Trustworthy AI

Chapter 3Robust and Reliable

At BAM Inc., Mariam was having calibration problems and customers were complaining. The parts coming out of two manufacturing plants in Vietnam were sub‐par. Shipments were being returned, costs were rising, and Mariam could not figure out why those plants were having so many problems.

She had already dug through the local processes, the human talent, the management approach, and a dozen other things that could be causing the problem. And as she was considering the timeline of this poor performance, it dawned on her: the problems started after they deployed the AI system that executed calibration on one type of high‐precision grinding machine.

What was mystifying was that the machines abroad were all identical, as were the AI systems. The German plants manufactured to customer specifications, as did the North American plants. Why would the AI system work everywhere except in Vietnam? It was a question that demanded an answer, as more shipments came back and the complaints piled up in her inbox.

With AI model training, datasets are a proxy for the real world. Models are trained on one dataset and tested against another, and if the results are similar, there is an expectation that the model functions can translate to the operational environment. What works in the lab should work consistently in the real world, but for how long? Perfect operating scenarios are rare in AI, and real‐world data is messy and complex. This has led to what leading AI researcher ...