The biggest challenge is not about technology in AI, it's about talent.

Dr. Jay Lee
Director, Industrial AI Center at the University of Maryland

In a recent episode of the Augmented Ops podcast, we took a deep dive into applied AI and the role of education with Dr. Jay Lee, Director of the Industrial AI Center at the University of Maryland. Titled "Rethinking Our Approach to AI," the discussion with Dr. Lee explores the value that machine learning and AI can bring to frontline operations, and how our educational programs must adapt in order to train the next generation of AI and ML engineers.

With extensive experience across industry, government, and academia, Dr. Lee explains how to apply machine learning tools to solve real world problems, and what is wrong with our current approach to AI/ML education. He highlights the need for new educational programs — like the one he is building at UMD’s Industrial AI Center — to train the volume of engineers necessary to meet the growing demand within the manufacturing industry.

The Foundations of Applied ML/AI

Dr. Lee lays out a framework for creating value with AI by applying it to solve real operational problems. This consists of three facets: technology, tools, and talent. First, technology serves as the foundation for an AI/ML solution. As the available machine learning technologies and techniques become more refined and new ones are developed, they can unlock new ways to address problems or make old methods more efficient. For example, the development of transformer models (first proposed in 2017) has allowed for significant performance improvements in Natural Language Processing (NLP) and other tasks.

Technology continues evolving, but we need to have tools to have a systematic way to do things… Then we need talent that can use the tool.

Dr. Jay Lee
Director, Industrial AI Center at the University of Maryland

Next, you need tools that allow people to leverage the technology and solve problems using that tech. For instance, although there had been a variety of technologies available for NLP tasks (such as OpenAI’s GPT-1, 2, and 3 transformer models), they did not see widespread adoption by the public. Only once OpenAI released ChatGPT — which provided an intuitive interface for interacting with those existing models — were they able to get this technology into the hands of the masses and enable them to use it at scale.

Finally, you need talent that is able to take advantage of those tools and use them to solve their problems. As Dr. Lee explains, “the biggest challenge is not about technology in AI, it's about talent.” For example, although ChatGPT is now widely accessible, it requires specific talents in the form of prompt engineering and understanding how the underlying model works in order to be able to use the tool most effectively. However, Dr. Lee argues our current educational paradigms are not set up to support the development of the talent necessary to meet the demand for ML/AI engineers.

Engineer at Mack Molding using the App Editor

Four Steps to Become an ML/AI Practitioner

Dr. Lee goes on to explain how anyone can be trained to become an effective ML/AI engineer, and lays out his vision for how a curriculum should be structured in order to accomplish this. His approach is based on four steps: principle-, practice-, project-, and professional-based learning.

Principle-based learning is the first and most basic step, in which students learn the core concepts of machine learning and AI. This can either be accomplished through self study, or by enrolling in a course that teaches these fundamentals.

Next is practice-based learning. As Dr. Lee explains, he provides his students with real-world datasets that have been collected across the industry over the course of many years, creating a realistic environment in which they can hone their skills. According to Dr. Lee, this phase is “the most important.” In this practice-based phase of study, students are tasked with applying the concepts they have previously learned to solve the problems presented in the data. Since the problem at hand has been previously solved, these datasets serve as a benchmark for student performance.

Then comes project-based learning. As Dr. Lee explains, this involves students going out into “a real manufacturing environment” and collecting data to build their own datasets against which they can test their abilities. Once students have demonstrated the ability to consistently solve real-world problems on the shop floor, using what they have learned, they can then progress to the final step.

Professional-based learning is the last stage in the journey of becoming an expert AI/ML engineer, and successfully navigating it indicates real mastery of the concepts and tools that students have been developing up until this point. This final phase involves guiding other students through the first three steps and acting as their teacher. “You have to be able to be like a [lean six sigma] master black belt,” explains Dr. Lee.


Educating ML/AI Engineers at Scale

However, in order to meet the rapidly growing demand of industry, Dr. Lee makes it clear that these educational efforts need to be carried out at a massive scale. “This is our goal: train 10,000 engineers in 10 years,” he explains. To accomplish this objective, Dr. Lee lays out three core pillars of his strategy — training engineers at scale, with speed, and systematically.

I'm not talking about training 200 people. I'm talking about 10,000 people, 100,000 people.

Dr. Jay Lee
Director, Industrial AI Center at the University of Maryland

For Dr. Lee, scale is critical. “You want to do big scale training,” he explains, “not one [person], it's 1,000, 10,000.” Without this kind of scale, we risk not having enough qualified candidates to meet the needs of industry, especially given the increased focus on domestic manufacturing. Equally important to the success of these efforts is the speed at which they operate. “You cannot just learn AI ad hoc,” Dr. Lee argues, “you have to develop AI, test it, implement it in two days.” This rapid learning model ensures that students are quickly brought up to speed with current technologies and methodologies, making them job-ready in much less time than traditionally required.

Conducting the training systematically is also key to the success of his strategy. Traditional ML/AI education is complicated by the fact that the models students use are not always in agreement. This means that one student may identify an issue in a certain situation, but another student using the exact same model may not. The result is that student performance is more difficult to assess accurately. Dr. Lee explains that “you got to have a consistent way to draw decisions” in order to be able to systematically assess how well students are doing. With these three principles, he believes that academia will be able to successfully train the volume of engineers that the manufacturing industry needs.

Rethinking Our Approach to AI

Check out the full podcast episode for further insights into Dr. Lee’s vision for the future of ML/AI education, and how these technologies can solve real problems on the shop floor.

Bottom CTA Global