More Than Meets The AI

The rapid emergence of innovative AI tools and models has made it possible for people to create in a multitude of new ways.

Don’t be fooled by the inherent fun in generative AI tools, underlying them is a real technology that will change the trajectory of process manufacturing.

By Claudio Fayad, Emerson

As the trend of generative artificial intelligence (AI) takes the world by storm, many of us are experimenting and having fun with it, exploring it, and creating unexpected results. While AI’s capabilities and hallucinations are quite amusing, and some may argue that they are in their infancy and need to mature to amount to anything people trust to navigate the complexities of the real industrial environment, it’s hard to deny that AI has the potential to thoroughly redefine the working world. This is particularly true in manufacturing, where the ability to process large amounts of data at a rapid pace is a stumbling block across nearly every industry. 

What we must not forget is that some of the world’s great innovations started as or were inspired by toys, such as PCs, which were preceded by gaming consoles. Consequently, the value of generative AI deserves a deeper look—one that explores the state of the technology, its current industrial uses, and the advancements and investments that give it the potential to grow into something truly transformative in the coming decades.

What is generative AI?

AI has evolved over the years to become a broad field of study with many technologies and solutions. Many people have differing opinions on what exactly constitutes AI, much less generative AI. One of the best modern definitions of generative AI is provided by Gartner: “Generative AI can learn from existing artifacts to generate new, realistic artifacts (at scale) that reflect the characteristics of the training data but don’t repeat it. It can produce a variety of novel content, such as images, video, music, speech, text, software code and product designs.” (Generative AI: What Is It, Tools, Models, Applications and Use Cases)

At the heart of this definition is the fact that generative AI is more than just a series of if/then logic. Generative AI does not just repeat, it creates. It is the difference between a system that can identify a picture of a dog amongst thousands of pictures of wolves and foxes, and a system that can draw an original picture of what a beagle might look like 300 years from now based on thousands of beagle photos spanning 100 years.

Why Generative AI?

Much of the use case for generative AI today is about fun—and for good reason as generative AI tools are quite enjoyable to use. On the most basic level, people are using generative AI tools to create unique content in ways that were not previously available to them. Non-artistic people find that they can use AI image generators to help create original representations of things they could previously only imagine. Generative AI is also contributing to new genres of music and entertainment with AI-generated instrumental sounds, music, and game development and testing tools. 

In addition, some of these fun tools provide value beyond play. AI camera technology in cell phones not only helps people take clearer and more vibrant pictures, it also offers them powerful editing features that were previously only available with complex, expensive software packages. People are also using generative AI tools to help them create fit-for-purpose professional headshots from their own collection of portraits, without ever needing to set foot in a studio. Perhaps most exciting is the explosion of copilots, helping people to be more productive and creative by exchanging ideas with a generative AI entity.

AI copilots can help operators on workstations in the control room, and out in the field using mobile devices. AI-generated image

A powerful process tool

The ability to generate specific artifacts based on previous training and prompts opens many different opportunities for AI to be used in real-world situations, in many cases applied to industrial plants around the globe. The process manufacturing world is the perfect place for the incubation of generative AI tools, as engineering plants, projects, processes, operations, and reliability issues are often complex, costly, and time-consuming. 

Today’s most advanced automation suppliers are incorporating generative AI into industrial plant design tools. This type of software relies on large datasets of previous plant designs and models to present an engineer with multiple options for preliminary plant designs, based on criteria provided using an intuitive interface. Such a solution aims first to eliminate the costly, time-intensive, repetitive early steps of plant design by relying on generative AI to complete those steps far more quickly, using a wider array of variables than a human engineer could handle. 

Generative AI and large language models (LLMs) can be used to streamline the path for control system configuration and upgrades. New software solutions can use generative AI to streamline the process of creating code for a new control system into a digital workflow. Specially trained LLMs can convert engineering documents, or a configuration of the existing control system, into a configuration file, fully documented. The software can automatically generate modern code with minimal or no errors—enabling earlier testing—along with faster, easier project completion. Moreover, the system learns with each new control-system implementation, adding content to its stored data and improving outcomes tailored to a user’s reality with each new project.

Though they are in their earliest stages, AI copilots, made intuitive through LLMs, are helping engineers more easily interact with control systems and control loops. Relying on deep data stores of successful automation strategies, such copilots can help engineers more quickly design successful and more fully tuned loops that provide increased performance and efficiency from the earliest stages of deployment.

Operators, too, are relying on early AI tools with alarm help solutions that recommend actions and provide decision support based on shared knowledge. With such solutions, experienced operators can share and store their input on what alarms mean and how to remedy them. Generative AI can put that information into natural language, contextualized for any situation.

Feedback defines the future

The current implementations of generative AI are exciting, but they are only the tip of the iceberg. New strategies and technologies are improving the technology every day, and those advances clearly illustrate the ways games and toys are becoming essential tools. On the most basic level, feedback loops continuously improve the quality of generative AI. 

Modern generative AI tools—like programming code generators, image generators, and even the plant design tools mentioned previously—use stored knowledge to provide a user with multiple options customized to their individual style. When the user accepts or rejects an option, AI increases its knowledge base, making it more likely to select an improved option in the future. This feedback is critical training for AI and, as it increases, so will the power of the technology.

Other, more advanced technologies are also improving the outlook for generative AI. Retrieval-augmented generation systems use external authoritative knowledgebases of specialized data as reference material for LLMs. Such a technology could prove to be a game-changer for industrial manufacturers because they could use general LLMs from their automation-solutions provider in tandem with proprietary authoritative knowledge bases. These technologies could then be used to design new process loops and control strategies—all customized to a manufacturer’s own unique operations, while simultaneously protecting their intellectual property.

LLM chaining is also opening exciting new opportunities for generative AI in the industrial sector. The technology empowers teams to connect multiple LLMs together for even more specialized processing. For example, a team could connect an LLM trained in using industrial data to generate operational technology insights to a different LLM trained to create verbose narratives. This combination could, in turn, feed into an LLM trained to translate into any language. This methodology would make it easy to create standardized, concise reports, or documentation and strategy guides, across world areas.

Fun drives function

There can be no denying that generative AI will have significant real-world implications. It also has the potential to create a paradigm shift in industrial process manufacturing. Consequently, the world is fortunate that there are so many generative AI games and toys available today.

Humans learn through play, and there is no better time to start learning about the AI technologies that may one day work alongside operators in the control room. Keep exploring and keep playing. Generative AI will keep learning. 

Claudio Fayad is Vice President of Technology of Emerson’s (St. Louis, emerson.com) Process Systems and Solutions business. He has held a variety of positions within the company. Initially based in Brazil, Fayad holds a bachelor’s degree in engineering from the Universidade Estadual de Campinas, an MBA from Fundação Dom Cabral, and a post-MBA from Kellogg School of Management.