Quality Assurance in and for AI

Abstract

With AI influencing important decisions in companies, organizations, and our private lives, the quality of AI models becomes a concern for managers and citizens alike. Sales success depends on targeting customers with personalized and fitting products. In the medical area, researchers want to identify Covid-infected persons by letting an AI model analyze how persons cough into a microphone. These two simple examples prove: AI passed the tipping point from harmless, academic experimentation to real-life solutions. The time has come that AI projects require auditable quality assurance and testing. Errors can impact financial results severely or threaten human lives. Data scientists, test managers, and product owners cannot continue treating AI models as magical, always correct black boxes created by brilliant-minded, impeccable, and sacrosanct specialists.