Trustworthy generative AI improves decision-making processes for industry

MDU is leading a new collaborative project in industrial digitization that aims to transform decision-making processes in industrial sections, using generative AI. The project focuses on creating transparent and explainable AI systems that can suggest actionable measures with explanations to optimize maintenance schedules and minimize downtime in industrial equipment in production processes.

"Advanced data analytics and artificial intelligence (AI) are revolutionizing industrial processes. A key player in this transformation is “prescriptive analytics,” a technology that not only analyzes data but also provides clear, actionable recommendations to improve various decisions. In industrial settings, the effectiveness of these recommendations depends on the AI ​​models behind them", says Shahina Begum, professor of artificial intelligence at MDU.

One of the most exciting developments in AI is generative AI (gAI), which has the potential to dramatically reshape how industries work, says Shahina Begum. Generative AI refers to algorithms that can create new content — be it images, text, or even entire virtual environments — by learning patterns from existing data. This ability allows gAI to offer insights and innovative solutions in a way that was not previously possible.

But the complex, black-box nature of gAI makes it difficult to understand how it arrives at its recommendations. The lack of transparency can raise questions about the security of the decision, which is critical in industrial environments where trust and reliability are critical.

"This project aims to address these challenges by developing more transparent and explainable generative AI systems, making AI-driven decisions both powerful and reliable for industry", says Shahina Begum.

Actionable for industrial equipment

The main focus of the "Trust Gen Z" project is on using gAI for forecasts in connected industrial equipment. In complex operational environments, the generative AI will monitor, detect and predict the condition of equipment and machinery by analyzing different types of data, such as images, text and tables. The information is analyzed and evaluated, after which the gAI makes actionable recommendations with clear explanations.

"The generative AI model can simulate different possible future states of the equipment, which helps companies foresee potential problems before they arise", says Mobyen Uddin Ahmed, professor of artificial intelligence at MDU.

By understanding the uncertainty and risk in machine performance, companies can take proactive steps to optimize maintenance schedules, reduce downtime and increase operational efficiency. Right now, there is a lack of reliable AI methods that can both predict and provide reliable and explainable recommendations, according to Mobyen Uddin Ahmed.

"Although gAI seems promising, there are shortcomings in current explainable AI (XAI) methods to explain decisions by gAI, which limit their use for recommendations. This project will solve these problems and support digitization, initially in the automotive and telecom industries".

In the project, which is financed by Vinnova's "AI for Advanced Digitalization", MDU collaborates with Ericsson, Volvo Construction Equipment and MainlyAI.