AI-driven, natural language processing (NLP) tools such as ChatGPT from OpenAI and Google Bard are enabling humans to have normal conversations with computers. While AI tools like these are capable of doing many things, using them as a research assistant is a common way to dip your toe into the AI pool.
Many of these NLP tools build their language models by using publicly available information. For instance, OpenAI relies on three sources of information: what’s available on the internet, information licenses from third parties, and information their users and human trainers provide. This “training process” not only supplies the AI tool with lots of information, but also allows it to pick up on language patterns so it can become a more proficient communicator.
At times, the response you get from an AI tool is not only impressively fast, but also dead-on accurate. At other times, the response can be a little off. That’s why tools like ChatGPT and Bard recommend fact-checking “important” information. But doesn’t that defeat the purpose of using an AI tool in the first place, where one of the primary benefits is reducing time to insight?
“With our asset management software, and even a lot of shop management software that’s out there, the assumptions are controlled,” said Brian Antonellis, senior vice president of fleet operations at Fleet Advantage, an asset management and data analytics firm. “We are taking industry-accepted norms and introducing them to an algorithm, so a fleet can have confidence in the assumptions that are being made.
“While we do get excited about what AI means, we’re also cautious about putting the right data into the system,” Antonellis continued. “Even today, you don’t really know what data and information is in something like ChatGPT. A lot of it is coming from the internet, and that information can be as much as two years old if you’re using the free version. So you have to be cautious. The last thing you want to do is pull a query on the wrong data. And I would be really careful about leaning in too strongly on anything technical, unless you have a verified data source and are comfortable with the assumptions being made, and are able to vet them through your OEM.”
Read more: How AI is becoming the ultimate shop assistant
We asked the free version of ChatGPT questions a fleet maintenance manager might have, including a few on predictive maintenance, and then compared the responses to what a human expert in the trucking industry had to say. In most cases, ChatGPT’s responses were somewhat general, whereas the human industry expert came back with a more refined answer.
Take a look and decide who you’d trust.
ChatGPT vs. Trucking experts
How often should I clean the DPF on my Kenworth truck?
ChatGPT: Cleaning the Diesel Particulate Filter (DPF) in a Kenworth truck depends on various factors like driving conditions, mileage, and manufacturer recommendations. Typically, it’s recommended to inspect and potentially clean the DPF every 150,000 to 300,000 miles, but this can vary based on your specific truck model and usage patterns. It’s advisable to consult your truck’s manual or the manufacturer’s guidelines for precise maintenance schedules tailored to your vehicle.
Person: Steve Hoke, president of Diesel Emissions Service: That response is spot on from a manufacturer’s point of view. But real-world results would prefer that a DPF be cleaned in the 100K-150K range for the best longevity of the DPF.
What is the best oil filter for a Detroit DD15 engine?
ChatGPT: The choice of the best oil filter for a Detroit DD15 engine may vary based on factors like specific model requirements and personal preferences. However, reputable brands such as Fleetguard, Baldwin, and Donaldson are often recommended for DD15 engines due to their quality and compatibility. It’s crucial to consult your engine’s manual or manufacturer recommendations to ensure you choose an oil filter that meets the required specifications for optimal performance and engine protection.
Person: Len Copeland, product marketing manager for Detroit Components: While we acknowledge that AI has become influential technology, we still recommend our own branded oil filters for our engines. All of the recommendations are on our publicly accessible Techlit webpage at dtnatechlit.com/portal-public/ (search for “Supplemental Oil Filters”).
How can I improve efficiency in my truck repair shop?
ChatGPT: There are several things you can do:
- Streamline the repair process
- Employee training
- Utilize technology
- Regularly maintain your repair equipment
- Clear communication
- Set standard operating procedures
- Customer service focus
- Performance monitoring
Person: Jacob Findlay, Fullbay founder and chairman: The best thing a shop can do is grab a pencil and blank sheet of paper, and draw out what the repair workflow is today. Identify the areas where time and/or cost tend to get out of control, and find ways to do things more efficiently. Also, be sure to measure technician efficiency, which is invoiced hours divided by total hours worked. Publish the results so the Hawthorne Effect can play out (which says people change behavior when behavior is being evaluated). Once the key metrics and processes are laid out, set some goals and continuously monitor them.
Person: Nick Pittinger, director of data services at Decisiv: Streamlining communication between the fleet, shop, and suppliers is one of the best ways to improve efficiency. When communication is consistent and clear, the repair event typically goes much smoother and faster. Leveraging standardized operations will also improve efficiency, and is much better than leaving each technician to figure things out for themselves. Improving workflows by automatically assigning tasks through a software platform will also help save time. Finally, shops should find ways to measure performance, analyze that data to identify bottlenecks, and use it to get a little bit better and faster.
ChatGPT vs. Predictive maintenance
For our predictive maintnenace showdown between man and machine, we asked the free version of ChatGPT some predictive maintenance questions, and vetted those answers with some human experts.
What indicators should a fleet manager look for to determine the readiness of their company for integrating predictive maintenance components?
ChatGPT: Fleet managers should scrutinize key indicators such as ensuring effectiveness in the current preventive maintenance process, considering fleet characteristics such as age and mileage, and evaluating engagement with telematics providers for dynamic solutions. Collaboration and customization play a pivotal role, aligning implementation plans with fleet goals. Integration into existing workflows, addressing education gaps, and prioritizing technician training are essential. Additionally, the company's capability to handle data influx and prevent overload should be considered. By comprehensively examining these factors, fleet managers can determine the preparedness of their company for the seamless integration of predictive maintenance components.
Andrew Bartholomew, associate product marketing manager, Fleetio: It’s not that complicated! An ideal indicator for us would be when an increasing proportion of downtime or repair spend is resulting from repairs that couldn’t be avoided with a comprehensive preventative maintenance system. Essentially, if a fleet’s preventive maintenance process is in order, it’s easier to think ahead; there aren’t going to be as many unexpected issues popping up and taking all of the focus.
How can a predictive maintenance algorithm and software adapt to changes in how a fleet operates or how they grow over time?
ChatGPT: A predictive maintenance algorithm and software can adapt to changes in a fleet's operations and growth by being dynamic and scalable. These systems can accommodate new vehicles, changes in regulatory compliance, energy transitions, and even the integration of autonomous vehicles. Telematics providers, who often offer these solutions, collaborate closely with fleets to custom tailor implementation plans, ensuring flexibility. The algorithms incorporate advanced AI techniques to process and prioritize fault codes. This adaptability allows the system to remain effective as the fleet undergoes operational shifts or experiences growth, ensuring that predictive maintenance continues to deliver meaningful insights and benefits over time.
Read more: How to unleash the power of predictive maintenance
James Rice, president, Uptake: In the last probably 6 to 8 months, I'd say our models and process with the algorithms have actually even gotten down to the asset level. We're taking a little bit more into account at the asset level as opposed to component or fleetwide data that says all 2,000 of my Freightliners should all act the same. The ones in California aren't gonna act the ones that you know same in New York. When you have the baseline foundation of what it should be but if you're in an extreme weather event or extreme temperature, we're not gonna automatically say there's a problem here.
We'll look and determine if there’s an anomaly and if it’s sustainable. Is it a weird thing for the day, or is it really outside the normal operating measure? That has been how this product is evolving and as data scales, we will continue to build out the power of these algorithms.
At what level of sophistication should fleets commence their implementation of an AI-driven predictive maintenance component?
ChatGPT: It depends on various factors, such as the fleet's size, operational complexity, and existing technological infrastructure. It's recommended for fleets to conduct a thorough assessment of their specific needs, considering factors like the age of their vehicles, maintenance history, and the desired level of automation. Starting with a phased approach, integrating AI gradually and scaling based on successful outcomes, can be a strategic way to ensure effective implementation without overwhelming the existing operations.
Alan McMillan, president, Intangles: No, it's really straightforward. We plug into their CANBUS through their OBD port. Even if they have something in the OBD port, we just put a pigtail in. In almost all of our installations there’s already a GPS device or an ELD device running in that vehicle. It's no problem to run alongside that and frankly the truck driver doesn't even know we're pulling all that information back to command center. It's all done through the cloud and naturally we're also doing things like driver behavior we do all sorts of predictions around that.
What challenges would you advise people to be mindful of during the integration of a predictive maintenance solution?
ChatGPT: The answer would depend on the insights and experiences of individuals involved in the integration of predictive maintenance solutions. They might discuss challenges related to technology adoption, data management, technician training, or overcoming hesitancy in embracing new approaches to fleet management.
Vedant Khattar, CTO, Pitstop: Don't think it will just start working out-of-the-box. Each company has a different threshold of what is critical, and what is minor, so you have to give it at least like 2 to 3 months, because every model takes some time to calibrate. You really need patience working with a new system like this so don't get frustrated if the system says something is a critical issue but when you check it was only a minor issue. Those incidents will happen. If it’s reported and then happens again, then it is a problem. But if it happens one time and the computer has learned and improves, it’s only going to get better from there.
ChatGPT vs. Emails
Finally, we all could use help phrasing things with the right balance of cold calculation and empathy, especially when breaking bad news to a customer. So, we wanted to see if the free version of ChatGPT was up to the task and asked the following question:
Can you write an email telling my customer that their truck repair will be done Thursday afternoon once a backordered part comes in?
Here’s what ChatGPT produced:
Dear [Customer's Name],
I hope this email finds you well. We appreciate your patience and understanding as we work on repairing your truck. We wanted to provide you with an update on the status of the repair.
Currently, everything is progressing smoothly, and our team is diligently working on resolving the issues with your truck. However, we have encountered a slight delay due to a backordered part that is essential for the completion of the repair.
We want to assure you that we are actively monitoring the situation, and the backordered part is expected to arrive by Thursday morning. As soon as we receive the part, our technicians will immediately proceed with the repair work.
Our goal is to have your truck fully repaired and ready for pickup by Thursday afternoon. We understand the importance of having your vehicle back in a timely manner, and we sincerely apologize for any inconvenience this delay may have caused.
If you have any questions or concerns, please feel free to reach out to us at [Your Contact Information]. We truly appreciate your understanding and cooperation in this matter.
Thank you for choosing our services. We look forward to completing the repair and returning your truck to you in top-notch condition.
Best regards,
[Your Full Name]
[Your Position]
[Your Company Name]
[Your Contact Information]
