Seite - 203 - in The Future of Software Quality Assurance

The Future of Testing 203 role.After all, back in 1998who would have thoughtwe would be havingbusiness cards todaywith terms like ‘Scrum-master’or ‘Product-owner’ . . . 2038 It may be near impossible to predict the future 20 years ahead. But let’s give it a try. In 2038 we’re probably driving autonomouscars, and robots may very well be part of everyday life. Artificial Intelligence will be a lot more intelligent than it is today and may even be replacing knowledge workers. Will there still be software developmentaswe knowit? Andwill therebesoftware testing? I personally believe we will still be needing people to develop software even though it’s probable we’ll be able to produce a lot more software with a lot less people. It is possible software has become intelligent enough to be testing itself. Software may be continuously runningself-diagnosticswhich indicate when something’s going wrong and the software in question might even be able to fix itself, at least toa certaindegree. ButIdon’tthinkcomputersandrobotswillhavereplacedeveryoneinvolvedwith the development of software. Simply because of one thing: Artificial Intelligence is only intelligent about certain things, but really unintelligent when it comes to some other things . . . This is clearly visible today, but I don’t expect it will be all that different in20years fromnow.Takea lookat current-daytools like Google AssistantorSiri.TheseAI’sknowalotmorethananyhumanbeingknows(because they have an endless supply of information constantly at their disposal). Some robots today are amazing at interpreting their surroundingsand figuringout what’s expected of them. Current prototypes of autonomous driving cars may very well already be safer than human drivers. However, even with all this computer power and all this data and intelligence there’s still one thing at which every AI sucks: understandinghumanbehaviour. A great example is Honda’s humanoid robot Asimo, which was developed a few years ago. In every single way this was a great feat of engineering. However, during a demonstration it failed horribly because of one simple misunderstanding of human behaviour. The robot didn’t understand why people would want to take pictures of it and thus concluded that when people were raising their camera or mobilephone to take a picture, theywere raising theirhands to ask a question.The robotfroze,repeatingoverandover“WhowantstoaskAsimoaquestion?”[3].And this interpretation of human behaviour is something current AI still doesn’t know how to do, even though it’s somethingwe humansfind veryeasy! We immediately understand what’s happening when someone is hanging out of the window of a train that’s about to leave to hug someone outside. They’re saying goodbye.Pretty straightforward, right? However, an AI might mistake it for someone trying to pull another person out of the train. There might be something wrong in the train, an evacuation might even be necessary . . . . I don’t believe AI will be much better at interpreting human behaviour in just 20 years from now. And therefore it’s very likelywe’llstillbeneedinghumanstodevelopsoftwarethatwillactuallydowhatan end-user isexpectingfromit.Andwhenwe’restillbuildingsoftwarewe’llalsostill need someone to act as the earlier mentioned quality conscience. Someone who’s