Estimated reading time: 6 minutes

The Right Approach: Old Philosophies, New Tools

About six years ago, I wrote a series on the Founding Fathers of Quality. With AI taking over the world, I wanted to posit how these quality gurus would apply this technology to their core quality management philosophies.

Quality management as a discipline was built by a handful of people who never touched a neural network, yet whose ideas map onto AI with almost eerie precision. Dr. Walter Shewhart, Dr. Edwards Deming, Dr. Joseph Juran, Philip Crosby, Dr. Kaoru Ishikawa, and Shigeo Shingo each built a philosophy around a specific lever for improving quality, including statistical rigor, reducing variation, prioritizing effort, eliminating defects, involving the workforce, and error-proofing the process itself.

Rather than AI replacing any of those levers, AI sharpens each one, and imagining how each of these individuals would actually put it to work says as much about their original thinking as it does about the technology.

Shewhart: AI Bound by the Scientific Method
Shewhart's contribution predates all five of the figures below and quietly underlies each of their frameworks. He invented the control chart itself, and with it the statistical distinction between "chance cause" (common) variation and "assignable cause" (special) variation that Deming would later carry forward and popularize.

Shewhart's real achievement was insisting that any claim about a process being "in control" rests on a defensible statistical basis, not intuition or a manager's gut feel. He also originated the Plan-Do-Study-Act cycle that Deming would popularize as the PDCA cycle.

Shewhart would likely be the most rigorous skeptic of AI. A model that outputs an anomaly score is, in his terms, an assignable-cause detector, but unlike a control chart built on known statistical distributions, most machine learning models don't come with a transparent, auditable basis for why a given score crossed a threshold.

He would push hard for treating model validation with the same seriousness he brought to control limits: back-testing against historical data, quantifying false-positive and false-negative rates the way a control chart's three-sigma limits quantify the odds of a false alarm, and refusing to treat a model's output as authoritative until its statistical behavior is understood.

For Shewhart, the danger of AI wouldn't be the technology itself, but rather organizations trusting a black box the same way they once trusted an unproven chart without doing the scientific work to justify that trust. "Explainable AI," in his framing, is the modern equivalent of the statistical theory that made the control chart trustworthy in the first place.

Deming: AI as the Ultimate Variation Detector
Deming's entire framework rested on the insight that most quality problems come from variation in the system, not from the people working in it. His statistical process control charts existed to separate "common cause" variation, the noise inherent in any process, from "special cause" variation that signals something has actually changed. He was fiercely opposed to tampering, meaning adjusting a stable process in response to normal noise, which only makes performance worse.

Deming would almost certainly see machine learning as an extension of the control chart, not a replacement for it. A model trained on process data can detect subtle multivariate shifts that a single-variable Shewhart chart would miss, with several parameters drifting together in a way no individual chart would flag.

But he would apply the same discipline to AI outputs that he applied to control limits: an anomaly score is not, by itself, permission to act. Any AI-flagged deviation still has to run through the same root-cause discipline before anyone touches the process, because a model that "detects" noise and triggers unnecessary adjustment is just tampering with better technology behind it. His 14 Points would likely gain a 15th in spirit: Understand your model as thoroughly as you understand your process, or you've automated the guesswork you were trying to eliminate.

Juran: AI as a Prioritization Engine
Juran's contribution was less about statistics and more about management focus, the Quality Trilogy of planning, control, and improvement, and his adaptation of the Pareto principle into the “vital few and useful many.” For Juran, the central question was always where to spend limited attention, and which a handful of causes account for most of the defects, cost, and customer complaints.

AI is, in a very literal sense, a Pareto engine at scale. A model that ingests warranty claims, inspection records, supplier data, and field failures simultaneously can rank root causes by actual financial and quality impact far faster than a manual Pareto analysis, and continuously rather than as a quarterly exercise.

Juran would likely push organizations to use AI first for the unglamorous work of cost-of-quality analysis, accurately attributing the financial impact of failures so that improvement projects get funded on evidence rather than on institutional politics. He'd also warn that a prioritization engine is only as trustworthy as the data governance behind it; garbage inputs produce a beautifully ranked list of the wrong problems.

Crosby: AI in Service of Zero Defects
Crosby's “Zero Defects” philosophy was as much cultural as technical, and he is famous for his claim that quality is free because the cost of prevention is always lower than the combined cost of failure, appraisal, and rework. He was deeply skeptical of “acceptable quality levels” that tolerated a planned percentage of defects, arguing that any tolerated defect rate becomes a self-fulfilling target.

Crosby would likely embrace AI-driven defect prediction enthusiastically, precisely because it shifts quality left, catching a likely failure during design or early production rather than paying for it in the field. Computer vision systems that flag a solder joint or a weld before it fails inspection fit his philosophy exactly: the earlier the catch, the closer the cost gets to zero.

But he'd push hard against a subtle risk of treating a model's confidence threshold as a new acceptable quality level, quietly tolerating a defect rate the algorithm doesn't flag rather than driving toward true prevention. For Crosby, the measure of quality was always the cost of nonconformance, and he'd insist AI systems report in those terms rather than abstract accuracy scores.

Ishikawa: AI as a Democratized Fishbone Diagram
Ishikawa built his reputation on making quality tools accessible to the entire workforce, not just engineers and statisticians. The fishbone (cause-and-effect) diagram and quality circles were both designed to provide frontline workers with a structured way to contribute to root cause analysis. His principle of company-wide quality control held that the people closest to the work often see problems that no dashboard can capture.

He would likely see generative AI's biggest contribution not in the algorithms themselves, but in translation: a natural-language interface that lets a machine operator describe a problem in plain terms and get back a structured cause-and-effect breakdown, or that turns scattered shop-floor observations into a properly categorized fishbone diagram without requiring a black belt to build it.

His skepticism would center on exactly what AI tends to erode if used carelessly: the participatory, cross-functional discussion that made his quality circles work. A model that outputs a root cause without a team ever discussing it together produces an answer, but not necessarily buy-in, and not the tacit knowledge that only comes from people arguing over a whiteboard.

Shigeo Shingo: AI as Poka-Yoke at Scale
Shingo's most durable contribution, poka-yoke, was about making mistakes physically impossible rather than catching them after the fact: A part can only be inserted one way, or a jig won't close if a component is missing. His related concept of Zero Quality Control argued that source inspection, built into the process itself, should make sampling inspection obsolete.

AI-enabled vision systems and sensor fusion are close to a direct realization of that idea at a scale Shingo couldn't have built by hand: a camera that halts a line the instant a part is misaligned, a model that predicts a tool is about to drift out of tolerance, and stops the cycle before a bad part is even produced. Shingo would push hard for AI to be embedded at the point of creation rather than the point of inspection; an error-proofing device, not a dashboard reviewed after the shift ends.

The Common Thread
None of these six would treat AI as a replacement for the discipline they spent their careers building. Each would use it to do what they already believed in, faster and at greater scale, including grounding decisions in statistical evidence, reducing variation, focusing effort, preventing rather than detecting, involving the workforce, and error-proofing the process itself.

The risk they'd each flag, in their own language, is the same one quality management has always faced: mistaking a more sophisticated tool as a substitute for management commitment, root-cause thinking, and a culture that treats every defect as a problem worth understanding rather than a statistic worth tolerating.

This column originally appeared in the September 2026 issue of I-Connect007 Magazine.