Sustainable Fractal Design: Maximizing ChatGPT and BIM Energy Analysis

DANIEL BREUL: Hello, everyone. Today I am talking about sustainable fractal design and maximizing ChatGPT and BIM Energy analysis. My name is Daniel Breul, and I am a structural technician at Gannett Fleming.

So to start, we'll talk about what is sustainable fractal design. Fractal design, by definition is architecture-- is an approach to building design that integrates the principles of fractal geometry and complex systems theory to create buildings that are responsive to dynamic patterns and processes and various different ecosystems.

It's essentially based on earthship design. The earthships that are in Taos, New Mexico use various design principles of solar heating and cooling as well as different reactive systems to water capturing and using planters and recycling. So it's loosely based off that, but it's a concept that brings it forward a little bit.

So the fractals that we're talking about and the fractals that people are normally used to is geometric fractals. These geometric fractals occur in what's called third-dimensional phase space. A phase space is where a complex system has every single possible outcome from its initial onset.

We're mainly concerned with fourth-dimensional phase space, so that would be geometric patterns but using mathematical patterns. Instead of getting 3D geometric patterns and third-dimensional phase space, it uses time as the coefficient to drive design changes that would occur over time. So the buildings are essentially reactive buildings.

And that right there at the bottom is a earthship diagram that shows the heating and cooling effect that it uses passively. So chaos theory-- the main part of fractal design uses chaos theory. And chaos theory, you might have noticed, in our daily day-to-day lives isn't something that you would think of very much, but it is something that is in the background.

So one good example is every hurricane season we get those spaghetti models. What those spaghetti models essentially are are using chaotic systems to determine where each hurricane would go, and because a hurricane is such a complex system, it is difficult to predict, which is why your weather patterns and your daily forecasts might be completely incorrect, and it's because these mathematical models are extremely difficult to calculate.

And our current ability, as you've seen in the graph below, shows that we have very little idea of what a complex dynamic system will do. And right now, chaos theory is the best way we have of trying to predict these systems, and it's predicted through the use of initial conditions, which brings me to chaotic attractors.

So a chaotic attractor, similar to the spaghetti model, are what is used to create these predictive systems. And these predictive systems can then be used to predict an environment like a complex ecology, and what fractal design does is it uses these chaotic attractors to understand any given environment.

So right here is a bifurcation sequence, and what the bifurcation sequence does is it allows you to calculate complex things. So right here is an example of it modeling a population growth of any given area, and what the R-value is is the rate of change that occurs.

So right here you see a system going from a state of simplicity to infinite complexity, and that is what we consider chaos when it gets to the part that looks like it's complex. That part-- the system becomes a chaotic system, and it's hard to understand where that pattern would go. But a bifurcation sequence is a good way of mapping that if you have initial conditions and an emphasis on initial conditions.

The better understood the initial condition of a system is, the better you can predict it down the line. And a lot of times, these graphs-- you can chart that specific picture right there of bifurcation. You can chart that down the line and put that on an actual graph that's easy to understand.

So speaking of chaotic attractors, complex dynamic systems-- so the purpose of these attractors is to understand complex ecology, and right now, when we build a building or any built environment, very little of the complex system that is being built in is taken into account. So what this system-- what this new type of architecture does is it uses system impact and building integration and then uses complex attractors and understanding of dynamic systems to predict what impact the building will have.

And when you put a building into a new system, it's hard to tell what will happen, so what we try to do is we try to design the initial conditions in such a way that we could help predict the outcome of what impact it will have, basically making the building into a new ecological paradigm within whatever ecology it's built in.

Right here are some examples of models that have been created to understand the different changes in an environment. So the one to the left right here gives you wind force, and the wind force is charted with the amount of time that a high wind is in any given area. And that will be deterministic of what kind of trees fall, and then when you look at those patterns of how many trees fall due to wind, you can start to model a predictive algorithm that gives you an idea and modeling of what the future environment will look like.

So how does this all fit in with ChatGPT? Well, what is ChatGPT? Well, it is a generative pre-trained transformer. ChatGPT is a natural language processor model developed by a company called OpenAI. What happened with this and why it's different than other AI is it is a chat interface, so unlike other AI models that we interact with, this one is something we are able to actively type in and chat with, hence the name, ChatGPT.

And ChatGPT is able to use machine learning to understand complex human speech in, essentially, texting. So it uses its interaction with humans to be able to learn what humans are doing, what their patterns are like, and thus it is able to learn from us to become a better helper to us.

That being said, there are some dos and don'ts of using AI, especially in a professional workplace, that I feel like needs to be said. One problem with it-- and a lot of people who have maybe tried it in the workplace might know-- is it can be sometimes unreliable. There is a case where a lawyer tried to use it for coming up with different case study cases to bring to his case, and they were made-up ones that ChatGPT came up with.

So it is always wise to double-check whatever you're doing, especially when you're using it in the context of BIM and engineering. It's always good to double check whatever work you've had to do, made sure you check it with either someone else, peer review, or double-check it with a factual database.

Another problem is confirmation bias. So after you chat, like I said before, it has machine learning, so it starts to learn your style. And it starts to predict what kind of answers you would desire apart from the answers that it would just give otherwise, and that could be a problem with confirmation bias because then you can start having a feedback loop with ChatGPT where you just continually confirm a narrative that you have in your head about something.

And this is super easy to do with engineering and architecture when you already have some sort of bias about a new design you're working on or a way of doing something. So in my opinion, it should be used as an assistant only.

Another problem is copyright issues. Sometimes it has a way of taking from somebody else's work that is otherwise copyright and spits it out to you without you knowing that to be true. Another really good way to use ChatGPT to get the best answers you can is be very specific with your questions. The longer and more specific the question, the better the response ChatGPT will be able to give you.

Here is an example of-- I asked it what kind of mistakes it made and, very similar to what I was saying, generate incorrect information, sensitive to the input-- that's why you want to be very specific with how you put your question to it-- and biases is a major part of ChatGPT.

So how did ChatGPT help me to develop this concept of fractal architecture? Well, one of the ways it helped-- it was a great learning tool. I knew what objective I wanted to accomplish with fractal architecture, but I was unclear on some of the ways to go about developing.

It was a great learning tool of opening up ways that I could develop my concept, and as I developed it further, ChatGPT better understood what kind of concept I was going for and that I wanted to use fractals in not a geometric sense that have been used in the past but in a fourth-dimensional sense.

And it was great at being able to adapt to what I wanted to do, and it was perfect for writing Python scripts, which-- a lot of what reactive design needs as it relates to Revit and BIM is a lot of Python scripts that you can import into Dynamo or that you can use to open the Revit API. One of the things that introduced me to combining the learning part with the Revit API is it introduced me to Visual Studio by Microsoft, which is a good way to be able to open up Revit API and write Python scripts.

It's also a fantastic brainstorming tool. Whenever I got stumped on a topic or something that I wanted to develop as part of fractal architecture, it was always a good way of bouncing ideas back and forth with it because it's able to understand the concept I want, and it is able to understand that I want to bounce ideas back and forth with it.

So why use AI? Well, AI is really good at machine learning. It's really good at adapting to your specific workflow and work style. ChatGPT can also remember past conversations, which is why I said before it was great at brainstorming.

Right here is a picture of a Waymo self-driving car, and similar to ChatGPT, the Waymo self-driving car uses interactions with its environment to machine learn about different variables that could occur in it. One time I saw a pedestrian going, and they waved the car forward. And the car learned that that was something-- that it can go ahead and make the turn.

That showed me that it had advanced machine learning, and from experience, like a human, from driving after time, it's able to get better and better, which is why they've been driving these with humans in them for so long while the machine learning was able to build a complete back repertoire of experiences from its environment. So similar to that, ChatGPT is able to learn from using it in the workplace.

So here are some of the Pythons that-- Python scripts that I got from it. So right here I was using it to try to have a solar analysis program in Revit and have an add-in which allows ChatGPT to interact with the solar analysis that I was running. And right here it gives you kind of an over idea of what you need, so it gives you lists. ChatGPT likes to give you lists when you have it do something complex, and these lists will be very superficial and very surface level.

But what you can do is you can use that to start getting deeper into what ChatGPT can do to help you. And so right here I asked it to help me with a specific step, and then we're able to drill down and get more specific with what we want the operation to do. And you can see right here eventually I asked it to write a Python script about a specific part of running the energy analysis.

So speaking of using BIM energy analysis, ChatGPT helped me put together a few generative design scripts where I could run iterative design on whatever I was working on. So right here is a building that I was running solar analysis on, and right below you can see the picture. That's the Revit Solar Analysis program being run.

And what I did is I used it to create a real-time reactive design. So what I was able to do is I was able to run this, take the data from the solar analysis, and run it into an external program. So I won't go over the external program too much since it's not an Autodesk product. But Revit is, and that is where the data and information is being collected from ChatGPT to be able to help drive the design of what I'm doing in real time.

So I have a parametric roof, and I'm able to change the shape of the roof depending on how much sun shines through and what the surface area of the sun is. And then I'm able to use ChatGPT and the other program to calculate the BTUs of that specific unit.

And what I have down there is a condo unit, and I can look at every specific one, and depending on the angle that the sun hits it, I'm able to change that in real time with generative design because ChatGPT enabled me to export the data to an external program and then re-import it and then in real-time change that design.

So like I was saying earlier, it drives parametric design with analysis feedback. That is real-time feedback, and it's not something you have to wait for because you can run it, get the data back immediately, and then run it again if it's not to your desired amount. And yeah, it adapts to what you do. So after you've done this a few times, ChatGPT gets better at understanding what you want to do and is able to streamline the process easier of having it go from Revit to another program to back into Revit.

Another way that I've used it with energy analysis isn't the analysis that is in Revit, but I used it to calculate tree and plant dispersion. So this is a good example of a fourth-dimensional phase space as far as it relates to fractal architecture, and what that does is it allows you to decide where would be the most beneficial place to place your building and how to orient it with what the given flora and fauna of any given ecology is.

And it does that by using the machine learning and advanced algorithms, and ChatGPT is really good at running these calculations as an AI and using them in other open source AI programs. And crunching those numbers, you're able to have a predictive model and a predictive system to where you're building will have the least amount of impact.

So when AI learning model meets sustainability-- these two are two things that are starting to be widely used in the sustainable communities one of the reasons is it creates a new ecological paradigm, a new system, so to speak. Our current cities and developments are made not with any sort of different paradigm, and the effect is that they end up being a detriment to that system instead of a help to that system.

What combining AI with is able to do is you're able to minimize ecological impact of these buildings, and you're able to encourage inhabitants to live within an ecological system. And one of the reasons is because when you have a reactive building that's dependent on it, it incentivizes people to live with that building and thus living with any ecological environment and its systems.

So what is the future of AI? Right now we're only seeing the beginnings of it. We're starting to see OpenAI like ChatGPT, and we're only going to start seeing it more as these things start to develop.

I was using ChatGPT-3. Right now ChatGPT-4 is the most cutting-edge as far as OpenAI is concerned, and there will only be more programs like that. And as these programs become more and more advanced, they are able to learn more and more off the previous programs, so every single iteration of these is an exponential advancement.

And another good thing about AI is that people can start using them as learning opportunities. They're good ways for you to learn, and they're really great for interacting with since it's not just a medium in which you're getting information in. You're also giving information out. It's an interaction that you're having. So it can almost be like a personal tutor for yourself as it is something that can chat back with you.

I see it as something that will be more and more integrated as far as BIM is concerned. Right there at the bottom, you see that ChatGPT plus pyRevit is something that is already being developed, and I could see that as being an extreme help, especially if you're a Revit user like me and you use pyRevit a lot, which we use quite a bit in our line of work. And it's quite helpful program, and so having an add-in of ChatGPT within Revit and Python along with pyRevit has been an extreme help, and I can only see that being advanced further in the future.

That is it. Thank you for listening.