The limits of Gen AI in manufacturing

Artificial Intelligence (AI) has revolutionized industries worldwide, and manufacturing is no exception. You’ve surely seen the hype: every industrial vendor has rushed to claim that their solution is “AI-Powered” in order to ride the wave of interest in the technology.

The recent AI hype has been mostly driven by the launch of consumer AI products that represent incredibly powerful breakthroughs in AI, and have wide applications for a variety of industries. Chat GPT is one of these well-known products. Generative AI tools like Chat GPT are commonly what springs to mind when people think of “AI” these days.

“AI” is a compelling buzzword for customers and investors alike. By integrating custom GPTs, software companies can market their industrial software as “AI-powered” or “next-gen,” even if the AI doesn’t meaningfully improve performance. This helps the software company differentiate from competitors, even if the added features don’t provide much value.

In manufacturing, GPTs and Gen AI are just a small part of a broader AI-driven transformation. The most valuable uses of AI in this sector extend far beyond the capabilities of GPTs and Generative AI. These applications of AI for manufacturing can have drastic impacts on quality management, supply chain management, process design and control, operational efficiency, and cost control.

First, let’s understand what Chat GPT and Generative AI are, and explore their strengths and weaknesses for manufacturing applications.

What is Generative AI?

Generative AI is a type of AI that generates original outputs. A tool like Chat GPT (Generative Pre-trained Transformer) is a type of AI model designed to generate human-like text. It can handle and understand large sequences of text, such as paragraphs or even entire documents.

Generative AI is just one type of AI. A tool like Chat GPT is actually a combination of different AI techniques expertly woven together, including:

• Generative AI: A type of AI that generates original outputs, in this case text-based responses
• Large Language Models (LLMs): Chat GPT contains a large-scale language model that is trained on vast amount of data to help it predict and generates text
• Deep Learning: The use of neural networks (transformers) to learn patterns in language and produce coherent responses
• Natural Language Processing: Application of techniques to understand, process, and generate human language, including grammar and semantics
• Reinforcement Learning from Human Feedback: Improves output using human feedback during fine-tuning for more relevant responses
• Self-Attention (Transformers): Focuses on different parts of text to handle context and dependencies
• Contextual Embeddings: Understanding words in context, which enhances the accuracy and relevance of responses

As you have probably experienced first-hand, a Generative AI tools like GPT’s strength lies in its ability to understand and respond to regular human language. This features gives it an incredible “wow” factor, as it appears to the user that Chat GPT is responding to us like a fellow human would. The same impact can be felt when we interact with humanoid robots, compared to other forms of robotics that look and move more like “machines”.

This “wow” factor is interesting, but misleading. It can give the impression that Chat GPT is much more intelligence and capable than it actually is.

The limits of GPTs and Gen AI in manufacturing

Many software companies have rushed to incorporate AI into their existing platforms. One of the fastest and easiest ways to do this is through adding a custom GPT to their tool. Unfortunately, these custom GPTs often don’t vastly improve the abilities of the existing software they are tacked onto. Here are three key limitations of the use of GPTs in manufacturing software:

1. GPTs lack industrial expertise

While GPT models are powerful for general text generation, they are not inherently expert in all topics, such as Industrial data processing and analysis. To provide value to industrial applications such as predictive maintenance, complex machinery diagnostics, or real-time control systems, GPTs need carefully curated data, specifically designed guardrails, and continuous evaluations to be effective.

2. GPTs are poor at analyzing manufacturing data

GPT models are not designed to analyze technical data like sensor readings, equipment logs, or environmental variables. They are poor at numerical inference on time-series and high-frequency data natively. Their outputs may seem plausible but could miss key insights, leading to flawed conclusions.

3. GPTs can “hallucinate”

GPTs are known to “hallucinate,” or generate convincing but incorrect information, particularly when provided with a lack of context or when being asked to provide information it is not equipped to do – such as process large amounts of numerical data. In industrial settings, this could result in incorrect diagnostics, inaccurate reports, or misleading operational suggestions, which could have severe consequences, such as equipment failure or safety risks.

When deploying GPTs in Industrial settings, the application matters. Uploading raw data to ChatGPT and asking for an analysis is unlikely to provide great results. But an LLM that is asked to synthesize and

The value of Gen AI in manufacturing

The goal of this article is to provide healthy skepticism about AI in the form of GPT in your manufacturing software. In manufacturing operations, there are some benefits that generative AI can add to an already powerful tool, such as:

• Making the platform simpler by allowing the user to perform functions or ask questions by typing in a request in plain language
• Translating analysis generated in the platform into plain language
• Automating documentation
• Improving platform search or “help” functions
• Augmenting training and onboarding

When choosing an AI software, do your research. Ask questions. If the tool you are evaluating has recently added AI, ask if it was a GPT or something more deeply embedded into the core function of the platform.

Generative AI for manufacturing product design

Outside of manufacturing operations, Generative AI has some very effective applications, particularly in the realms of design, prototyping, and innovation. According to an article by McKinsey & Company, here are three areas that Generative AI can assist during the product design phase:

From the same article, the following case study was outlined:

“Industrial designers at an automotive OEM needed just two hours with the help of gen AI to create the initial design concepts for 25 variations of a next-gen car dashboard with a touch screen interface, charging surfaces, instrument panel, and other components. These concepts were then further refined by the design team using an image-editing software to provide stakeholders with a clearer picture of where the industry was going and how to evolve component interfaces, form factor, color, material, finish, and more for the latest models of electric vehicles (see images below). Without gen AI, creating images with similar detail and quality would have taken at least a week with many more iterations. This process empowered designers to bring a product experience to life in a far more tangible manner and in a fraction of the time.”

McKinsey & Company

The use of Generative AI in design is still emerging, and there are also ethical considerations about AI in product design to untangle before it becomes a widely adopted design method.

AI designed for Manufacturing operations

While companies may be adding custom GPTs into their industrial software to capitalize on AI hype, the real benefits of AI for manufacturing operations lie in specialized, domain-specific models that address the complexities of industrial environments.

The most impactful applications of AI in manufacturing tend to directly impact manufacturing operations, and utilize the vast amount of manufacturing data available from modern factories. AI tools that make a big impact on production are not LLMs or Generative tools. They instead use machine learning, anomaly detection, predictive analytics, computer vision, or optimization algorithms. They include:

• Predictive Quality
• Vision Inspection
• Augmented Reality Inspection
• Smart energy management
• Production planning and scheduling
• Self-guiding and self-adjusting robots
• Advanced Process Control
• Predictive Maintenance

What differs in these applications of AI from Generative AI is that they contain algorithms that were purpose-built to complete specific tasks right in the production environment.

It is critical to understand how any tool works before integrating it in your manufacturing environment. When all levels of an organization understand what AI is, the various types of AI, and what they can actually do, they can make informed investments that will positively impact their business.

Know your manufacturing AI

As the AI ecosystem continues to evolve, manufacturers will see even greater benefits from its use, particularly as formerly separate systems become more interconnected and capable of learning autonomously. Generative AI will continue to develop and become a useful part of an integrated ecosystem, helping to retrieve and explain information on the shop floor.

It is critical for manufacturers to educate themselves on applications of AI for manufacturing, not only to select effective tools to use today, but to keep up in the future… because this is just the beginning of the changes that AI will bring.