18. Information ethics as a theoretical foundation for ethical assessment and moral design of AI systems

Abstract

AI systems have transformed the way we consume, inform ourselves and interact, but they are entering spheres of activity where the particular vulnerability of users makes it all the more urgent to have adequate ethical evaluation tools. For example, in Quebec, a company that provides information systems and data management solutions for student records to colleges has developed a system that allows them to target students at risk of failing different courses. The aim is to design a tool that would optimise the use of data already recorded in electronic student records in order to add value by analysing them using an AI system. Unlike the manual student tracking program, this software would make it possible to cross-reference a very large amount of available data: high school grade point average, results of ministerial evaluations, absenteeism rates, results of mid-term evaluations, first language, etc. The system would make it possible to automatically identify the student's needs and expectations. It would make it possible to identify the particular needs of students and which students are considered at risk of failing or dropping out of college. They could then get quicker access to success support resources that are appropriate to their needs. Given the high dropout rate among the college population and the political pressure on college administrations to promote success, it is not surprising that several colleges are considering implementing this system. Student associations are concerned about the implementation of such software, given that its use requires the processing of students' personal data. Some professors are also concerned about the psycho-social impact of the implementation of this monitoring program on the student population. In addition, administrators are concerned about the accuracy of the predictions made by this algorithm and whether students from certain groups may be unduly identified as being at risk of failure. It appears that classical ethical theories, stated at a high level of abstraction and focusing on the moral character of individuals' actions, are of little use when we attempt to weigh up the ethical benefits and risks associated with the use of such an AI system in education. Furthermore, while the many charters and declarations that have been published in recent years can be a useful resource for identifying the broad principles that AI systems should respect, the fact that these are not based on a clear ethical theoretical framework that would guide the interpretation of these principles in context, compromises the effectiveness of their application. We believe that the development of an appropriate theoretical framework for these principles is necessary in order to be able to provide practical tools for the ethical assessment and moral design of AI systems.

Ähnliche Arbeiten