is focused on whether implicit human bias exists and therefore might be transferred to AI applications. The research is limited to exploring implicit gender and ethnic biases in human respondents only. Pre-test and post-test questionnaires are completed by each respondent to collect sample data, e.g., gender, age, and ethnic background. A toy task was administered that required the respondents to rate very similar resumes of fictional job applicants. Respondents rated each resume from five (excellent application) to one (poor application or poorly qualified for the job). We discovered that implicit gender and/or ethnic bias seem not to exist in respondents aged 19-29, the majority of our sample. While the data shows that respondents did associate names with specific gender and ethnic groups, they did not use this information in rating the resumes. One rather disturbing result was noticed in the rating of Rebecca Johnson and Shaniqua Johnson. Of the ten resumes these two were rated as average; however, the eight other resumes were consistently rated higher. The small sample size of thirty-nine respondents may have resulted in this finding. This result suggests that further research is required; even though the data analysis overall suggests a lack of implicit gender and/or ethnic biases.