Since the June/July issue of the magazine is dedicated to technology, it would be a mistake to ignore the biggest story in technology of the last year: ChatGPT. Every corner of the culture, from entertainment, marketing, and product development in far ranging and different industries is now forced to reconcile with the rise of "AI" as a platform.
For contractors, it doesn't seem obvious at first how algorithms like ChatGPT, as well as the myriad of others springing up every day, would be useful in any way. A chatbot AI can't produce asphalt or go out and pave a road...yet. However, could it potentially act as a tool for contractors? Could it be useful as a source of information? How accurate or reliable would it be if you asked it questions related to mix design or compaction? The best way to find the answers to these questions would be to conduct an experiment and report our results.
Dispelling the Myth
Firstly, it is vitally important to mention that there are several important caveats. The most important might come as a shock to you, when you consider how most of the media, news, or other internet outlets discuss artificial intelligence. Calling these platforms "artificial intelligence" is a lot like calling the popular two-wheeled motion scooters my daughters love "hoverboards" when they don't hover in any way whatsoever, or another example might be that of crypto currency. These "AIs" are similar in the disparity between their names and what they actually are is vast and often misunderstood by the general public, and that misunderstanding is sometimes taken advantage of by others.
In a recent article on the subject for America Magazine, Jim McDermott wrote this:
"These programs generate complete and seemingly original works in an instant. They can also communicate with a person in a way that resembles actual conversation. But in fact, for the time being anyway, these programs are not sentient but just a very complex form of the kind of predictive text bot you find using Gmail or Google docs. ChatGPT-3, for instance, was trained on and is informed by 500 billion “tokens”—words or phrases culled from books, articles and the internet through which it interprets and responds to the prompts given to it. In calling these programs “artificial intelligence” we grant them a claim to authorship that is simply untrue. Each of those tokens used by programs like ChatGPT—the “language” in their “large language model”—represents a tiny, tiny piece of material that someone else created. And those authors are not credited for it, paid for it or asked permission for its use. In a sense, these machine-learning bots are actually the most advanced form of a chop shop: They steal material from creators (that is, they use it without permission), cut that material into parts so small that no one can trace them and then repurpose them to form new products."
When we talk about ChatGPT and "AI" it should be noted that it isn't "artificial intelligence" in any way that you might expect. It isn't thinking, or even approximating thinking. It is "scraping". It uses complex algorithms to aggregate data and information already created by human knowledge and documentation. It can mimic, it can reproduce, and it can predict based on previously established data, but it doesn't invent something novel. Which, in a way, makes it ideal as a potential tool for this experiment. We aren't interested in its ability to generate something new, but whether or not it could be used as a way of conversationally accessing established fact.
The Experiment: Having A Conversation with ChatGPT
Before diving into any of the questions or final analysis, I do think that this platform is interesting enough to at least play around with yourself, should you be curious to do so. It is fascinating to interact with, even more so when you know it isn't actually intelligently thinking. However, I do think that, near the end of the experiment, I began to distinguish its constraints more obviously the longer that I carried on the conversation.
If you are interested in indulging your curiosity, the first thing you will need to do is head to Chat.OpenAI.com and create a user account just like you would for any website. After you answer some basic questions about usage, and data sharing, you can start up a new conversation with ChatGPT in about five minutes. Once I completed these steps, I started in asking it some basic questions.
The questions I chose were based in my own curiosity, and from the gaps in my personal knowledge. So, if you find the questions strange, I actually think it plays an important function in the experiment which I will explain near the end. For the moment, however, bear with my naïve or seemingly strange line of questions.
Noel: What is the best asphalt mix grade for paving residential streets?
ChatGPT: The best asphalt mix grade for paving residential streets depends on several factors, including traffic volume, climate, and budget.
In general, residential streets typically require a mix that is durable enough to withstand regular vehicle traffic but also cost-effective. A common asphalt mix grade used for residential streets is a dense-graded hot mix asphalt (HMA) with a nominal maximum aggregate size (NMAS) of 12.5mm (1/2 inch) or 9.5mm (3/8 inch). This type of mix is designed to provide a smooth, long-lasting surface that can withstand moderate traffic volumes.
However, if the residential street experiences heavy traffic, such as buses or trucks, a higher-grade mix may be required, such as a Superpave mix or a stone matrix asphalt (SMA) mix. These mixes are designed to withstand higher traffic volumes and provide increased durability and skid resistance.
Noel: What asphalt mix grade is the most versatile while also being the most sustainable and environmentally friendly?
ChatGPT: One type of asphalt mix that is often considered to be sustainable and versatile is porous asphalt. Porous asphalt is designed to allow water to drain through the pavement and into the ground, which can help reduce stormwater runoff, improve water quality, and reduce the risk of flooding. This mix can be used for a wide range of applications, including parking lots, residential streets, and low-speed roads.
Porous asphalt can also be mixed with recycled materials such as reclaimed asphalt pavement (RAP) and recycled concrete aggregates (RCA) to reduce the demand for virgin materials and lower greenhouse gas emissions associated with asphalt production. Furthermore, the use of porous asphalt can help reduce the urban heat island effect, which can help mitigate the impacts of climate change.
However, it's important to note that porous asphalt may not be suitable for high-traffic applications, and the maintenance requirements for this type of pavement can be higher than for traditional asphalt.
With my first questions, something immediately became apparent: I was using ChatGPT like Google search. Decades of internet experience taught me that this is how you type questions into a search bar. Thankfully, a more experienced ChatGPT user was assisting me with the experiment, and was able to offer some advice. They explained that I didn't need to think of each question as a standalone event, but rather, an ongoing conversation on a topic. It would remember and reintegrate previous portions of the interaction, allowing my questions, as well as its answers, to become increasingly complex.
I started a new line of questions with one I knew the answer to, and then decided to probe ever more specifically the limits of its answers:
Noel: What are the best practices for the milling and repaving of a section of interstate road?
ChatGPT:
- Conduct a thorough inspection: Before milling and repaving, conduct a thorough inspection of the existing pavement to identify any structural or surface defects, drainage issues, or other problems that need to be addressed before repaving.
- Plan the project: Develop a comprehensive project plan that outlines the scope of the work, project timelines, traffic control plans, and equipment needs.
- Milling: The milling depth should be based on the condition of the existing pavement and the desired final pavement thickness. Careful control of the milling depth is essential to prevent damage to the underlying layers of the pavement.
- Prepare the base: Once the existing pavement surface has been milled, the base should be inspected and any defects, such as soft spots or inadequate drainage, should be repaired. The base should then be leveled and compacted to provide a stable foundation for the new pavement.
- Mix design: The selection of an appropriate mix design is crucial to ensure that the new pavement meets the required specifications for performance and durability. Factors to consider include the type of pavement, traffic volume, climate, and materials availability.
- Placement: The new pavement should be placed and compacted using the appropriate equipment and techniques to ensure proper density and surface smoothness. Adequate compaction is critical to the long-term performance and durability of the pavement.
- Quality control: Conduct regular quality control checks throughout the project to ensure that the work meets the required specifications and standards.
- Maintenance: Once the new pavement has been placed, it's essential to conduct regular maintenance, such as crack sealing, to prevent water infiltration and extend the pavement's lifespan.
At this point, I noticed two things from its answer. The first was that it is very good at mimicking the style of best practice articles, which wasn't a surprise when you consider that it could draw upon the internet's vast wealth of previously published work. The second thing I recognized was that it implemented a certain vagueness with some points that stuck out to me like when a high school kid is trying to "pad" out their literature essay with generalizations that aren't technically false, but aren't specific enough to really be of any real value, or show that the author actually did the reading, for that matter.
It felt concentrated in ChatGPT's answers in regards to mix design and placement, so, I pressed the matter and utilized my home state of Ohio as a basis for specificity.
Noel: What are best practices for the milling and repaving of a section of interstate road according to Ohio's mix design requirements?
This question only yielded a set of identical answers to my previous question, except it added the qualifier "Ohio" when referencing the required specifications. However, what if you didn't know the required specs for Ohio, as was the case with me? Depending on the next question I decided to ask, I felt I might be close to hitting a dead end, but I was rewarded, instead, with some progress.
Noel: What is the proper asphalt mix design spec for Ohio interstate paving?
ChatGPT: The Ohio Department of Transportation (ODOT) has specific asphalt mix design specifications for interstate paving projects. The ODOT 2018 Construction and Material Specifications (CMS) manual outlines the required specifications for asphalt mix designs. Here are some key specifications for asphalt mix design for Ohio interstate paving:
- Aggregate gradation: The aggregate gradation must conform to the ODOT 2018 CMS manual's requirements for specific asphalt mix types.
- Asphalt binder: The asphalt binder must meet the requirements of ODOT specifications for the specific mix type. The binder grade is determined by the expected high and low temperatures during the pavement's service life.
- Asphalt content: The asphalt content is determined based on the aggregate gradation, asphalt binder properties, and expected traffic loading. The total asphalt content should be within the specified range for the particular mix type.
- Compaction: The asphalt mix must be designed to achieve a minimum compaction of 92% of maximum theoretical density.
- Performance testing: The mix design must be tested for properties such as air voids, voids in mineral aggregate, and Marshall stability to ensure that it meets the specified requirements for performance and durability.
This was the first mention of the ODOT 2018 CMS manual, and while that was interesting, it acknowledged a separate source that could provide detailed answers, and it didn't provide the specs in detail in its own answer. I decided to pull out all the stops at this point, and ask a wildly specific question.
Noel: According to the ODOT 2018 CMS manual, what would be the proper mix design for a 2 mile stretch of highway on I-77 south near Cleveland, Ohio?
ChatGPT: To determine the proper mix design for a 2-mile stretch of highway on I-77 south near Cleveland, Ohio, additional information is needed such as the expected traffic volume, pavement thickness, and climate conditions. The specific mix design requirements will depend on these factors.
It was the very next thing it said in its answer that provided the first real instance of this experiment surprising me.
ChatGPT: However, I can provide a general overview of the mix design specifications for a Superpave HMA 19.0 surface course, which is commonly used for interstate highways in Ohio.
According to the ODOT 2018 CMS manual, the required specifications for a Superpave HMA 19.0 surface course are as follows:
- Aggregate gradation: The aggregate gradation must conform to the ODOT 2018 CMS manual's requirements for Superpave HMA 19.0 mix.
- Asphalt binder: The asphalt binder must meet the requirements of ODOT specifications for the specific mix type. For a Superpave HMA 19.0 mix, the required asphalt binder grade is PG 76-22.
- Asphalt content: The asphalt content for a Superpave HMA 19.0 mix should be within the range of 4.8 to 5.8 percent by weight of total mix.
- Compaction: The Superpave HMA 19.0 mix must be designed to achieve a minimum compaction of 92.0% of maximum theoretical density.
- Performance testing: The mix design must be tested for properties such as air voids, voids in mineral aggregate, and Marshall stability to ensure that it meets the specified requirements for performance and durability.
After about an hour of conversing with the language model, this was the first time it felt like it was "talking" back to me. It detected a gap of needed information in my question. I wasn't aware of the Superpave HMA 19.0 specification, and thus didn't include it in my question, and ChatGPT offered it as as example of a "commonly" utilized standard, and then provided a specific list of details about it. Although it took awhile, eventually, the algorithm handed me some information that was genuinely new to me.
Drawing Some Conclusions
I didn't list every question I asked ChatGPT about asphalt and the industry, instead, focusing in on one thread of that conversation that resulted in something interesting. Other questions did lead to dead ends, and I spent a lot of time analyzing it's answers to find out why I struggled to gain new insight from the platform which had access to much more information that I did.
The answer is that ChatGPT and the generation of language models like it, are only capable of one level of reflective analysis. That means it is only pulling in data related to your exact question, and if you don't know exactly what you don't know, then it can't really help you. Not like real people can.
For example, it was only able to help me with those specifics when I realized there was a term I was unfamiliar with (Superpave HMA 19.0), and could include it in my question by name. When you are talking with a human being who has real experiences and expertise to draw upon, they are able to grasp what you aren't saying, and offer information you didn't know to ask about.
Therein lies the first caution related to using this as a tool in our industry, it's going to leave things out, because you didn't ask -- not that any serious contractor was likely considering using ChatGPT for jobsite questions, but that doesn't mean a next-generation, young upstart contractor wouldn't. This reveals it's second weakness, which is that even when it does offer correct information, it doesn't yet have the ability to cite sources like we do in the industry. If we make a claim, we provide the expert, engineer, OEM, or research that backs up that claim. ChatGPT style language models have been known to completely fabric sources and remix information, much to the chagrin of students looking for an easy path to essay writing.
Computer, End Program
By the end of my nearly three-hour session with ChatGPT, it became apparent that it was most effective at mimicking conversational intelligence when I treated it like characters on Star Trek do the ship's computer. A google search that fulfills the original promise of AskJeeves (if you don't know what that is, then I am currently turning to dust).
It's weaknesses, though, outline what it could truly become useful in doing for contractors and business owners. If you trained the platform on your company's previous jobsite, scheduling, resource, and accounting data, you would be able to ask it questions like "based on the last year, what would be the most efficient employee schedule to reduce call-off?" or "Based on the last six month of material costs and upcoming projects, what future projects should be prioritized to maximize margins?" and it should be able to give you something useful.
The platforms that emerge to simplify the training of such an AI on your company's data will likely find a lot of demand for that functionality. In the meantime, I offer the final question I asked ChatGPT for this article, and it's full reply.
Noel: If you were an asphalt contractor, would you trust an AI language model enough to implement its guidance?
ChatGPT: As an AI language model, I can provide accurate and reliable information based on my training and the available data. However, it is ultimately up to the asphalt contractor to decide whether or not to trust and implement my guidance.
While I can provide useful insights and suggestions, I am not a substitute for the knowledge, experience, and expertise of a professional asphalt contractor. My responses are based on the data and information available to me, and may not take into account all of the variables and nuances that a contractor may encounter in the field.
Therefore, it is important for contractors to use their own judgment and expertise when making decisions about their operations. However, I can be a valuable tool for contractors to use in conjunction with their own knowledge and experience to make informed decisions and improve their operations.
