2022, Vol., No.38

PHILOSOPHY·THOUGHT·CULTURE 2023 KCI Impact Factor : 0.5

pISSN : 1975-1621 / eISSN : 2092-0873
https://journal.kci.go.kr/insewtdgu

In this paper I investigate the problem of bias that threatens the rationality of data-based decision-making in the data-driven society we are entering, and seek ways to minimize that threat and to enhance the fairness of such decision-making. With this purpose in mind, I examine recent criticism about COMPAS that has been used for criminal correctional management in several states in U.S.A. COMPAS has been criticized that it (re-)produces biases in the evaluation of the criminals that varies depending on the criminal's skin color. I analyze two sources of such biases, the dataset and the algorithm. This analysis reveals that the data are not neutral image that reflect reality as it is, but product of measurements that embodies the perspectives of data-producers. Such intervention from specific perspective does not in itself endangers the fairness of data-driven decision-making. However, the bias in which the properties of the sample data deviate from those of the population reduces the effectiveness of data-based decision-making and thus undermines its core value. Therefore, a data-driven society should keep an eye on the biases in data-based decisions, perform regular assessments and revisions of the data processing algorithms and supplement the dataset itself coping with the result of the assessments.

Level 3 automated driving car is coming. The most crucial issue in safety management of level 3 automated driving car is considered to be reducing risks in takeover, where a driver takes the control back from an automated driving system(ADS). Takeover may undermine not only a driver’s own but also others’ rights and freedom, it raises an serious ethical concern. However, implementing safe takeover is facing a number of technical problems. This article attempts to analyze ethical, normative implications of such problems. Through this analysis, one might find out what specific obligations of both the driver and ADS have. To this end, first, this article surveys the definition of level 3 automated driving car, the notion of takeover and related concepts. Second, the problem of individual differences in takeover and that of driver monitoring are presented, and norms they suggest are articulated. Both tell us that for safe takeover, ADS is obliged to be personalized and the driver is obliged to comply. These observations indicate the importance of diriver-vehicle coordination that has not been emphasized enough so far in ethical approaches to automated driving car.

As artificial intelligence systems exert more influence in various areas, the need for artificial intelligence ethics increases and its role is expanding further. Today, interest in artificial intelligence ethics guidelines for solving problems with artificial intelligence systems tends to be discussed as if artificial intelligence ethics represents all ethics and compliance with artificial intelligence ethics guidelines can solve many problems caused by artificial intelligence systems in our society. In this paper, I would like to argue that excessive interest and emphasis on these artificial intelligence ethics guidelines and principles have room to overlook the fact that artificial intelligence ethics is “human ethics” and that “artificial intelligence ethics as existing ethics” should be discussed. First, rather than being discussed separately from the society in which artificial intelligence technology will be used, the principles of artificial intelligence ethics should be dealt with in the context of the society in which the technology is being used. Second, among the principles of artificial intelligence ethics, efforts to mitigate artificial intelligence bias to protect the value of fairness have room to reduce the meaning of fairness.