Engineering Application of Artificial Intelligence & Machine Learning (Part 3)

Impact of Domain Expertise on Engineering Application of AI

Outline

Part 3 — Today’s Problems of Engineering Application of AI in Academia and Industry

AI Experts or Traditional Statisticians with Minimal Understanding of Engineering

Traditional Engineers

Engineers with Superficial Understanding of AI & Machine Learning

Conclusions

Today’s Problems of Engineering Application of AI in Academia and Industry

Today, Artificial Intelligence and Machine Learning is a technology that almost everyone is interested in. This technology has proven to accomplish so much while being incorporated in so many areas such as social media, online search, stock market, the interaction of companies with their individual clients such as Netflix, banks, insurance companies, travel agencies, and even in security. Therefore, it makes sense when engineers express interest to incorporate this technology into their everyday work.

Given the fact that the engineering application of AI and Machine Learning is not exactly the same as the application of this technology to non-engineering related problems, there have been some issues associated with how some scientists and engineers in academia and industry have been using this technology to solve engineering related problems. This is the main context of this section of the article.

In order to explain what needs to be done and what needs to be avoided when AI and Machine Learning are used to solve engineering related problems, let us start with the name of this technology: “Artificial Intelligence” and “Machine Learning”. “Artificial Intelligence” has to do with “Intelligence” that is not natural but rather “Artificial”, and “Machine Learning” has to do with “Learning” that is applied to “Machines” and computers, not to humans. Therefore, the two words to be defined and discussed are “Intelligence” and “Learning”.

“Intelligence” is defined as “the ability to acquire, retain, and apply knowledge, learn from experience, adapt to new situations, and solve problems” and “Learning” is defined as “the process of acquiring new understanding, knowledge, and skills”. After defining these two items, let’s start by asking two simple questions about engineering problem solving:

1. Is general, human level, “Intelligence” enough to solve engineering related problems?

2. Do humans need to “Learn” anything specific to be able to solve engineering related problems?

I think the answers to the above two questions are quite obvious. To become an engineer, humans need to go to university after they get their high school diploma. Then after four years of “Learning” at the university, they become engineers with bachelor’s degrees. Furthermore, “Learning” at the university goes beyond just providing information to the students through books, videos, and articles. Engineering students go through an extensive learning process in four years before becoming engineers. “Learning” at the university includes “Teaching”. These simple answers to the questions regarding “Intelligence” and “Learning” means that application of Artificial Intelligence and Machine Learning in engineering related problem solving (a) requires engineering knowledge and domain expertise and (b) requires “Teaching” through detailed understanding and communication of the data to the machine learning algorithms.

This goes way beyond just throwing the available data to the AI & Machine Learning algorithms to solve problems, as mostly done when people use traditional statistics for data analysis. Today, the main issues associated with the engineering application of Artificial Intelligence and Machine Learning in the academia and industry has to do with three major problems:

Problem #1: When AI or statistics experts that have minimal understanding of engineering are identified as the main leaders and scientists to address and solve engineering related problems using data,

Problem #2: The involvement of traditional engineers that do not believe that AI and Machine Learning is a science and technology and claim that AI and Machine Learning is a hoax (same as Covid-19) and should not be used to solve engineering related problems, and

Problem #3: The engineers that have a minimal and superficial understanding of AI and Machine Learning and are mostly interested in it from a business and marketing point of view rather than science and technology.

Let’s go into a bit of detail on these three issues.

● AI Experts or Traditional Statisticians with Minimal Understanding of Engineering

One of the major problems that many companies in several industries started with, and some still are doing, is hiring AI experts and/or statisticians to apply this technology (Artificial Intelligence and Machine Learning) to their industry. About two years ago, in a Petro-Talk[1]I explained the main problems with this approach in the Petroleum Industry. Those who do not have engineering domain expertise usually use the data that is provided to them and try to find “Correlations” between the parameters that are shared with them. They do not pay much attention to finding out whether the data that is provided to them include all that is required to accomplish the objective that is being addressed.

[1]https://www.youtube.com/watch?app=desktop&v=Simc0Kd4sPY&list=PL1AUiFrtrjsQzYYBKXBhrl75Bq5SMBKPR&index=3&t=0s

Lack of domain expertise does not allow them to identify what data is actually needed in order to solve the problem that they are dealing with. Is it possible for a professor that has no knowledge about a specific engineering technology to teach engineering related courses to the students? In other words, can a political science professor teach thermodynamics? Or vice versa? As far as natural “Intelligence” is concerned, dealing with engineering related problems requires engineering domain expertise. Therefore, the same would be true when it comes to artificial “Intelligence”. A domain expert engineer that becomes an expert practitioner of AI and Machine Learning will be able to identify and use, or even generate the required features that can be used to “Teach” the AI & Machine Learning algorithms how to solve the specific engineering related problems.

Becoming an expert practitioner of AI and Machine Learning requires a detailed understanding of how AI & Machine Learning algorithms work beyond just the mathematics of these algorithms. The mathematics of the Machine Learning algorithms are not highly complex. Solving engineering related problems using AI & Machine Learning goes way beyond only mathematics and statistics. It requires an understanding of neuro-biology. It requires the knowledge of the philosophy of this technology and its differences from our traditional approaches to engineering problem-solving.

As mentioned in an article from Stanford University; “Artificial intelligence (AI) is the field devoted to building artificial animals … and, for many, artificial persons … Such goals immediately ensure that AI is a discipline of considerable interest to many philosophers, and this has been confirmed (e.g.) by the energetic attempt, on the part of numerous philosophers, to show that these goals are in fact un/attainable. On the constructive side, many of the core formalisms and techniques used in AI come out of, and are indeed still much used and refined in, philosophy[2]”.

[2]Artificial Intelligence; Stanford Encyclopedia of Philosophy. First published; Thu. Jul 12, 2018. https://plato.stanford.edu/entries/artificial-intelligence/

To cut the story short, using AI experts and/or statisticians that have minimal understanding of engineering domain expertise, to solve engineering related problems usually generates very poor results and sometimes, at best, will create highly mediocre outcomes. This has caused many managements of industrial companies to blame AI & Machine Learning technology rather than realizing that it was the incorrect decision to use the wrong individuals to guide their company in this area. The petroleum industry is one of the top industries that have been making such mistakes for years.

● Traditionalist Engineers

Traditionalist engineers refer to those that believe the only way to solve an engineering related problem is through the development of a mathematical formulation that describes the specific physical phenomenon. They believe that what they have learned, experienced, and have been dealing with for a long time is the only way to solve engineering related problems. To many of them, this specific, traditional approach to engineering problem-solving is more a religion than science. They keep saying that” AI & Machine Learning is a hoax”, or “I tried it, but it never works”.

Until about a decade ago, these types of engineers were the toughest problems against the application of AI & Machine Learning in engineering related problem solving. Some of them still exist and do all they can in order to annoy those engineers and scientists that are interested to learn and working with this new technology. Fortunately, given the positive behavior of the new generation of engineers and scientists, the current traditionalist engineers are no longer a problem. Many of them still try to use as much politics as they can, to make sure that the application of AI & Machine Learning in engineering related problem solving can be avoided. As time goes on, the number and the problems associated with these types of individuals keep going down.

● Engineers with Superficial Understanding of AI & Machine Learning

Today, the most important problem related to the application of AI & Machine Learning in engineering related problem solving has to do with some engineering domain experts. One of the main problems associated with the application of AI & Machine Learning by some engineers is when they try to achieve solutions without following the main characteristics of AI & Machine Learning. From their point of view, as long as some mathematics that incorporates machine learning algorithms is used to achieve their solution, then they feel comfortable calling it an AI-based approach. What they miss is to correctly answer this question: “Why would you want to use a Machine Learning Algorithm to achieve a solution?”

One answer is: “Because people (management of my company or my client) are interested in AI & Machine Learning, therefore, I can even use Machine Learning algorithms to perform “Linear Regression”. Well, this is true. It allows you to exercise Machine Learning algorithms to better learn how to use it. However, this particular approach has only an “academic” value that allows you to learn, but it does not have a “realistic” value because it does not follow the main reason “WHY Machine Learning algorithm must be used”.

Another answer is: “I tried to use actual data (actual measurements), but it did not work. Maybe I did not have enough data, or maybe the actual data was too uncertain and had too much noise and maybe it was too complex to achieve a solution. Therefore, I decided to generate data from the equations and combine it with the real data, then it significantly increased the possibility of success for pattern recognition”.

Well, it makes sense, because data that is generated by mathematical equations, is based on existing correlations and therefore it guarantees that patterns from this data can be recognized by the Machine Learning algorithms. Again, this has an “academic” value that allows you to learn how AI & Machine Learning algorithms work, but it does not have a “realistic” value because it does not follow the main reason “WHY Machine Learning algorithm must be used”.

Let’s be clear; trying multiple approaches to achieve reasonable solutions is the right thing to do. However, it is important to be able to correctly judge the essence of the approach that is being used and it is extremely important to pay attention to the scientific reasoning of “WHY Machine Learning algorithm must be used”. Unfortunately, in many cases, the judgment seems to be based on marketing and business success rather than scientific correctness. Under such circumstances, the lack of realistic and scientific success of using AI & Machine Learning in engineering related problem solving is clarified.

It seems that those engineering domain experts that end up doing what has been explained above and are also being called “Hybrid Models” have a superficial and limited understanding of AI & Machine Learning. In general, this specific problem associated with engineering domain experts can be divided into two categories:

1. Those that try to combine data generated by mathematical equations with actual measurements when trying to solve engineering related problems using AI & Machine Learning. This is called the “Hybrid Model”, and
2. Those that try to use specific machine learning algorithms that have been developed to solve non-engineering related problems to solve engineering related problems such as Convolutional Neural Network (CNN that was used to solve the non-engineering related problem such as “image recognition”) and Long-Short Term Memory (LSTM that was used to solve the non-engineering related problem such as “caption generation” for images).

Conclusions

Engineering application of Artificial Intelligence and Machine Learning is different from the non-engineering application of this technology. Engineering domain expertise is an absolute requirement to correctly solve such problems. Traditional Statistics and a combination of actual data with data that is generated through mathematical equations should not be used in the engineering application of AI & Machine Learning. While the future of engineering will be very much impacted by this new technology, currently many industries (specifically the petroleum industry) are going through a non-realistic version of this technology that has much to do with business and marketing rather than science and technology.