AI in Manufacturing: Separating Hype From Reality

Manufacturing: The Unique Challenges of AI Implementation

In manufacturing it is being used for quality control and supply chain optimization. But manufacturing is a different beast, and AI is not necessarily the solution to every problem. In some industries, there’s a big difference between marketing to dealers and consumers.

As an example, a popular company that builds portable generators—the kind homeowners would use to guard against power outages—might want to consider their use of AI and really consider what the company wants to accomplish, because this company is mainly selling to dealers, not necessarily consumers.

For the most part, dealers already know what they want in terms of size and unit capacity. Since the dealer is simply buying the product for sale in his or her own shop, it most likely would not be necessary to use generative AI to figure out everything about that dealer to try and sell them more. Dealers usually have a set number of units needed, types, costs and specifications. They are probably not veering off that path.

The point is, AI is not a one-size-fits-all solution, especially in manufacturing, and for tasks like Quality Control or predictive maintenance, traditional machine learning models have often proven more efficient.

For example if a company is using extruder lines and it needs to create a way of rejecting pieces with too many scratches, traditional machine learning methods often prove to be more efficient. Using an AI Large Language Model for the same quality control issue means not being able to input domain-specific knowledge about the process, tolerances or defect types. You may fine-tune, but that fine tuning likely will not match the ease of integration the traditional approach would create.

Also, the final result would be an extremely large model with over one hundred billion parameters, causing increased costs and extreme inefficiency.

Of course Generative AI is certainly powerful, but it’s still at the beginning, and therefore it’s a little limited where manufacturing is concerned. It carries risks, ranging from getting information wrong to misinformation to losing control of itself.

The Future of AI in Manufacturing: Balancing Innovation and Efficiency

In manufacturing facilities, AI is being used to upgrade existing software and hardware. Robots that have been around for years used for sorting and packing items on production lines are getting AI upgrades to better identify and adapt to new products.

Quoting software that has been used since the late 1990s is getting AI upgrades to help create better queries on its existing platform. Sensors are now being used by large companies on machinery and equipment to help monitor vibrations, heat, humidity and other factors being reported for maintenance.

So for now, AI is mostly software upgrades for the manufacturing sector. It may be saving lives elsewhere, but in manufacturing, the older systems are generally found to be more efficient.

And that’s because AI is still really just getting started. We’re really only in the first stage of the LLM phase, in which models are being built, trained and tested. Today’s AI applications are built around an AI model that is already beginning to look old, which is a testament to how quickly things change in AI technology.

Final Thoughts

Even OpenAI’s CEO Sam Altman said on a tech podcast earlier this year that the current version of ChatGPT isn’t great compared to what’s on the horizon.

Sometimes people are expecting AI to solve problems and bring ideas to the table that no one else ever thought of. And it does do that for other industries. But manufacturing companies generally fall into the category of B2B, and AI won’t necessarily work miracles for them—yet.

‍