Making AI Outputs Match the Intended Goals (Validity)

Abstract

This chapter focuses on validity in artificial intelligence. It explains that the quality, representativeness, and context of data influence how well AI systems can perform. The chapter discusses three ways of judging AI validity: technical, interpretive, and normative. This includes correctness, interpretability or understanding ability, fairness, and respecting ethical standards. Now this chapter goes on to notice difficulties at the time of saving AI practice from entirely new areas, during which strict application of tried and tested verification methods, such as cross-validation or bias testing, has been applied to real problems set. What case study discussions highlight is that failure to comply with its common employment rules may have far-reaching effects, including: this costs money, and what goes wrong cannot easily be restored; biased decisions; compromised safety. The results reveal that AI systems must be closely monitored and continually changed in order to maintain public trust, fairness, and compatibility with societal values.

Ähnliche Arbeiten