In this article the author deals with how machine learning can address the issues of fairness and the effect algorithmic decision making formulated based on bias data negatively effects peoples lives. The author focuses on three different application of Machine Learning in which this ethical issue must be addressed.
The first being COMPAS a criminal risk assessment software utilized by the justice system in order to asses potential recidivism risk. In this software data such as an individuals misconduct history, affiliation, employment status, etc is used to to aid a judges decision of the probability such an individual would be a reoffender. The author argues that due to the data the system uses to formulate its probabilities potential bias against African American individuals is present.
The second application is the use of machine learning in medical prediction. The hypothesis tested by the author is that because large pool of data collected in order to make medical predictions stems from medical experiments with a majority white male subjects, it may not be compatible with equality of accuracy across demographic groups.
In his third application high stake decision-making such as prediction of life expectancy the Author explores the tension between inequality aversion and accuracy.
In this article the author deals with how machine learning can address the issues of fairness and the effect algorithmic decision making formulated based on bias data negatively effects peoples lives. The author focuses on three different application of Machine Learning in which this ethical issue must be addressed.
The first being COMPAS a criminal risk assessment software utilized by the justice system in order to asses potential recidivism risk. In this software data such as an individuals misconduct history, affiliation, employment status, etc is used to to aid a judges decision of the probability such an individual would be a reoffender. The author argues that due to the data the system uses to formulate its probabilities potential bias against African American individuals is present.
The second application is the use of machine learning in medical prediction. The hypothesis tested by the author is that because large pool of data collected in order to make medical predictions stems from medical experiments with a majority white male subjects, it may not be compatible with equality of accuracy across demographic groups.
In his third application high stake decision-making such as prediction of life expectancy the Author explores the tension between inequality aversion and accuracy.