기계학습의 활용에서 발생하는 편향과 차별: 학습자료의 활용

Abstract

The use of artificial intelligence is expanding to various fields of society as Korea enters the 4th industrial revolution. Artificial intelligence based on machine learning causes problems such as bias and discrimination due to biased data along with the positive aspects of improving national and individual productivity and quality of life in society and economy. Therefore, in this study, the level of errors in prediction and classification that may occur by using biased learning data was verified. Specifically, this study verified the error of prediction that is expected to be obtained using artificial intelligence by applying the biased algorithm to actual data. The main results are as follows. First, prediction level decreased when the attributes of the population represented by the learning data and the actual data were different. Second, errors in prediction and classification increased when the learning data contained predictors that had a biased influence on the criterion variable. This study emphasized the importance of unbiased learning data to properly apply the algorithm using artificial intelligence.