My AI Was Wrong!

Th2     Start Time : 09:45     End Time : 10:30

Talk Abstract:

“My AI was wrong,” but isn’t it more like “I was wrong about my AI”?

The difficulty in testing AI comes from the ambivalence of this proposal. To differentiate between the 2 assumptions mentioned, the presentation will briefly show what an AI is (historical), how it learns (machine learning) and how it can be misled, voluntarily or not (cognitive bias or learning bias), and in which sectors of our daily lives they intervene today, and all this with playful illustrations. This is to emphasize the difficulty of testing AI, on the one hand because of the unpredictability of the expected result (problem of the “classical” test oracle) and on the other hand by the shortening of the development cycle of all these new features (IoT, Robotics…).

To enable us to overcome those problems, the presentation focus is placed on the metamorphic testing method. This will allow to present a pragmatic method of testing AI using heuristic tests, but also respecting the good principles of agile testing, essential in this context. Concerning testing, this revolution is clearly in line with previous ones (agility, continuous testing, devops…) and requires some specific skills (data science, probability…). We will see how to create a good test strategy for AI testing.

AI can also help (or even replace…) the tester in his work, and this seems essential with the acceleration of changes and the increase in the technicality of the systems. So rather than being afraid of AI, you have to use it and tame it. Finally, it becomes certain that testing with an AI will quickly become the only way to “keep pace”. So we will also see how to test with an AI.

Key Takeaways:

  1. Learn about metamorphic testing methods for AI
  2. Understand what AI is, and how it learns
  3. Understand how AI can be mislead (biased) during its training

Back to Programme

    • Speaker


      Laurent Bouhier - Director, acpqualife Sàrl, Switzerland

      Laurent Bouhier is a French computer science engineer with more than twenty years of professional experience in the test. He is also a an experienced trainer and speaker. He lives in Switzerland for 3 years to develop testing and agility for different companies. Laurent is now an expert in DevOps and Artificial Intelligence (AI) testing. He is now working on the development of test management using deep learning, and participate to some AI development and tests.