Adversarial Testing for AIs: Get Ready to Fuzz Things Up!

With the “AI Boom”, testing AI will be a vital skill for every tester. But, how can you test a system with few expected results or test oracles? This talk will discuss the vital skills that Quality Engineers offer to AI/ML development as well as different tools and techniques that you can use to start testing AI. Ben will discuss how you can automate “Adversarial Testing”, a technique which provides inputs to see if an AI produces undesirable responses. You will learn a technique known as “Fuzzing”, in which testing provides an array of inputs to see if a program produces a result that is never desirable. Fuzzing will be compared with another AI testing approach, Generative Adversarial Networks (GANs), illustrating how you can adapt your test automation skills to formulate Fuzz test assertions. Gain an understand of how you can further apply your test engineering skills to Fuzz Testing. In Fuzzing, there are vastly more possible inputs than could ever be executed. There is a need to reduce the number of inputs, without increasing negative risk. Join Ben to discuss tools like “glitch tokens” and wrappers, which you can use to optimize Fuzz testing, further upskilling as a tester in the AI Spring.