Subject Matter Experts Still Needed

One of the first things I learned many years ago about basic process control is that process control can only be as good as the process variable measurements fed into the control algorithms. In other words, even the best control algorithm can’t control the temperature or level in a reactor precisely without good temperature or level measurements or the flow of fluids through a pipe without good flow measurements.

That thought came to me this morning while reading an interesting article in the Boston Sunday Globe newspaper entitled, “The Bias in the Machine.” (Never mind that it’s Friday morning as I write this – it usually takes me several days to get around to reading this typically thought-provoking section of the Sunday paper.) The author’s premise, is that the artificial intelligence algorithms frequently employed today to help humans decide who will get a mortgage loan, a job offer, and so on often (unintentionally) share the biases of the humans who wrote those algorithms as well as the inevitable biases of the data used to “teach” the AI algorithms that learn by identifying patterns in the data.

While noodling this around over my first cup of morning coffee, another thought suddenly popped into my still-sleepy head; a pattern I had identified while reviewing many recent ARC case study Insights on how important it is for software engineers and/or data scientists to work closely with plant “subject matter experts” (process control engineers, operators, and maintenance technicians) when developing the computer programs and advanced analytics applications intended to either help plant personnel optimize the plant, or – in the not-so-distant future – possibly do so directly, without human intervention.

This is important because, while computers, automation, and other technology can help take human error out of the plant operation equation to “replicate the performance of the best operator in the plant on the best operating day,” there is also plenty of opportunity to inject human error (and/or ignorance) into the computer-based automation and analytics.

My point here is that – in these days of rapidly increasing IT/OT convergence and the loss of basic plant expertise due to attrition – it’s critical to keep in mind how important it is for IT and OT experts to collaborate closely to meld their respective knowledge and skillsets. Because just as it’s possible for automation, optimization programs, analytics, and other IT to capture and replicate the knowledge of the best plant engineers, operators, and maintenance people; it’s also possible replicate the old “garbage in/garbage out” paradigm that applies to even the most basic process control.