Seite - 128 - in The Future of Software Quality Assurance

128 G.Numan 3.1.4 Over-fitting Over-fitting occurs when the labelling is too diverse and too manifold for the purpose of the AI system. If we want to see patterns and groupings, a high granularity of labels compromises the outcome, making it unusable because of its indecisiveness. 3.1.5 Outliers Outliersareextremeexamples thathave toomuchinfluenceon thealgorithm.If the first cat your1-year-oldchild sees is a Sphynx,a naked race, this will have a major impact on his concept of a cat and will take multiple examples of normal cats to correct. 3.1.6 ConfoundingVariables Pattern recognition and analysis often requires combining data, especially when causal relations are looked out for. Confounding variables occur when different data patterns are associated for data analysis purposes that have no real causal relation. It has often been believed that drinking red wine could evoke a migraine attack,becausedrinkingredwineandmigrainesreportedlyoccursequentially.New research has shown that a migraine attack is preluded by changes in appetite, such as a craving for red wine. Drinking red wine is a side effect and not a cause of migraine! 3.2 Over-confidence inAI AI can perform some types of mental activities on a scale and with a velocity and precision that is unachievable by humans. The algorithm of AI is not directly accessibleoradjustable.FromthistheintuitioncanbeeasilyobtainedthatAIcannot be judgedbyhumanstandardsand is superior. Intellectual lazinessandcomfortcan beanimportantmotivationtooforuncriticallytrustingAI.Whoquestionstheresults ofGooglesearch? A possible consequence of over-confidence is the transfer of autonomy to an instance outside of our individual or collective consciousness. AI does not need to achieve self-consciousness to be able to do this, as sci-fi teaches us. It takes over- confidenceor laziness.