The next era of generative AI in industry

Exploring developments in generative industrial AI and their implications for industry.

AI has delivered tremendous value for industry over the past few years. Innovations in machine learning and neural networks enabled solutions like predictive maintenance or generative design. However, with the recent breakthrough in generative AI new opportunities emerged, which — beyond all the hype and excitement — are delivering real value to industries.

With industrial copilots, domain know-how in industries plagued by a lack of skilled labour is being democratised and AI-powered human–machine interfaces are making technology more accessible. Additionally, large language models (LLMs) are playing a transformative role in industrial applications by acting as ‘translators’ between APIs. This capability simplifies communication and integration across diverse tools and systems, breaking down silos and enabling more efficient workflows.

At the same time, there are many obstacles that need to be overcome. For example, the data hunger of LLMs cannot be satisfied as long as data sovereignty and trust between partners remain unresolved. Developments are therefore fragile and difficult to predict.

The evolution of industrial generative AI: key trends

Industrial foundation models

Industrial foundation models are pre-trained on industry-specific data, enabling faster and more accurate deployment of AI solutions. They provide a standardised starting point, saving time, resources and energy due to scaling effects. The models capture industry complexities, leading to informed decision-making. They also facilitate knowledge transfer and collaboration across sectors.

What this means is that an industrial foundation model is trained on the ‘language of engineering’ and goes beyond typical LLMs. It supports not only text and images but also 3D models, 2D drawings and other complex structures like time-series data specific to industry.

Tailoring AI solutions to industry needs improves efficiency and performance. However, challenges include data availability and privacy concerns. Industrial partners need to team up and share use cases and their specific training data — so trust is essential here.

Agentic AI

Agentic AI, especially in combination with generative AI, will be a game changer. Agentic AI refers to the use of AI systems that possess a certain level of autonomy and decision-making capabilities in the industrial context. These AI systems are designed to act as agents, performing tasks and making decisions on behalf of human operators or organisations.

Agentic AI in the industrial context involves deploying AI systems that can independently monitor, analyse and control various aspects of industrial operations, such as predictive maintenance, quality control, inventory management or optimisation of production processes. By leveraging agentic AI, industries can benefit from increased efficiency, reduced costs, improved safety and enhanced decision-making capabilities.

Generative AI combined with agents will change the way we engineer and design. AI will take over more tasks that were previously done by humans, such as designing a specific element or product. With agents, we will first simulate and optimise the process, then derive the parameters for the factory or plant, and then design the system, which will then be built.

A common marketplace (an Agentic Economy) will form where agents can be ordered or bought. Agents will also need to be able to search for other agents in such stores, so a standard for agents and their meta-description may become important. Agents will also have the ability to pay for third-party services. At some point, you will have a network of agents, orchestrated and controlled by a ‘master agent’.

End-to-end quality management across multiple agents is one of the biggest challenges, however. This is especially true when an agent economy is established that combines a wide variety of agents from different companies.

Multimodal LLMs

Multimodal LLMs represent an exciting trend in AI, enhancing language processing, improving computer vision and enabling machines to better interact with the real world. They combine language understanding with visual perception. These models process data from text, images and videos to gain a deeper understanding. They can generate accurate descriptions, analyse visual data with textual information and enhance natural language understanding.

Multimodal LLMs have applications in computer vision, autonomous vehicles and robotics. They improve object recognition and scene understanding, and enable machines to follow complex instructions. Multimodal LLMs have the potential to revolutionise the processing and generation of industry-specific data such as time series, 2D and 3D models or machine vision in the same way that conventional LLMs have revolutionised text and speech processing.

Challenges include acquiring and labelling multimodal datasets and addressing biases. A good approach would be to pre-train multimodal LLMs to the extent that the local effort required by the end user remains low. This increases acceptance and safety and enables faster scaling.

Edge models

Industrial edge architectures represent a growing trend in the field of AI, and involve the deployment of AI algorithms and processing power at the edge of industrial networks — in closer proximity to the data source. This approach facilitates real-time data analysis, decision-making and automation, thereby reducing reliance on cloud-based processing. It addresses latency and bandwidth limitations, enabling industrial systems to respond promptly to critical events and make timely decisions.

This approach also enhances trustworthiness and cybersecurity by keeping sensitive data and AI algorithms within the local network. Additionally, it offers cost savings by reducing the need for continuous cloud connectivity and associated data transfer costs.

However, the decision to adopt industrial edge AI depends on factors such as algorithm complexity, data volume and the need for centralised monitoring and control. Achieving a balance between edge and cloud computing is crucial for ensuring optimal performance and scalability.

Specialised hardware

Specialised hardware — like GPU-enabled or language processing unit (LPU)-enabled edge devices — is an emerging trend in industrial AI. These devices provide high-performance computing power at the edge, enabling real-time processing of AI algorithms. GPU and LPU integration allow for parallel processing and accelerated performance, resulting in faster execution of complex AI tasks. This local processing reduces latency and reliance on cloud resources, making it crucial for time-sensitive applications.

Specialised hardware also supports more advanced AI models, leading to enhanced insights and improved performance. However, careful consideration is needed for factors like power consumption, scalability, and compatibility with existing systems to ensure seamless integration and optimal performance.

The impact: future scenarios

Scenario 1: From assistance systems to autonomous systems

In the future, industrial AI will experience a significant transformation, shifting from assistance and recommendation systems to an increasing number of more autonomous systems in the realm of industrial automation. This trend is expected to extend to the automation domain in the next five years, as experts foresee the need for more autonomy in factories due to a lack of skilled labour. Advanced task-planning systems are likely to play a crucial role in this transformation.

In manufacturing, autonomous robots will have taken over repetitive and labour-intensive tasks on the assembly line. These robots work collaboratively with humans, seamlessly integrating into production processes. Equipped with advanced AI algorithms, they can analyse data in real time, make decisions and adapt to changing production demands. This level of automation could drastically improve efficiency, productivity and product quality.

The logistics sector has already witnessed a revolution through the implementation of autonomous systems. Smart robots navigate autonomously through warehouses, retrieving and delivering goods with precision. They optimise inventory placement, track stock levels and streamline order fulfilment processes. As a result, order processing has become faster, errors have been reduced and customer satisfaction has improved.

In industries such as energy and utilities, autonomous systems play a crucial role in monitoring and controlling complex infrastructure. AI-powered systems continuously analyse data from sensors, predict maintenance needs and can automatically schedule repairs or adjustments. This proactive approach to maintenance ensures the smooth operation of critical systems and minimises downtime.

Besides providing support at individual steps in existing workflows, next-generation industrial generative AI solutions are expected to enable shortcuts in the workflow process, spanning multiple individual tasks, from intent to outcome. This advancement will further enhance productivity and efficiency in industrial processes.

The rise of autonomous systems in industrial automation has led to increased productivity, cost savings and improved safety. These advancements bring us closer to the progress levels projected by economists for a world filled with autonomous systems, compensating for the limitations of human labour.

However, it is crucial to also address the challenges of potential job displacement and ethical considerations that arise from this transition. It must be clear that AI is not intended to replace humans, but to support them. Less challenging, less complex tasks can be handled by AI, leaving humans to focus on tasks that require deeper understanding and creativity. Careful planning and management are necessary to ensure a smooth integration of autonomous systems while addressing these concerns.

Scenario 2: Autonomous industrial AI for flexible, customised and high-speed processes

Autonomous industrial AI has the potential to transform production processes, outperforming human intervention in terms of speed and flexibility. These intelligent systems can analyse vast amounts of data in real time, enabling split-second decisions to optimise operations.

However, achieving this level of automation requires trust in AI’s performance. Manufacturers must have confidence in the accuracy and reliability of AI algorithms to delegate control. By placing trust in AI, businesses can unlock the full potential of speed, flexibility and efficiency.

As AI capabilities evolve beyond our current understanding and experience, the importance of human involvement will diminish, with the extent of autonomy granted to AI systems determining the possibilities. The key question is whether we will limit what is achievable, or whether we will embrace the realm beyond our current understanding.

Mastering the new era of generative AI: a holistic strategy

To ensure readiness for the advancements and challenges of industrial AI in 2030, it is essential that stakeholders adopt a comprehensive strategic approach, incorporating a number of key elements.

General considerations include:

• Fostering a culture of innovation within the organisation that embraces AI technology: This can be achieved by promoting collaboration, providing training and resources for employees to upskill in AI-related areas, and fostering an environment that encourages experimentation and learning from failures.
• Exploring which processes can benefit most from the power of AI: Increasing efficiency, reducing material and energy consumption and shortening lead times are also key to achieving sustainability goals.
• Remembering that industrial AI differs significantly from end-user AI: It must meet the rigorous requirements and standards of industrial environments, with cybersecurity, harm reduction, legal compliance and the mitigation of bias in training data ensured through binding standards and rules.
   

To fully take advantage of the future of industrial AI, organisations must enable an industrial ecosystem-centric approach. What is true for digitalisation is all the more true for AI: no one can do it alone. True creativity and value creation can only happen in a space where partners, customers, vendors, developers and experts come together to explore and develop best-in-class solutions.

We must remember that industrial AI can only be as good as its training data, and that it benefits from collaboration and co-creation. The old ways of protecting data to realise your interests (such as intellectual property) will change in the long run — in the emerging age of AI, those who share their data with partners, customers and experts in the best way will be the ones that succeed.

Image credit: iStock.com/Just_Super