Seite - 78 - in The Future of Software Quality Assurance

78 R. Marselis that can do continuous delivery of AI systems. The classical test engineer has to evolve and incorporate new skills like data analysis, AI algorithms, or (as we will see at theendof this chapter)weather forecasting. In thischapter,wewillelaboratefirstonthetestingof intelligentmachines.After that we will focus on testing with intelligent machines, which means the use of intelligentmachines to support testing. 2 TestingOf IntelligentMachines Artificial intelligence can (and should) be tested. In this chapter, we talk about the testingof AI. Since AI solutionsare quite new,experience in this field is scarce. Testingof AI has toformulateandevaluatecompleteandstrongacceptancecriteria thatverifythe outcome.Theoutcomeisdeterminedby the inputdataanda trainedmodel.Testing those is the core activity. The quality of cognitive IT systems that use artificial intelligence needs to be assessed. The challenge in this case is in the fact that a learningsystemwillchangeitsbehaviorovertime.Predictingtheoutcomeisn’teasy becausewhat’scorrect todaymaybedifferentfromtheoutcomeoftomorrowthat is alsocorrect.Skills that a testerwill needfor this situationare related to interpreting a system’s boundaries or tolerances. There are always certain boundaries within which the outputmust fall. To make sure the system stays within these boundaries, thetestersnotonlylookatoutputbutalsoat thesystem’s input.Becausebylimiting the inputwe can influence theoutput. 2.1 SixAngles ofQuality for MachineIntelligence People have been assessing the quality of things for centuries. Since the invention of thesteamengine theneedfora structuredapproachtoqualityassessment rapidly grew. After the creation of the first computers in the 1940s people realized that these “decision-making and data-processing” machines again needed a new angle to quality, and the first approaches to testing were published. Nowadays, we have test methods such as TMap, approaches (e.g., exploratory testing) and techniques (e.g., boundaryvalue analysis) to establish the quality of IT processes and to build confidencethat business successwill beachieved. Different angles can be used to assess quality. Some angles are known from a traditional need for quality (think mechanical or electrical). These angles are brought to life again because the digital age brings new technologies (e.g., 3D printing in themechanicalworld). The quality approach for a machine intelligence solution must address the followingsix anglesofqualityshowninFig.1.