Upskilling the Workforce in Trained AI Expertise with General Motors' Jeff Abell

In the fifth of our series of A3 interviews with AI leaders, Jeff Abell, Chief Scientist for Global Manufacturing and Director of Materials & Manufacturing Systems Research at General Motors, addresses improving education in the workforce around industrial AI. Abell, who sits on A3's Artificial Intelligence Technology Strategy Board, recommends a variety of strategies based on foundational training and initial disciplines. Check out his advice on how to upskill existing employees on industrial AI and get started on AI projects. 

Jeff Abell

Jeff Abell
Chief Scientist for Global Manufacturing and Director of Materials and Manufacturing Systems Research
General Motors

How are companies successfully addressing the lack of trained AI expertise in the workforce?

The lack of trained AI expertise in the workforce has been a common challenge that most companies are currently facing. We have groups of talented domain area experts and groups of knowledgeable AI developers and data scientists. The real challenge here is how do we bridge the gap and synergize the expertise between AI experts and domain experts. There are several viable options and strategies that companies can adopt to form a multi-faceted approach and address this issue. Ultimately, the gap can be narrowed or even eliminated by establishing training and development programs to expose technical staff to situations and roles where they can either gain domain expertise or enhance AI-related skills and abilities. Given the rapid pace by which AI tools are being developed and made available to people with wider and increasingly diverse levels of mathematical and computer science training, the challenge is not so much in acquiring great depth in such disciplines, but rather in understanding where, when, and how these technologies can be best deployed for the problem at hand. In this sense, staff with strong domain expertise need to first attain a solid, broad foundation in AI, but then focus on searching out the training that is most applicable to their area of expertise. For example, a manufacturing quality engineer working on inspection systems may need to fine tune their knowledge of deep learning as it relates to advanced vision systems. 

For staff possessing a strong background with perhaps even graduate-level skills and abilities in AI technologies, the challenge of how to close the gap is somewhat more situation dependent. When initially faced with a technical problem that demands deep knowledge of science or engineering, the staff person cannot unilaterally nor blindly develop an AI solution without first getting familiar with the basic manufacturing “physics” of a given process or operation. There are a number of strategies that can be taken ranging from seeking out foundational training in relevant the discipline(s), retaining services of experts in the field, or what is often very effective, creating multi-disciplinary teams through which collaboration with domain experts can offset any deficit in a given subject matter. In the end, it is the continual and targeted education of both types of technical experts supplemented by frequent and close collaboration that ultimately can yield the most feasible and sustainable approach to ensuring an overall workforce that is literate not only in AI technologies but in how to best deploy them to difficult and challenging industrial problems.

How would you advise companies to choose their artificial intelligence projects – and what questions do they need to answer before they begin?

A few key factors come to mind when we start talking about what to consider before beginning any artificial intelligence projects. If we consider the efficient frontier, it's important for companies to carefully assess achievable performance and likely risks associated with each potential AI project to understand the trade-offs and choose the projects that align with your company's risk-reward tolerance. This is similar to a stock-portfolio and efficient frontier analysis in optimization.

Another related consideration is to completely understand the impacts of potential AI projects. AI projects can have diverse applications, from automating routine tasks, improving decision-making, enhancing customer experiences, optimizing operations, to enabling new business models. However, AI is not “one size fits all” solution and may not always be the best approach. It is crucial to understand the potential impact of AI projects on company's goals and objectives to ensure that it aligns with business strategy and creates opportunities. For example, do we see any headroom for further improvement if AI is introduced as compared to the expected effort? If the answer is yes, then we need to further assess each AI project in terms of the value it can bring to our company versus the investment that has to be made. 

After estimating the impact of a potential AI project, the next question to answer concerns AI readiness, which includes the readiness of data and infrastructure to support the AI solution. Data is a critical component of AI projects, and the readiness of your data is an important consideration to ensure that the AI solution can be developed, launched, and sustained. This requires the assessment of the quality, quantity, and accessibility of your data for each potential AI project. High-quality, relevant, and well-curated data is essential for training and validating AI models. The data collection, storage, and processing capabilities of your organization need to be evaluated to ensure that the necessary data infrastructure and resources are in place to support the AI projects being considered on an ongoing basis. Moreover, AI data readiness is not all about whether there is enough good data, it is also about whether sufficient “bad data” is present to represent all the scenarios and situations that the AI model and solution must comprehend. Data infrastructure in manufacturing applications is particularly critical owing to the variety of plant-floor systems and equipment data acquisition points that need to be aggregated in a cohesive manner to be transferred over and across networks, databases, servers, and applications into the final AI solution. The connectivity, pathways, and robustness of this infrastructure needs to be able to support the AI solution requirements else the underlying validity, functioning, and accuracy of the model(s) can be put at risk.

An interesting, but related, question companies should raise is the possibility of selecting a suboptimal strategy to accommodate a future AI solution. It may be optimal in a short-sighted manner to manually document activities of a process like maintenance, record keeping and so on. After all, it can be cost-effective and require less resources to plot a hand drawn chart and hang it on a board. We don’t foresee using such information beyond immediate production needs. So, we don’t have the need to curate such information long term. However, if we would think beyond today, we could see that all production information and actions taken have lessons that we can learn. When a cohort of such information is assembled, there are trends and information to be derived. To achieve such intelligence, investments in digital record keeping and subsequent visualizations are needed. Manual plots on boards need to be replaced by displays and dashboards with greater functionality. Marking with a pen would have to be replaced by digital data entry that require software and hardware and the necessary maintenance that comes with such added infrastructure. This investment does not improve charting or communicating information that would impact the immediate production. If we can make this investment, we would enable AI that could provide us more future value on the investment today.

What AI application for industry are you most excited about and why?

While AI is not necessarily that new in manufacturing, it was not very prevalent. The most fundamental AI has been implemented as traditional statistical process control techniques through computer-based feedback mechanism. Today, we are able to make very complex decisions from very complex and pseudo-stationary processes. The paradigm of big data and AI has given rise to a new philosophy that we refer to as “process monitoring for quality,” a technique which has been successfully applied in several quality prediction scenarios including welding applications. These methods not only enable quality prediction, but pave the way for extracting in-depth information about the process itself. This works to provide information regarding process stability as well as root-cause information. We envision these techniques being used in a wide variety of manufacturing processes in the future. Beyond the use of AI on digitally acquired and curated data, we are also interested in looking at large language models for natural language processing. Plant operators reports process concerns, corrections, and maintenance events as verbatims can provide a tremendous amount of insights about how the plant floor is operating. The ability to parse such information to generate actionable items to improve production yield and quality could be a significant benefit to the automotive industry. 

Dr. Jeffrey Abell is Chief Scientist for Global Manufacturing and Director of Materials & Manufacturing Systems Research at General Motors. He is responsible for global manufacturing research focused on vehicle electrification, lightweight systems manufacturing, automation, and smart manufacturing. He led a research team that played a key role in bringing the Chevy Volt advanced battery to production and has held other leadership positions in product development and manufacturing at GM, Delphi, and DaimlerChrysler, including two international assignments. He has written numerous technical publications, been granted several patents, and has a strong track record of successful manufacturing technology implementation. He was twice presented with the Boss Kettering award, GM’s highest recognition for technical innovation. Jeff has a bachelor’s degree in Mechanical Engineering from General Motors Institute (now Kettering University), and graduate degrees in Systems Engineering from Oakland University, and is a licensed Professional Engineer.