The next industrial revolution – Multi-agent systems and small Gen AI models are transforming factories

Factories are transforming and becoming smarter through the introduction of powerful multi-agent AI systems.

In this blog, we’ll take a close look at how these revolutionary AI-powered systems can help drive the factories of tomorrow. 

A lesson from history 

The industrial revolutions of the past can be described in two ways: firstly, as the emergence of new types of power. The transition from using humans and animals to using steam power in the 18th century was a significant revolution that enabled huge productivity gains as well as transportation innovations and urbanization. Secondly, the industrial revolutions marked the emergence of specialization: splitting up work into smaller tasks, with dedicated humans or machines for each part of the process. This enabled standardization and mass production. 

Coinciding with this, education and knowledge became specialized as well – people were only trained on their individual part of a process. Eventually, the innovation of machinery introduced automated reactivity to factory processes. Machines could now use condition-based “if this, then that” actions to complete a task. 

In today’s factories, we are seeing the emergence of innovative multi-agent AI systems, which reflect the above themes in many ways, while also exhibiting some differences. In this blog, we’ll take a closer look at some of these new developments. 

What are multi-agent AI systems? 

Multi-agent AI systems consist of autonomous agents or bots equipped with AI capabilities that work together to achieve a desired outcome. An agent in this context can be defined as  “an entity which acts on another entity’s behalf.” In these multi-agent systems, AI agents cooperate to achieve the goals of people who own certain processes and tasks. 

Multi-agent systems can be thought of as having five dimensions in terms of complexity when compared to a single agent system: 

1. Single to multi – adding more agents. 

2. Homogenous to differentiated – having fundamentally different roles between agents. 

3. Centralized to decentralized – removing the need for a single/central point of orchestration. 

4. Generic to specialized – adding in different backgrounds and knowledge to create different expert agents. 

5. Reactive to proactive – agents that can act independently in response to changes in the environment, without needing to be prompted. 

Are there parallels with the previous industrial revolutions that suggest agents might accelerate the next one? 

Let’s take the principles of multi-agent AI systems and apply them to a smart factory.  

• Each machine can have its own AI agent, while multiple machines or types of work can be managed by supervisor agents.  

• Most industrial tasks require multiple machines to work together, either in a streamlined, one-piece flow or in batches. Even machines working in “islands” need to be coordinated for the work in progress to be controlled, with no idle time. This requires that many different roles need to be assigned to different agents.   

• Adding a decentralized AI management layer can be very beneficial for a factory. There are many advantages to having sub-teams of agents with the ability to act independently of each other and run different areas of a factory to meet objectives.  

• Each machine works in a different way, and each area of a factory requires specialized knowledge. Therefore, each agent needs its own pertinent information to be able to act effectively. Higher levels of agent specialization would be very valuable to a smart factory. 

• Agents would benefit from autonomously determining when and how they need to act, rather than waiting for permission or being told when to do so. If agents were connected to the market, they could independently decide what to do. For example, an agent might exhibit this reasoning: “although the plan says that we have to produce this mix, I will change it because I think that there will be an increase in that particular product due to X and Y.”  

Multi-agent AI systems deliver clear improvements to factory processes and outcomes, including reduced downtime and increased optimization and efficiency. We also have the ability to add AI agents to data processing tasks, such as image and video analysis. This unlocks the potential of understanding input data in ways that were not possible before.  

Unlocking new ways of understanding data in smart factories 

In-line process control (IPC) is an approach that provides immediate feedback and adjustments based on real-time monitoring to maintain desired performance, quality, or output. If this is done well, it improves efficiency and reduces waste. However, the approach is difficult to implement, especially in systems based around humans. There are many data sources that need to be reviewed and understood in real time, and very experienced individuals tend to be the ones relied upon for this task. This experience is hard to acquire, potentially expensive, and still may not be sufficient to get the best results. This is, therefore, a great area of opportunity for multi-agent AI systems, which are very good at taking in lots of information, understanding what it means, and making real-time adjustments.  

Let’s look at two examples of how this works. First, let’s say you are making potato crisps, and need to understand how the cooking time of the chrisps differs depending upon the size and growing conditions of the potatoes. This can be a complex problem involving lots of disparate data sources that a multi-agent AI system could cope with well. The system could also help to determine the root cause of any problems that arise. 

A second example: if you are processing rubber in an extrusion line, the composition of the raw materials, their current mechanical and thermal characteristics, and the line parameters all influence the quality and speed of extrusion. This is a very complex problem, and in-line process control performed by an AI multi-agent system could add a lot of value. 

Another advantage of this application is that it can be integrated into factories of varying levels of infrastructure quality. Sensors may not be perfect, and information from outside the factory may have data quality issues, but removing even some of the problems will give great productivity and quality benefits. This can be especially true if costly manual inspections could be streamlined, alongside the more obvious benefits of reduced waste.

Multi-agent AI systems are revolutionary for factories 

We see parallels between the industrial revolutions of the past and what we are seeing today in multi-agent AI systems being adopted into factories. The difference now is that we are not transitioning power sources from people or animals to steam, or substituting humans in physical parts of processes. Instead, we’re allowing AI to perform tasks where it is beneficial to do so, and where it can perform the task better than the human. It is also worth bearing in mind that the real world is messy, and multi-agent AI systems can help us have more resilience and be more flexible.  

New innovations like real-time AI processing on edge can accelerate the next AI-powered industrial revolution, and give similar productivity benefits as seen in the first one. The edge component is critical, as it is more responsive than cloud, permitting real-time control. It also offers higher levels of data security, enables off-line operations (which are critical to factories), and significantly reduces the cost of the operation. 

However, AI will likely not be operating alone. I believe we will have human-AI hybrid systems for quite some time, and this is in no way a bad thing. It will be essential that humans and AI work effectively together – because for AI systems to bring value, they need to empower people, rather than replace them.  

This blog article was written in collaboration with Ramon Antelo (Capgemini Engineering)

About the Generative AI Lab 

We are the Generative AI Lab, expert partners that help you confidently visualize and pursue a better, sustainable, and trusted AI-enabled future. We do this by understanding, pre-empting, and harnessing emerging trends and technologies. Ultimately, making possible trustworthy and reliable AI that triggers your imagination, enhances your productivity, and increases your efficiency. We will support you with the business challenges you know about and the emerging ones you will need to know to succeed in the future.  

We have three key focus areas: multi-agent systems, small language models (SLM) and hybrid AI. We create blogs, like this one, points of view (POVs) and demos around these focus areas to start a conversation about how AI will impact us in the future. For more information on the AI Lab and more of the work we have done, visit this page: AI Lab.