Ethics in Computational Design: Infusing Equity into AI and Generative Design for Building Design Processes

YAEL NETSER: Hello, we are here today to discuss ethics in computational design infusing equity into AI and generative design for building design processes. My name is Yael Netser. I'm a licensed architect in Israel. I also own an ATC called Revitalize specialized in Revit and BIM implementation with 15 years of experience in this field.

I'm also a PhD student in Carnegie Mellon University for architecture, engineering, and construction management. My research focus is social and ethical impacts of BIM, generative design, and generative AI on the architectural design process and the built environment.

MICHAEL BURSHTEIN: Hi, I'm Michael Burshtein. I'm an architect at AGD. I'm licensed in Michigan, and I specialize in corporate commercial project and community-focused design.

YAEL NETSER: We're going to start with a bit of introduction. What is this class actually about? And before that, we're going to speak about what this class is not about. Even though it's very tempting, we are not asking, will AI replace us as architects? Is it even able to replace us, is the technology good enough, are human designers better than machines, will AI be the end of humanity?

Even though all of these are very valid questions, we are here today to discuss something a bit different. What this class is about is we're assuming that AI is a tool which will be incorporated into our design processes as architects. We don't know yet to what extent it will be incorporated and what part it will play. But we are asking, in the case that AI is incorporated into our processes, what may society lose when architects rely more and more on AI on their design?

Here's the agenda for today. We just passed the introduction. Then we're going to introduce generative design and generative AI use in architecture, kind of state of the art at the moment. We're going to discuss ethics in architecture, very briefly, because of course, it's a very huge topic. And then we're going to focus on ethics for using generative AI in architecture, and eventually, we'll get to how we can prioritize social sustainability and equity into our design.

So let's talk a bit about generative design and generative AI in architecture. First, we'd like to introduce the terms to anyone who is not familiar. What is the generative AI? Generative Artificial Intelligence, or GAI, is a group of algorithms capable of generating either text, images, or videos based on statistical models that are learning from a database. They learn themselves, and then they try to predict and generate some new data based on the algorithm and the databases that they learned from.

So in this group of services, we have many, many services, for example, text generating chatbots ChatGPT, Copilot, and so on. We got the text-to-image generators, DALL-E, Midjourney, and many more. And we also have some text-to-video generators.

For the AEC industry, these types of services are very new, only lately introduced, and they can help us with programming, with preparing documents, renderings, and so on. Sorry.

What is generative design? Generative design describes a pretty different algorithm. These algorithms are actually iterative design processes that generate outputs based on specific constraints and metrics defined by the designer. The designer can see the results, pick the best one in their eyes, and then iterate the process again.

Example of generative design algorithms in architecture, there are many, many services like this one. This is just an example from Sidewalk Labs, but the idea of all of them is pretty much the same. You have metrics. Here we can see residential units, walkability, access to transit, daylight access, construction cost, and energy usage.

The algorithm generates a lot of design options, and it gives grades to each metrics. And you can see the overall score in the top and then you can decide, based on other things, not only what the generative design measured, what is your preferred design option.

So these services are completely different, and they are even pretty much complementary. While generative design is quantitative, you define, you execute procedure, and you have deterministic controls. You get pretty much a predictable, result. And you can understand where it came from. Generative AI is a qualitative algorithm. It trains, it learns something, you get some kind of outcome based on statistical models, and it can be unpredictable.

If we look at services starting to emerge of AI for architects, we see many terms. A lot of people use just AI. A lot of these services, you can see on their website, they are saying AI powered, AI-powered design. We can also see the terms generative design and generative AI in the descriptions of these services. But some of them even mentioned both of them.

So we have to give a disclaimer here. We are architects. We are not programmers, so we can't exactly be sure what technology each of the services we are using, is utilizing. So we are going to focus on what they are going to do for us and not discuss the technology itself.

So let's see some of the generative-design best apps available in the market at the moment. I'm sure most of you know them or even use them. So if we look here, we see many services that can generate design based on some kind of constraints or metrics.

We have Autodesk Forma that uses automated nesting takeoffs and real-time environmental impact analysis. We have HYPAR that does design automation platform for buildings. You can create your design, you can see a lot of metrics for cost and energy use and whatever quantitative data you want, you can even. Introduce some images for facades and the program will design your building for you in according to that. We got TestFit that does neighborhood design, parking design, also interior design, and Finch, which uses graph technology to design buildings with the interior.

When we look at generative-AI based apps, and again with the disclaimer that this is from our understanding, here is an example of three of them. Midjourney a very robust text-to-image processor. We have a Veras that we can see in Revit, where you can upload an actual model and not just generate an image based on text, but actually work on your model and ask for different design using prompts. And we have Maket, which can transform pictures and also do generative design. Turning over to you.

MICHAEL BURSHTEIN: So I will show you some example that we've been using at AGD. The first software, it's called TestFit. We've been using it for a couple of years. And here, you can see some picture of a study, apartment study that we did.

And some pros to the software, it's really a time saving. And it's good enough for early planning purposes, but it just produces typical layout. So if we want something more complex or unique, then we'll have to manipulate it manually. It's create the most efficient box it can fit on a site. You can see it as a pro or a con, depends on your point of view.

There's some concern about the layouts not being tested. None of the project that we did in TestFit, actually matured to a real project, so we didn't have the opportunity to real test it and see if it's work. And so those TestFit may lose the concept of relation to the neighborhood, to the client need, or community personality.

So another program that we just started using called Veras. And we use it for a marina study in Keego Harbor, Michigan. The client wanted a high-level render for a fundraising reason. By making a very simple, am just masses in Revit, and exporting this image into Veras and give it some direction that we want people, some trees, restaurant, the type of building, the day, and the season we were able, after some back and forth and changing some stuff, to get to this image.

During the process, we also changed the model according to what we were trying to get. So this is the final image from the Revit project. And this is what we got from Veras, which is pretty impressive. And it saves us a lot of time. We didn't need to program the building, to decide about the whole elevation, just to create one image.

Some of the funny things that we encounter during this process is, for example, the deck that goes nowhere or some floating island and some floating people. And this one that looked like there is a really big problem with the drainage on the boardwalk.

YAEL NETSER: So after seeing some examples of real-life use of these technologies in architecture, I'm going to present something, a survey that we did to see some statistics about this. So what I'm going to introduce is part of an ongoing research at CMU.

We can see the names of the research team here. This is a multidisciplinary research project. We are working from the Department of Civil and Environmental Engineering, Department of Architecture, and also Heinz College for Information Technology and Management. And the research is trying to ask what are the current uses of generative AI in the AEC industry, as a whole, architects, civil engineers, project managers, and everyone in between. This is funded by the Block Center for Technology and Society in CMU. And the title is Empowering Civil Engineering with generative AI; Opportunities, Risks and Future Directions.

So one of the first questions we asked in this survey, which at the moment, we have results from around 240 participants. We would be really grateful if you could also answer our survey. I will put a QR code at the end of the presentation.

So the first question was, how often do you use generative AI in your work at the AEC industry? And we can see here the professions of the people who filled out the survey architects, budget analysts, civil engineers, compliance managers, construction managers, cost estimators, and all the professions you can see here on the screen.

And we can see that almost all of the professions are using generative AI to some degree. Some of them much more than others. And we can see that architects are among the professions that use generative AI less. I think we can assume that this is because the nature of this technology, which is pretty new, only recently evolved to images and 3D modeling, which is a very big part of the profession. But we can still see that about 30% of the architects use it a few times a week or more, and even almost 50% use it more than once a week.

Focusing on architects, we are moving to a more detailed question we had in the survey, which was for each of the tasks mentioned below, how much do you think generative AI will be able to help you with this task? And it goes in five levels from strongly disagree to strongly agree. The strongly disagrees represented in red here, and the strongly agree is represented in green.

We have many tasks here, and of course, they are only a small part of everything that architects do. But I think that we can see, again, that a lot of architects believe that generative AI can help them to some degree with these tasks. Because if you look at the midpoint, represented here in a vertical line, then we have much more green on the right side than red on the left side.

But still we can see that some of the tests are seen as more fitting to using generative AI than others. So first, let's look at the ones that architects do not think that they are good for generative AI, or at least most of them do not think so. So these tasks, if we try to think of what is common between them, they are describing direct design, detailing, supervising.

These are tasks that are supposed to be very precise and have a lot of responsibility and accountability to them. So this may be due to the novelty of the technology, but it can also point to an issue of trustworthiness of the technology for architects.

On the other hand, we can see that many tasks are perceived as suitable for use of generative AI, and the top ones are create graphical representations, perform marketing activities, analyze costs, prepare procedural documents, contracts, disclosures, and incorporate green features into the design. So these tasks, architects perceive that they are good candidates for using generative AI tools.

We can see from the examples from HED and also from the CMU research, that as architects, we have definitely been starting to use generative AI more and more. And now we're getting to the questions, are we considering ethics while we do it? And before we discuss this question, let's introduce some terms of ethics in architecture in general.

Of course, ethics in architecture is a complicated subject with many fields mentioned in it. While looking at the AIA Code of Ethics and Professional Conduct, we can see that even in the first canon, in general obligations, social, and environmental impact are mentioned here as very important, and they are also mentioned together.

Looking at both of the AIA Code of Ethics and also at some literature, I collected the ethics values specific to social sustainability and equity, which is the subject of our talk today. So what we can find here, is respecting and conserving natural and cultural heritage, promoting health, safety, and welfare, transparency, public interest, advocating for the public realm, promoting sustainable design, fairness, and human rights. And as architects, of course, we are all familiar with these values and as ethical architects, we strive to promote them every day in our projects.

The question is, will AI be able to weigh ethical issues as we do? And when asking this question, of course, everyone knows the dilemmas of ethics in AI. For example, popular culture, science fiction, we can see from the movie I, Robot, which came out some years ago but still very relevant. And the problem there was that even though there were some ethical rules presented to AI, it was able to deduct some other rules and decide that for the good of humanity, it can harm human individuals, which is problematic for us.

And this is a simplification of the actual book, I, Robot, which was written even in the 1950s. This book contained several short stories, each studying an ethical dilemma presented to a robot which made him malfunction and discussing why that happened in logical terms. I really recommend reading that book. It's amazing how relevant it is even more than 70 years after it was written. So here is a quote from that book saying, "Reason is an excellent thing, but it has its limits, and such limits cannot be transcended by any means."

So let's look at ethics in AI. And of course, again, this is a very huge topic, and we're just going to touch it briefly. But I'm going to present some ideas from research. The question of ethics in AI became very relevant and kind of a hot topic, a few years ago when self-driving vehicles were introduced. And all of these questions were raised on whether a self-driving car should preserve the life of its occupant or the life of someone on the street jumping by and so on.

And some researchers introduced the idea of presenting an ethical layer into the algorithm which will contain all of the beliefs, desires, intentions, plan, library the algorithm will interpret it, and then it will take action based on the logic of the ethics in the layer.

However, a very interesting research I ran into called "The Dark Side of Ethical Robots" from 2018 by Vandereist and Winfield, did a very simple experiment and asked the question whether while introducing ethical robots, are we actually opening a door to unethical robots?

What they did here is they had the robot help a human make a decision on which cup to select. And at first they presented it with an ethical layer, and it helped the person move towards the right direction. But then they edited one line in the code, and the robot became competitive and even aggressive and started directing the human to the wrong direction.

And their claim is that if there is an ethical layer in the algorithm, then it becomes very easy to hack it and make it very unethical. So they raised the question, should we even have this layer, or should we just not trust robots with ethical decisions? And in addition to these ethical questions, we also have the question of bias, which is very important, as well.

I think this is a very common experiment that showed that there were extreme failure rates in face detection based on race. This research is from 2022, and it showed that only 50% of African-American faces were detected correctly compared to almost all of the other subjects.

And this is because the data that the algorithm learned from was not distributed equally. More than half of the frames in that specific model that they tested, contained Euro-American subjects. African-American subjects were only represented in 8% of the frames, 8%, and African-American females were even only 1%.

So how can the algorithm recognize well when it's not presented with the right information? And I believe that this question is relevant to algorithms who are going to learn about architectural design, as well.

For all these reasons, about a year ago, the Biden administration issued an executive order on safe, secure, and trustworthy artificial intelligence, calling for research and funding for development of any kind of projects who will promote new standards for AI safety, equity, and civil rights, protecting Americans privacy, and standing up for consumers, patients, and students while also supporting workers.

So we understand that this topic is very important and that our society is starting to deal with it. Let's see what are the ethical concerns for using AI specifically in architectural design?

So for architectural design, I didn't find any clear code of codes of ethics yet, either in the ALA or in Europe. RIBA has some initial thoughts but not something concrete. But I did found in research, things that are discussed for AEC industry and arts and may be very relevant for us, anyway.

So data privacy and bias are, of course, valid concerns. And when we talk about creative arts, we're also asking if we can balance the augmentation of human creativity with the preservation of genuine artistic expression. This is for all the arts, but it's still very relevant for architects.

We need to ensure accessibility and inclusivity of the technology. And questions about authorship and intellectual property become very big because these AI text-to-image generators mix and match a lot of things that they know from the past, and they may use parts of designs or styles of specific architects. Who has the intellectual property in this case, and who gets the authorship?

So we need transparency, and also, we need accountability because in architecture, a lot of our design has to comply with code and law, and eventually we need to sign off and be sure that it is safe. So we need explainable AI models providing clear insight into their decision-making process, so we can trust them. And we can see that in the results for the same survey in CMU.

The solution for all these concerns, as presented in research, is to involve diverse stakeholders, including community members, in the AI model development, to ensure technology aligns with the population's best interests.

In our survey at CMU, we also asked the participants to what extent do they think that these ethical and social concerns are relevant to the AEC industry. And we can see that for almost all of the concerns, we have more than 60% of green, which means that most of the people somewhat agree or strongly disagree with these ethical concerns, which are bias in decision making, loss of human oversight, and lack of transparency, data privacy, and security, intellectual property issues, job replacement, and made up facts and hallucinations.

So all of this information raises some ethical and social and policy issues. The image you see here on the right, is taken from a website called Maket that I showed before that can create designs, either by text-to-image processing or by generative design. And they actually have a section in their site for homeowners saying, stop wasting money on architects and 3D renderings. Why do you need them? Just come to us. We'll generate everything for you.

And this raises a very tough question. Can generative AI really replace human art, human consideration? What about professional integrity that we have as architects? What about our responsibility and accountability, our ethics and values? And to answer these questions, I think that we need to understand and present what are the comprehensive aspects that architects consider while designing a project.

And they are many more than what is represented in the generative-design processes. And for the generative AI, we just don't know what conclusion it deducts from what it sees. Can we count on the robots to weigh these questions responsibly and ethically?

So when AI learns from existing buildings, it can learn from past mistakes that we as architects actually fix them and don't repeat them in our new designs. What qualities will the occupants and society lose in the process? So let's look at the values that are being considered in the design process to try to figure that out.

We created a very simplistic graph of values divided by stakeholders, owner, public design team, current residents and neighbors and future residents and the public. Oh, sorry. And many values can contradict each other.

For example, for a project to be sustainable, it may cost more. And the architect may want to create an innovative project, but it may cost more. So the owner will be reluctant and also maybe the public will not receive it so well because they are looking for respect for history. As architects, we consider all of these things and make decisions about them all the time.

But who makes the decision? Unfortunately, a lot of the times, it's the person with the money. And what values will be considered by AI? Well, maybe they will be considered by the person who is operating the AI.

And here we have actually an opportunity. Maybe we can utilize AI to represent the public and use its powers to analyze texts so that we can ask the public a lot of questions and then summarize them using text generative algorithms.

Here are some of the values that we discussed before. We try to place them within the white circle. You can see the AI-led decision-making processes with values that are more quantitative, like financial gain, the development costs, energy performance, and so on.

But outside the circle, we can see a lot of intangible values that are still very important innovation, transparency, wellness, health, respect for history, equity. And it is our ethical responsibility, as architects, to be able to incorporate them into the design, as well. So we believe that it's very important to have a human in the loop for this process and to make sure that these values are not overlooked.

MICHAEL BURSHTEIN: So let's talk about how to prioritize social sustainability and equity. So what are social sustainability values are? We gather the values that are mentioned the most in the literature, which are social interaction, architectural identity, sense of security, flexibility, social participation, well-being and quality of life, social equity, and social capital.

I'm going to show two case studies that will demonstrate some of those values. Now, it's important to mention that those projects did not use AI. But we'll show the value that needs to be addressed in those projects.

The first project is a black-box theater. It's a very interesting project because it contains a lot of stakeholders with a lot of needs and wants in a very small space. The location is the University of Detroit Mercy, in Detroit, Michigan, in the lower level of the Student Union building.

So the goal for this project was to design a performance space for the Theater Department. Currently, they are using a space that is a driving distance, and it's outside of the campus.

The first value that I'll show is the social interaction, participation, and capital. We have three parameters that help achieving this value. The first one is location.

The project is located inside the Student Center, but it also can be accessed from the outside. So both give the student an opportunity to interact and to see and be engaged but also open the possibility for the community to come and use this space for performance and such.

Another parameter is the collaboration spaces. We added some spaces for student to just hang out, or collaborate, or even in participation in the shows in public speaking to do some improv groups and some singing.

So another parameter is views and connection. And this is a thing that was important from a programmatic aspect and to create interaction and a sense of security and the visual connectivity and also transparency.

You can see from the stage manager and also from the sound and light booth into the performance area. You can also pick when you are in the lobby into the performance area. And as you go through the ramp, you can see more and more.

The second value is flexibility. So as I mentioned before, there was a lot of needs and wants in this project with a very small space. So first of all, we had to prioritize what are the programs that really need to go in this space and then try to combine them and create spaces that can do more than one programs. You can see the programs. They are color coded by the spaces that they are in.

| one example is the stage and seating area. So a black box to, if you don't know, like it says, it's a black space that you can have a lot of options on how you want to organize the seats and the stage. So here, you can see some of the option, like a lecture type, or an arena type, a stack type.

This is the backstage, that during performance accommodate the actors for makeup, for costume change, dressing room, storage of the scenery. But during regular time, can be as a custom shop for repair and alteration, also for classes. And here, we combine the ticket office together with the concession and the queuing line, into the lobby area to create gathering places.

The second project is a new recreation wellness center in Redford Township, Michigan. Here, we work with the users and project stakeholders to develop a shared language to shape this project. Each individual and organization bring their experience, concern, and perspectives to the development of a building design.

In our role as architects, at least in part, to bring together this multiplicity of perspective into a single project, to make it custom to the identity and character of the Township of Redford, and to integrate it into the existing urban fabric, so it will serve the community and partnership for decades to come.

As the project developed, we are making sure that stakeholders understand the context of key decision, and the community stays informed by being transparency about how we make our decisions.

So the first value that I want to talk about is flexibility. For truly support a healthy community, you need to provide recreational space with flexible design to accommodate a variety of active lifestyles and community partnerships. Recreational design that promote health, at its core, is important to the success of a thriving recreational center.

More than just the inclusive of a generic, multi-purpose room, our focus on programmatic flexibility considers synergy between users. So you can see here the programs and how they combine together.

So in order to get this flexibility, we introduced some elements like a movable partition that can change the size of the room, no walls between program that allowed for overflow and for combine of activities, and also big opening on the first floor to allow program to use the indoor outdoor and to get more room. Another thing is an option to make the building grow without really interfering the current programs that are in.

Here, you can see the lobby area with the children place in the back. And here is one of the multi-purpose room with the big garage doors that can allow for the program to grow and to use also the outside.

Another valley is the social interaction, participation, and capital. We achieve that by transparency. Usually, a lot of the gymnastic activities doesn't have windows. But we purposely open it up and allow for viewing from one program to another. There's a lot of views between, and I'll show it next in some rendering how we can see between the programs.

Another parameter is the inclusiveness. We introduce some gender neutral and family restrooms. We expand the ADA. We make it more accessible for seniors and people with disability. And the third one is community collaboration by giving spaces to partnership and that you can rent.

So this is a view that you can see from the running track inside the gymnasium and vice versa. And you can see all the different type of activities that can happen here. And this is one example of cardio and weightlifting, weight area that is connected to one of the multi-purpose room and also to the hallway on the second floor.

YAEL NETSER: So we hope that when you saw all of these values incorporated in design by human architects, you can try to remember them and remember to ask the AI to use them and incorporate them into new designs. Because as architects, it's natural for us to use these values while we design for efficiency. But when we ask computers to do it, we may be more specific in our instructions.

So we would like to open the floor for discussion. Now, when we put some leading ideas of how can we use generative AI to help society in our design. Consider social sustainability and equity issues in your design. Sustainability does not end with energy efficiency. It has many other values that are just as important.

Make sure you ask the right questions. Keep humans in the loop while giving tasks to AI. Be aware of AI bias and ethical issues in decision making. Ask for transparency and accountability from the products that you are using and utilize AI for the public good.

Also, please consider filling out our CMU survey right after class. Here is the link for that. Thank you for listening.