AI: Understanding the Very Knowledgeable Apprentice

This is the HVAC Mentors Podcast. Welcome everybody. I'm Mike and I am so happy to have you join us today. I am extremely happy about this series that we're doing. This is a technician's guide to artificial intelligence. We are talking about AI in the trades, in our jobs, and what it means for us as professionals and as technicians. Follow along. I think you're gonna enjoy. What we have in store for you. So this is episode 2, what AI actually is. So I think part of the mystique of artificial intelligence, part of what makes it sort of a little scary, to be perfectly honest, and a little controversial, is the fact that a lot of people don't necessarily understand what AI is. In fact, one of the joys that I've had in preparing for this series has been researching AI and finding out exactly how it works, what the programming is, what the intentions of the people who made it are or were. I'm really excited about today's episode because we're going to just unpack a bit of what AI actually is, not down to the code. It's not, that's really not necessary, but how it works, what the theory behind AI is, and really Why it's not scary at all. So as we get started, I just want to jog your memory from the last episode. So we started with a story, right? Refrigeration technician responding to a service call for a walk-in cooler, smart gauges logging pressures, probes logging temperatures, service notes getting written without you having to write them, interactions with staff and managers being logged automatically. And data tags being recorded so that we have models, serial pressures, voltages, refrigerants, all of that vital information on the system. Parts availability was checked automatically, and all of it was happening behind the scenes while the technician was still standing in front of the equipment. At the end of the story, we introduced the tool making it possible, artificial intelligence. But before we start talking about how AI might show up in the trades or whether it might replace us, the technicians, or for that matter, whether it'll help us. We need to start with a much simpler question. What is AI? Let's unpack that a little bit by unpacking some of the myths surrounding AI. So first of all, when most people hear the words artificial intelligence, they picture something like a robot brain, something that thinks, something that understands, something that makes decisions the way a human does. Picture is not really accurate. Artificial intelligence doesn't actually think is the thing. It doesn't understand, and it definitely does not have intentions. AI is simply software that recognizes patterns. You know, you've heard of a search engine. Think about it as a pattern engine linked with a language engine. It's designed to recognize patterns and make predictions based on those patterns. So when you say XYZ, it's going to say based on a probability, the correct response is Probably ABC, but that's all it is. There is no conscious consciousness. There is no awareness. It's just patterns and predictions. The rest is ones and zeros. A simple way to think about this is if you imagine you had a helper who had watched thousands and thousands of service calls, okay? This is a person who has followed you around maybe for years. They've seen you diagnose plenty of compressors, circuits, Check superheat, inspect airflow, analyze system pressures. They've watched you do it all. Okay, now imagine that helper is standing next to you on a job. They might say something like, hey, the last 100 times I saw suction pressure that low with superheat that high, the system was low on refrigerant. That helper isn't necessarily right every time, but they have seen a lot of patterns and they can point out possibilities. Artificial intelligence works in a very similar way. It learns patterns from large amounts of data, and then it uses those patterns to make predictions. The crazy thing about AI is while it's been talked about and it's been becoming more and more prevalent lately, it's not actually new. We've been using forms of artificial intelligence for quite a while. You've used it in your car to navigate to a destination. GPS is a form of AI. If you've used Siri, you've used AI. Spam filters in email are a form of AI. They recognize patterns in unwanted messages. Credit card companies use AI to detect fraud. Voice-to-text on your phone, that's another form of AI. It's basically interpreting your speech patterns and converting them into text. Search engines use AI to rank and sort information. So artificial intelligence has already been quietly working behind the scenes in many parts of our lives. What's different now is that AI is becoming much easier for regular people to interact with directly, and that's where tools like chat-based AI come in. So I already mentioned ChatGPT, and if you remember in the last episode, I gave you a challenge. Take a photo of any data tag on any machine that you have worked on while you're out in the field, upload it into ChatGPT and ask it, or any chat bot application for that matter, and ask it, what can you tell me about this? Don't tell it what it is or how you encountered it. Just ask it, what can you tell me? And give it that picture. Now, if you did try that, then you're probably surprised by how much information it can give you. So over the last few years, you've probably started using and playing with things like ChatGPT and other conversational AI systems. These tools do feel really different from traditional software because it's very conversational. You type a question or you can speak a question into it, and the AI responds almost like you're having a conversation with another person. And that can make it feel like it understands what you're asking. But the truth is something a little bit different. The chat-based AI doesn't actually know things the way that humans know things. Instead, it predicts language. So these systems are trained by developers on enormous amounts of written text. We're talking about books, articles, websites, technical documentation, online conversations. During training, the AI studies patterns in how words appear together. It learns things like, for example, when people say compressor, they also often mention refrigerant, suction pressure, or superheat. When someone asks a question, certain types of answers usually follow. So when you type a question into a chat AI system, it doesn't go searching through a database for the answer. Instead, it predicts the most likely sequence of words that would form a helpful response. In other words, it's predicting language. That's why these systems can sound so natural. They've learned the patterns of human communication. In fact, here's a funny story. So today I was out on a service call on a commercial dishwasher at a nursing home, and I was working on my service notes and putting pictures from the job into ChatGPT. And ChatGPT will also sometimes give you suggested routes of diagnosis. And things, and you can choose to ignore those and you should. Not that they're wrong, but because ultimately you make the decision on how to go. You're the technician. But what I did was I basically took a picture of each stage of the wash, rinse, sanitize cycle on this dishwasher. And ChatGPT suggested to me that it was overfeeding wash detergent because of the high amount of foam, which There was a high amount of foam, but the reason was that the dishwasher was actually underfeeding water. So ChatGPT is intelligent. It recognizes patterns and it assigns a most probable response to that pattern. And in AI's case, it got hyperfocused on the image of the soap suds instead of the diagnostic information that I had given it. In text, I had put in that I tested the fill solenoid and found that the valve was sticking and that's why we were underfeeding water. But it kind of glazed past that and paid more attention to the picture that it was analyzing, which I had to laugh because a part of the way it interacts with you is it does learn your mannerisms and your way of speaking. And so it just seemed funny to me because I'm much more of a visual communicator, like visual learner. And so I could see where I would possibly make that same mistake and focus on the photo and not on the words, the text. So again, it's using your words to predict a helpful response. In other words, it's predicting language. It's not actually sentient. It's not conscious in the same way that we are. It's not conscious at all. That's why these— but the fact it's not conscious at all, but the fact that it's designed to predict language is why it can sound so natural. They learn the patterns of human communication, and their developers have designed them to predict your means of communication, not to take advantage of you in any way, but to simply make it more comfortable for you to use them. So the reason this matters is that it helps us to avoid 2 common mistakes people make with AI. The first one is that sometimes we can tend to think that AI has some kind of a magical superintelligence. It doesn't. It's got what information it's been given. The second mistake that we make is assuming AI is useless because it makes mistakes. But the truth is that we make mistakes. We are far from useless. We make mistakes all the time. The real value of AI is that it can process huge amounts of information very quickly and recognize patterns that might take a human much longer to spot. Case in point, if I were to just write out my technical notes on my work orders, they look jumbled, garbled. They're just really hard to understand, much like the contents of the thoughts inside of my head. But because I use AI to organize my service notes, I can create bulleted diagnostic steps and a nice clean flow for how the service call progressed, but At the end of the day, AI still needs a human in the loop, not you, the user, but the developer giving it the information and managing the information and managing the algorithm. So as we bring it back to those of us who are in skilled trades, artificial intelligence can analyze system data. It can organize information. It can suggest possible diagnostic paths, but it can't climb a ladder. It can't drag your tool bag up. Trust me, I've asked ChatGPT to please load itself onto a giant drone and fly my tools up to the rooftop to work on a rooftop unit many times. It refuses. But it can't do all that stuff. It can't smell a burnt motor or motor winding. It definitely cannot replace the experience and judgment of a technician standing in front of a piece of equipment. AI is not a replacement for technicians. It's a tool, just like digital gauges. Just like multimeters, just like combustion analyzers. The technicians who learn how to use these tools effectively become better technicians, and artificial intelligence may simply become the next tool in that toolbox. So now that we understand a little bit better what AI actually is, I want to introduce the next episode where we're going to dig into something that confuses a lot of people. And it confused me at first. Why do chat-based AI systems sound so human? We kind of touched on that earlier, but we're going to dive a little bit deeper. Why do they feel like you're talking to a person? What does that mean for how we should think about using them? That's coming up on the next episode of the HVAC Mentors Podcast. Thank you so much for joining me today, and I look forward to seeing you next time. Bye for now.