If you’re reading this, chances are you’ve heard about artificial intelligence (AI) and are curious to learn more. This post won’t try to convince you whether AI is a passing trend or a long-term shift. Instead, it’s here to give you a high-level overview of the basics to help you understand key AI concepts. Future posts will dive deeper into specific topics. The goal of this blog is to equip business owners, managers, and team leaders with enough insight to speak confidently about AI and make informed decisions as it becomes more relevant in manufacturing.

What Actually Is AI?

Artificial intelligence (AI) is the field of computer science focused on creating systems capable of tasks that normally require human intelligence. This includes problem-solving, learning, perception, language understanding, reasoning, and adaptation. AI systems analyze data, identify patterns, make decisions, and improve performance through experience—all based on algorithms and models.

Basics of AI Systems

You might interact with AI in many ways—from web-based tools like ChatGPT to AI features integrated into existing software. Here are the key elements of AI systems that help them operate:

Model

At the core of any AI system is a model: a mathematical tool or program that takes input and generates output. Initially, AI tools like ChatGPT only processed text as both input and output. Now, they handle images, spreadsheets, voice, and more, making them what’s termed multimodal. For instance, you can prompt ChatGPT with text while adding an image for context, and it will combine both to respond.

Most models today are multimodal, yet the term Large Language Model (LLM) is still widely used. Here are some examples:

GTP-4o – OpenAI

Claude 3.5 Sonnet – Anthropic

Gemini 1.5 Pro – Google

LLama 3.2 – Meta

The models themselves are, at their core just equations and code. Think of it as the engine of a car—it powers the vehicle, but you don’t interact with it directly.

To interact with the models you need a User Interface (UI). The companies who make the model typically offer their own UI to allow the general public to easily use their models.

OpenAI has ChatGPT

Anthropic has Claude

Google has Gemini

Meta has Meta AI

Typically the UI will be named similarly to the base model. Anthropic has models called:

Claude 3.5 Haiku

Claude 3.5 Sonnet

Claude 3 Opus

And they call their UI “Claude”.

AI companies also make their model available for other companies to use through what’s called an Application Programming Interface (API) . An API is a set of rules and tools that allow different software applications to communicate with each other. It acts as a bridge, enabling one program to use the functions or data of another program without needing to know how it’s implemented. APIs can also allow an AI model to use external software which really opens up the realm of what’s possible.

When you integrate an AI model into your own software, the AI company will charge you on a pay per usage basis. One example of a company using OpenAI’s model in their product is Microsoft. They integrated GPT-4 and other models into their Office suite of products under the name Microsoft CoPilot.

Think of CoPilot like an assistant built into your Office applications but unlike ChatGPT, it has access to your documents to help you work more efficiently. You can use it to summarize information, help you draft a PowerPoint based on various documents, summarize a Teams meeting into action items, and a bunch more.

All the major AI companies offer usage of their model through an API to allow developers to create products using AI. This also allows for customization of how you interact with the model.

Fine-Tuning

Some businesses take it a step further and fine-tune models to specialize in certain tasks by training them with specific data.

For example, an AI model like Claude 3.5 could be fine-tuned using a combination of process data and maintenance records to better predict equipment failures before they happen. The process data would include key operational parameters (e.g., temperature, pressure, flow rate) captured over time, while the maintenance records provide information on when failures occurred and the conditions leading up to them.

By fine-tuning the model with both types of data, it learns to recognize patterns and subtle shifts in the process data that precede equipment failures. This helps the model make more accurate predictions about when maintenance may be needed, ultimately reducing downtime and improving efficiency.

The process of fine-tuning adjusts the model’s parameters to focus on specific scenarios by optimizing its ability to recognize relevant patterns, but it can be costly in terms of both time and computational resources.

Alternatives to fine-tuning include:

• Prompt Engineering: Carefully crafting inputs to get desired outputs.
• Few-Shot Learning: Providing the model with examples during the interaction.
• Retrieval Augmented Generation (RAG): Using external data sources to improve responses.

The key point here is that fine-tuning might not always be the best option for your use case. Whether you should fine-tune really depends on things like how complex your needs are, what kind of data you have, and how much you’re willing to spend. It’s a tricky decision, and it’s definitely worth diving into more detail in a separate post.

Present Day Applications of AI

You don’t need to be a software company to take advantage of these models either. Lots of companies have been leveraging AI to create internal tools to improve their workflows and efficiency. There are plenty of no-code and low-code tools to help you accomplish this without being a software engineer.

Since the focus here is on Manufacturing businesses let’s look at a few present day applications. In future posts I’ll do a deep dive on each of these to show you what they can look like and how they’re made.

Customer Service Automation

You can have a chatbot on your company website or built into your social media to respond automatically to customers in a controlled way.

Customer service bots can be built with guardrails to control what types of questions they respond to and what will trigger them to pass the customer off to a real person.

The benefit here is providing a better experience for the customer. They receive an initial response rather than having to wait for someone to call them back or respond to their email. If you look at all your customer service inquiries you’ll notice the same questions get asked over and over again. Structuring an AI agent to respond to these will save your team time to focus on more complex problems and provide faster solutions to your customers.

Internal Sales Tool

If you’re in manufacturing, you probably have a bunch of catalogs, guides, and various information scattered around. Some of your senior employees may have a lot of that information memorized, but what happens if those people retire or leave the company? Having an internal sales tool loaded with information about your company and its products eliminates the time spent going back and forth internally to gather info for your customer.

This could be a chat tool integrated into Microsoft Teams or even WhatsApp, where outside salespeople could text a query and receive a response instantly when out in the field.

“Do we have stock of X?”

“What configurations does X product come in?”

“What’s our lead time on X?”

Research Tool

AI is incredibly powerful for research and learning. Tools like Perplexity provide answers with citations, making it easier to verify information. This can help you quickly get up to speed on a topic or find data without sifting through numerous documents.

Laying the Foundation for AI Adoption

We’re still in the early stages of AI development, and it’s natural to feel uncertain about how to move forward. Adoption takes time, but you can prepare for the future by:

1. Developing an AI Use Policy: Establish guidelines for responsible AI use.
2. Creating a Digital Knowledge Base: Organize your internal data for AI integration.
3. Sourcing Software with an API: Ensure new tools can interact with AI systems.

In the upcoming posts, we’ll explore each of these topics in more detail, offering guidance on how to approach them effectively. This is all about setting a solid foundation for whatever AI strategies you choose to implement down the line.

Getting Started with AI

If you found this useful and want to explore AI, my best advice is to start small. Try using tools like ChatGPT for various tasks in your organization. Just go to https://chatgpt.com/ and test it out. For example, you could use it to draft emails, generate quick content ideas, brainstorm, or even summarize lengthy documents. Talk to it like you would another person and you’ll get the hang of it quickly. The best way to understand AI’s potential is to experiment and see what works.

Till next time.