Applying Agentic AI in Intelligent Asset Management and Maintenance of Large Industrial Systems.

-Distributed Agentic Intelligence Reform (DAIR) as a framework for Adaptive, Resilient Data-Decision Workflows augmented with AI.

by Henrik G?thberg & Mikael Klingvall, - Dairdux September 2024

(Adapted from Dairdux key note at EPFL Intelligent Maintenance Conference 2024)

?

Abstract

Agentic AI represents an emerging paradigm in artificial intelligence, characterized by autonomy, adaptability, and purpose-driven action. Trying to anticipate the implications of this trajectory. ?We argue that this agentic understanding will influence/impact all down-stream Applied AI research, development of AI-software and use of AI in real-world systems. IF these approaches are normalized/adopted in the AI-research community.

This text explores the significance and impact of Agentic AI in the domain of Cyber-Physical Asset Management and intelligent maintenance, where integrating human and AI decision-making is critical for proactive, safe and scalable system operations.

We discuss macro trends that make agentic reform not just important but urgent, and propose a framework for how agentic reform can lead to more resilient, adaptive systems and industrial eco-systems.

Finally, we outline Dairdux’s approach to agentic reform, centered on the Booster-Reacher heuristic, and explain why unlearning outdated models is essential for navigating the challenges of future AI-driven environments.

?

Introduction

Agentic AI has emerged as a pivotal concept in artificial intelligence research. Particularly in the realm of intelligent maintenance we see both great promise and huge risks of these technical approaches. Intelligent maintenance requires the ability to make decisions autonomously in different parts while ensuring resilience and adaptability within complex adaptive systems. (Digital Twins, Industry Metaverse/Multiverse, Social Machines, industry 5.0, Systems of Systems, Dark factories,? are all ideas and concepts that rest on rationales and arguments based on the same “agentic” ideas but framed in other applied research fields than AI-Computer Science.)

Our entry point based on practitioner reality and change readiness, is that the most viable path to use AI in Industrial System context will, for the foreseeable future, be about Augmented Intelligence with Human in the loop. This article seeks to answer two key questions: What is agentic AI, and how does it integrate with intelligent maintenance to achieve Augmented Intelligence? Given the current trends in AI and the volatile, uncertain world we operate in, the need for reform is more pressing than ever.

?

Defining Agentic AI

“Agentic” simply means behaves like an agent. Any specific agent-definition is framed by its environment it acts and co-exist in. For example.??A tiger as an individual will get one agent-definition and its agency if it’s in the wild. And another one in the Zoo.? To better encapsulate and understand the word agency which give the frame of any agent we lean on Daniel Pink’s work on motivation which focuses on Purpose, Mastery, and Autonomy.

In the context of AI, “agentic” refers to systems that perceive their environment, make decisions, and act autonomously to achieve specific goals across various contexts in constant feedback loops with the other agents in the framed environment. Other agents for example being the humans/teams prompting (requesting/framing the goals – in THIS case the decision support needed) or the other AI-agents in an Agentic AI workflow.

Unlike traditional AI, which focuses on narrow tasks, Agentic AI patterns more effectively adapts dynamically to new challenges and allow us to work on more complex problems and work at a higher abstraction level. Making it ideal for Intelligent Asset Management and Data-Decision workflows of intelligent maintenance systems that operate in complex, ever-changing environments.

?

The Need for Agentic Reform in Intelligent Maintenance?

Agentic reform is the alignment of human and artificial agency to create adaptive, safe, scalable AND proactively maintainable systems. This reform goes beyond simply adding AI to systems; it involves restructuring workflows, data management, and team architectures to ensure seamless collaboration between AI and human operators.

Paraphrasing ?Mary Parker Follett from 1925, The machine and the human workflows supported by the machine, and/or the human workflows of the machine are bound together. This principle underpins the core of agentic reform, which seeks to create cohesive human-AI systems that enhance decision-making and operational safety.

?

Macro Trends Driving the Urgency for Agentic Reform

Several macro trends are accelerating the need for agentic reform in intelligent maintenance:

Extreme Investments in General AI: Significant investments in AI are pushing the development of more autonomous, adaptable systems.

The Three Vs—Volume, Variety, and Velocity: The explosion of AI-driven use cases and AI-techniques, creates a massive opportunity/risk landscape to navigate in all different parts of an industrial system. Systems then need to be composable to adopt to innovation, with different innovation-to-obsolesce cycles for different parts, and dynamically adapt to growing complexity of operations.

Modern Software Practices: Agile development, Dev/Ops. domain-driven design, and modular/distributed patterns like Data Mesh or Federated Machine Learning are changing the way we think about how ANY system and technology are organized, built and deployed, offering more modularity, composability, flexibility and resilience.

Operating in a VUCA World: Volatility, Uncertainty, Complexity, and Ambiguity demand adaptive systems capable of responding to unforeseen events, as highlighted by Nassim Taleb’s concept of “Black Swan” events.

These trends underscore the necessity of moving away from rigid, monolithic systems and workflows toward modular, composable, adaptable solutions and workflows.

?

Implications for Intelligent Maintenance when adopting AI

The shift toward agentic reform has several key implications for the future of intelligent maintenance:

?Adaptability over Efficiency: In a fast-evolving landscape, adaptability is more critical than squeezing out marginal gains in efficiency.

Modularity over Monoliths: Modular systems allow for containerization of uncertainty, flexibility and partial replacement and innovation. While monolithic systems are prone to breakdowns in the face of rapid change.

Convergence of Human and System Decision-Making: AI systems must work in tandem with human decision-makers, enabling higher-level decision-making while handling routine tasks autonomously.

The convergence of AI and human decision-making highlights the growing importance of addressing the whole ?sociotechnical context of data-decision-action workflows in our designs, Where human-AI collaboration and alignment is essential for safe and scalable operations.

?

Dairdux’s Core Beliefs and Agentic Reform Principles?

At Dairdux, we believe that agentic reform is the key to creating resilient, adaptable systems. Our approach is summarized by the Booster-Reacher Heuristic, which states that every agent—whether human or artificial—should enhance the performance of other agents within the system. This heuristic is supported by three core principles:?

Team as Agent: Teams, not individuals, are the primary operational units, the pivotal unit of change and innovation, working with autonomy, mastery and purpose. (high-cohesion with a manageable team cognitive load for relevant execution and decision making)

Agents Boost Reachers: Each agent in the system should serve to boost another, ensuring interdependence and cohesion of the eco-system.

Product as Interface: Clear interfaces between teams, systems, and products foster modularity, composability and resilience. (low coupling between agents for dynamic adaptability)

These principles enable organizations to maintain adaptability while managing complexity and cognitive load.

?

What Needs to Change: Unlearning Outdated Models

?To embrace agentic reform, organizations must leave behind outdated models of management and operation:

?From Static Optimization to Dynamic Adaptation: Traditional efficiency goals, objectives and models are no longer sufficient in a VUCA world. Decision-Making, systems and processes need to adapt continuously to survive and navigate unintended consequences.

From Division of Labor to Cross-Functional Collaboration as key concern to enable agentic org/work design: Silos inhibit innovation. Cross-functional teams that collaborate effectively are better equipped to handle complex problems.

From Economies of Scale to Economies of Learning: Success now depends on how quickly an organization can learn and adapt, not merely scale.

From Command-and-Control to Agentic intelligence and decision-making: Decision-making must be distributed to the lowest possible level, empowering teams to act autonomously while aligned with broader objectives. Furthermore. Feedback loops needs to be done in data and code (structural capital) and can not be relying on human capital alone.

?By unlearning these old paradigms, organizations can build more resilient, adaptable systems that are prepared for the complexities of modern AI-driven maintenance.

?

Heuristics for Distributed Agentic Intelligence Reform

Heuristics serve as guidance in decision-making on design in agentic systems. In super complex environments with a need for alignment on agentic shaping from algorithms to org. structures. Heuristics is as good as it gets for common understanding and discussion across disciplines and system and organizational layers.

Dairdux emphasizes the importance of feedback loops and modularity to ensure both autonomy and interdependence within teams. These heuristics should apply at multiple levels, from individual algorithms to organizational processes. Som further refections underpinning the Booster-Reacher Heuristic:

Self-Stabilizing Systems: Systems should self-correct based on real-time data, minimizing the risk of cascading failures.

Balancing Autonomy and Interdependence: Teams should be empowered to make decisions independently while staying connected to the broader organizational framework.

Cohesion and Resilience: Internal cohesion within teams and modularity between teams ensure resilience in the face of disruption.

?Heuristics like the Booster-Reacher Heuristic guide organizations in fostering and shaping distributed agentic intelligence and modular system architectures.

?

Conclusion: The Path Forward

Agentic reform is essential for the future of intelligent maintenance. As AI-driven decision-making becomes increasingly integral to operations, organizations must adopt flexible, adaptive structures that allow human and artificial agents to collaborate seamlessly. The principles of agentic reform—autonomy, modularity, and interdependence—offer a framework for building resilient systems that can navigate the complexities of a VUCA world.

?

In the words of W. Edwards Deming, “It is not necessary to change. Survival is not mandatory.” In the context of agentic AI and intelligent maintenance, reform is not just an option—it is a necessity for future success.

?

References

Pink, Daniel. (2009). Drive: The Surprising Truth About What Motivates Us.

Drucker, Peter. (2014). Innovation and Entrepreneurship.

Follett, Mary Parker. (1925). Creative Experience.

Taleb, Nassim Nicholas. (2007). The Black Swan: The Impact of the Highly Improbable.

Deming, W. Edwards. (1986). Out of the Crisis.

Grove, Andrew. (1996). Only the Paranoid Survive.

Toffler, Alvin. (1970). Future Shock.

Collins, Jim. (2001). Good to Great.

?

?