Featured Product
This Week in Quality Digest Live
Innovation Features
Jennifer Chu
High-speed experiments help identify lightweight, protective ‘metamaterials’
Michael King
Augmenting and empowering life-science professionals
Rich Nobliski
Helping narrow the manufacturing skills gap with 3D tech
NIST scientists perfect miniaturized technique
Kristi McDermott
Technology and what the future requires for patients and providers to thrive

More Features

Innovation News
New tool presents precise, holistic picture of devices, materials
Enables better imaging in small spaces
Helping mines transform measurement of blast movement
Handles materials as thick as 0.5 in., including steel
For companies using TLS 1.3 while performing required audits on incoming internet traffic
Accelerates service and drives manufacturing profitability
Process contains leaks, improves thermal performance in buildings
Delivered by SpaceX launch to demonstrate in-space capabilities
Recognized among early adopters as a leading innovation for the life sciences industry

More News

Dirk Dusharme @ Quality Digest


What Is AI?

What it can and can’t do, and how it can set you free

Published: Monday, February 13, 2023 - 12:03

One of the biggest problems when reading about any kind of innovation in the press is the prevalent assumption that everyone understands the topic and how it works. Whether it’s cloud computing, edge computing, cold fusion, controlled fusion, or recently, artificial intelligence (AI), experts and reporters start extolling the technological miracle without offering even a simple explanation about what it is.

Articles about AI have been especially annoying, in part because of its name—AI is neither artificial nor intelligent—but also because it conjures up images of machines that think and reason, which is far from the reality.

When Jeff Dewar, CEO of Quality Digest, was at MasterControl’s Masters Summit 2022 in Salt Lake City, he had a chance to sit down with Viktoria Rojkova, MasterControl’s vice president of AI/ML and Data Science, and ask her to explain just what AI is. Her answers were clear, concise, and easy to understand.

Although the ultimate goal of AI is to create machines that perform tasks that would normally require human intelligence, first we must teach machines to learn. In other words, machine learning (ML) is used to achieve AI.

In short, AI ML looks for patterns in really large sets of data after initially being trained on known pattern sets. It is analyzing the probability that one set of data (the unknown, whether tabular, text, speech, or sensor data) is related to the training data. What are the odds that data x are related to data y?

“Mainly it’s a mathematical process of finding patterns in data,” says Rojkova. “And just like for humans, the more examples of these patterns that you have, the more accurate your predictions can be.”

Rojkova uses a common human resource problem as an example. How might AI ML predict whether an employee will be late with their training or some other project?

“As a human, you do it by looking at their past history, how a person is procrastinating, and how many delayed and overdue notices and tasks he or she already has,” says Rojkova. “Based on that, you can make a prediction and a decision.”

Rojkova is describing something that probably all of us have experienced and use every day. We see (and remember) that a person is regularly late for their tasks and predict this person will be late for this other task as well... it’s their pattern. As humans, we’re wired to recognize patterns; it’s a shortcut for faster human decision making.

Rojkova says ML does the same thing. But, rather than being based on instinct—a human property—ML is based on probabilities. A machine will pull up the historical records of human behavior and look for a pattern of missed dates and overdue notices, for instance. Based on those patterns, it can develop a probability for whether this person will miss the due date for a current project. The important thing is that along with detecting patterns (even patterns that humans might miss), it can do this analysis for millions of tasks and trainee—and do it very quickly.

As it’s been defined, AI is about more than just learning patterns. It can generalize based on those learned patterns. “For instance, a human might intuitively understand or learn that an ellipse is a derivation of an oval, or that a rectangle is probably a broader case of a square,” says Rojkova. “Machines right now, with artificial intelligence, are able to make that step.”

AI in the real world

When it comes to AI’s application, many vendors wave their hands and say, “But... but... it has AI,” as if that somehow tells you what it’s doing. But ask them just what that means and how AI is being used in their product, and you often get a blank stare.

Rojkova, on the other hand, gave some great, practical examples of how MasterControl is experimenting with AI. One of the most interesting was using it to anticipate what the U.S. Food and Drug Administration (FDA) might be doing and what their red flags are during an inspection. This is especially important information for the life sciences industry and for any FDA-regulated company that wants to avoid the dreaded Form 483 warning notice. Wouldn’t it be nice to know the FDA’s hot buttons, or the kind of violations currently getting the most scrutiny?

“By understanding the main themes and categories of FDA citations, we can help companies increase the likelihood of a successful audit,” says Rojkova. “So with that in mind, we took a machine-learning model that is learning a context.”

Rather than looking at tabular data, as in the training scenario above—which might have been accessing an Excel spreadsheet with employee performance—this experiment used natural language processing (NLP), a subfield of AI that deals with processing, analyzing, and generating human language to derive the most likely meaning of a sentence or phrase. In this case, it's used to analyze all the publicly available data the FDA collects on 483 notices, which includes the short and long descriptions.

Any data miner could access these data, but as Rojkova points out, it wouldn’t know what to look for or how to understand the context. MasterControl, as a life sciences company, does understand the context.

Viktoria Rojkova describes how AI can determine FDA inspector priorities by looking at Form 483 data. View the entire video here.

As the AI processes the large amount of 483 data, it develops groups of sentences or words that are very similar to each other but very different from other groups. Data scientists can then dive into each of those groups to understand the context of that group and what the context might be.

With the 483 data, you might see that the machine is flagging a group of descriptions that is all about gloves, for instance. “And the machine is smart enough to understand that the gloves only can be used in a context of sterilization,” says Rojkova. “It’s amazing to realize that there is no single word of ‘sterilization’ in a sentence. The model is learning the context.”

Furthermore, having learned the context, AI can understand that a particular citation was for a sterilization issue even though neither gloves nor sterilization was mentioned (for instance, the citation might mention “bare hands”). Because the machine has learned the context of groups of words, it’s able to statistically predict what they refer to—in this case, a violation that involves sterilization.

It’s possible then to understand what kind of citations the FDA is issuing, and in what context, in a particular industry. Armed with that, companies can ensure they have safeguards in place so they’re not making the same types of mistakes flagged by the FDA.

Another way of thinking of it is that AI helps align the priorities of industry players with that of the FDA, something long discussed at industry events.

In the end, what AI and ML do is take the drudgery out of finding patterns (looking for the needles in the haystack) and instead allow employees to focus on innovation, risk management, problem solving, and the other tasks that humans are best at and enjoy more.

“Mundane data can be crunched and analyzed by a machine so much faster, and everything that is above, that is your job,” says Rojkova. “Now, you can create, you can think, you can brainstorm and initiate instead of being buried in a pile of Excel files. In other words, it frees you up to use your limited time to do the things that you wanted to do but never had time to do.”


About The Author

Dirk Dusharme @ Quality Digest’s picture

Dirk Dusharme @ Quality Digest

Dirk Dusharme is Quality Digest’s editor in chief.


Narrow AI for Quality Improvement Projects

Narrow AI limits itself to a single task.

Years ago, I found ways to use Excel PivotTables to digest large amounts of defect data and turn it into improvement projects consisting of control charts, Pareto charts and Ishikawa diagrams.

Doing it manually took many hours to many days, but that analysis was the source of many multi-million dollar improvement projects. 

This is where the invisible low-hanging fruit exists in any company.

I later automated the analysis to collapse the time into just a few minutes. 

Taking a repetitive human task like data analysis and automating it isn't full-blown AI, but it's a step in the right direction.