AI in manufacturing: What really matters now

Ms. Garg, you deal a lot with AI and the resulting changes in the manufacturing industry. I was wondering how AI has changed your daily work.

Arti Garg: I sometimes use AI to refine things, such as texts I have written or talking points. Sometimes it is also faster to verify facts using AI. But you have to be careful and always return to the primary source to verify it, as AI often makes mistakes. Nevertheless, I think AI can be quite helpful in this type of work.

In my position, I am no longer paid to write code, but that doesn't mean I don't enjoy doing it anymore. That's why I have experimented with some tools and applications for code development to try things out quickly.

It's great that you still find time to pursue your passion.

Garg: Not as much as I should, but yes, a little.

Regarding AI for businesses, there are many applications. In your opinion, what are the most important applications or use cases of AI in industrial manufacturing at the moment?

Garg: It has existed in the industry for a long time, but in my opinion, it is still one of the most important applications: the use of AI, data science, and machine learning models to detect anomalies and make predictions about industrial plants based on the collected data.

This allows companies to use the data generated by the plants and understand their behavior. This makes it possible to turn unplanned outages, which can be extremely disruptive, into something planned. If a plant needs to be shut down for maintenance, it can be done at the most convenient time. While this is not a brand-new AI application, it is still one of the most important.

If we fast forward to what is possible today, the capabilities of generative AI and large language models (LLMs) combined with agents open up many new possibilities.

This makes it easier for users to work in these very complex industrial environments, where there are often different types of plants. Some plants generate a lot of data. I like to call it reducing cognitive load. With the Aveva Connect Industrial AI Assistant, an operator can enter a question about the operations in a plant and quickly receive information without having to write code or call a specific AI interface.

Study: outdated systems cost industry millions

When automation becomes a brake: a new study quantifies the massive losses caused by outdated systems annually - and shows how open architectures can put companies back on the path to innovation.

Anja Ringel

Aveva pursues the vision of “connected industries of the future.” What does this look like in practice for a medium-sized company?

Garg: One of the solutions we are working on for our smaller and medium-sized enterprises is what we call “Industrial Accelerators.” These are pre-configured solutions within our Connect platform.

Connect is very flexible: on one hand, you can build something customized, which is often desired by very large companies. However, smaller organizations sometimes prefer things to be pre-configured. We develop these pre-configured solutions that allow for customization but reduce the startup time. Currently, we have an accelerator in the water industry and are exploring this approach in a number of other sectors.

Some companies I've spoken to have said they struggle to have the right data available to work with AI. What would be the first steps to get the right data?

Garg: Indeed, it is a challenge to implement AI if you do not have the right data. Many of these data are generated or captured through our platform. These data are crucial to be able to use AI.

However, sometimes the problem is that the data is available but not properly contextualized or accessible. To truly implement AI effectively, you need all the data elements. One thing we want to help with through our Connect platform is to bring these things together to create the right context. This way, AI can understand how new, potentially incoming operational data can be used to provide decision support and solutions.

In an interview with your colleague Markus Herrmann, he said, among other things, that the business case must be clarified before starting the implementation of new technology. That applies to AI as well, right?

Garg: Yes, that is always the case. In short: make sure you are solving a useful problem. I've probably been saying this sentence for over ten years when I talk about AI, and it is always true. It may be tempting to say, "I have data, let's see what I can do with it."

But it is really important that you solve a useful problem. When AI really works, it changes the way you do things, the processes, and the way you work - and that's good. However, you must be ready to make these changes. And in your organization, you will only be ready to make these changes if it solves a problem.

In the keynote at the Schneider Electric innovation summit, we learned that there is a 78 percent growth in organizations using AI. But it was also mentioned that there are challenges associated with it. What challenges does the manufacturing industry face?

Garg: One of the challenges is data readiness, which I have already talked about. However, when we talk about challenges that can arise after an organization has implemented AI, the issue of skills is important. Employees need to be ready to use AI. While we can build many safeguards and guardrails into AI solutions to prevent them from delivering inaccurate or faulty information, ultimately, the end users are responsible for using them correctly. It is equally important for end users to develop an understanding of where the limits and capabilities of AI lie.

This can feel intimidating because it is a new technology. But this is nothing new. For example, when you use an Excel spreadsheet, you know what it can and cannot do. If you enter an incorrect equation, you get an error message and know how to deal with it. It is precisely this comfort that needs to be built with AI. I think this is an important part of workforce training today.

Apart from AI, what are you working on right now?

Garg: That's a good question. At Aveva, we are also in the midst of transitioning to a hybrid and SaaS platform. This requires considering many other technologies that form the underlying infrastructure for SaaS. My team is working closely with the R&D team to identify the right solutions for building this modular, secure SaaS platform.

My focus is partly on understanding the transformations in the sectors we serve, which could drive the desire for new tools - many of which are AI-specific. To assess whether these tools could be used, we also need to understand the technologies that the respective sector already possesses.

I like to use the example of the power grid for this. It is becoming much more decentralized, the generation sources are getting smaller, and many are intermittent and renewable. This fundamentally changes what a grid operator must do to deliver power effectively and drives the need for more AI and decision support. To assess where this would be helpful, you need to understand these other technologies.

We are also looking far ahead and dealing with things like quantum computing and robotics. A lot is happening in the field of robotics right now. There may be a transition where the focus shifts from the mechanical aspects of robots to a much stronger focus on the AI brain in the robot. That is really interesting. It includes the AI part, the sensor part, the data integration, and the miniaturization of the system to a very small edge computing system.

Given all the technologies that are emerging and that we already have, cybersecurity is becoming increasingly important, isn't it?

Garg: Cybersecurity has always been important for our field. In recent years, especially in connection with the very powerful generative AI and foundation models, I have learned that AI often highlights existing cyber risks.

Issues that appear to be a risk posed by AI may have been due to data security policies not being correctly implemented. For example, a simple request to the AI to search all available data can quickly reveal that no permissions are required to access a SharePoint that should not actually be accessible.

AI can also help protect data from security risks.

Garg: It could be used for that. However, sometimes it inadvertently exposes security risks. I have seen organizations that have dealt intensively with risks and used AI to address a number of cyber risks that may not have been fully addressed in the past.

One last question: What advice would you give to CEOs who are still skeptical about how important AI can really be for their business model?

Garg: AI is already being integrated into everyone's daily life very quickly. Most people are used to seeing AI in their personal consumer devices - whether it's email or phone - and knowing what it can do. The speed at which this technology is being adopted and evolving is likely much higher than before.

It is therefore very useful to be aware of what AI is capable of and to know how to leverage its potential to achieve positive impacts on business.

An important advantage is AI's ability to train new workers and address the current workforce transition. A large part of the workforce will retire in the next few years and these employees have decades of knowledge.

The use of AI to capture this knowledge and make it accessible to the next generation of workers is of great benefit. For many companies, even smaller ones that may rely more heavily on one or two experts, this is a tremendous potential of AI.