Part 1: Option B
This lecture was very interesting as it was given in 2019, much before the AI boom that can be seen in the world today. The speaker focused on how, at the time, there was a large gap when it came to building information modelling and building information optimization or automation. He dubbed this space the “messy middle”. While a human can come up with multiple designs on their own, these designs are not often optimized to the level a computational machine can produce design alternatives and evaluate their pros and cons. Throughout the lecture he focused on analyzing different areas in the AEC field in which these optimization workflows can be or have been applied to create the best solutions for the clients.
One of the most interesting applications he talked about was modular construction. He talked about how other industries use robotics to optimize their workflows and wondered why this could not be implemented in the AEC industry. The conclusion found was that construction does not produce Toyota Camrys as each project is unique and not all projects are unique construction. So, while the whole project can not become a copy and paste project, there is some merit in modular construction when there are areas of the building which are repeated. With this modular construction idea, a QR code system was used in one project which allowed the construction workers to understand where each of the repeated parts was supposed to be placed specifically and aided in organization and speed of the project.
One thing that genuinely was surprising when watching this video was the continued emphasis that this increase in automation is not taking jobs from the industry. With the rise of AI/ML, one of the biggest fears of workers is a decrease in job security. Mr. Allen demonstrated in each of his examples that the machine was implemented to make the workers more efficient, not replace them entirely. One of the best examples of this was the automatic brick laying device that was created. While it laid the bricks quickly, the human was still on site walking behind the machine and ensuring the quality of the laying was up to far and fixing any mistakes. While I do think this is true to a certain extent, the implementation of machines such as fully autonomous construction vehicles will take the place of workers who have been formally trained to level sites and move soils for construction.
The tool that I find most interesting that I would like to try myself is the steel facade optimization tool that was used in the Harley Davidson project. This project had a certain budget that the designer did not want to exceed. Through this tool they were able to understand the number of linear feet of steel that was used in the design and understand how much steel was in each component as well. While this was more of a brute force tool, it took the guessing out of the equation when trying to create a design that met the design criteria but was also under budget.
Part 2:
The first point during this quarter that I felt like I was met with friction and felt limited in the current tool was when completing Module 3, the Give Me Shelter Assignment. This assignment was very fun because the tools that were being learned in class were being applied to a real-life situation, creating a bus shelter. I was able to manually flex my form and see many different design alternatives but I spent a decent amount of time changing my parameters and then re-running my code to see the new form. In future modules, it was taught how to create many design alternatives but specified ranges of inputs had to be given. An augmented version of this workflow that would have been beneficial would be to be able to give the program the form, tell the program what the purpose of the form was, and it be able to create design alternatives based on the knowledge of previous designs that are implemented in practice. For example with the bus structure, this tool would be able to size the dimensions of the structure to the dimensions of existing bus structures so that one could visualize what the item would look like in practice. This would be beneficial if trying to follow the status quo but if one was trying to create a product that was totally new, they may not want this augmentation and think that the point of friction found is ideal in order to be able to examine many different solutions that are not like the norm.
The second place I felt friction was when completing Module 6, the evaluating design alternatives assignment. When making the custom nodes, I followed the procedure from class and ensured all of my nodes matched the examples though one of my custom nodes still did not work. There were no error messages given in the run which was confusing and did not explain why no output was being given. In many programs, there is an AI chatbot on the side of the program which has been highly trained in the program and is able to answer questions related to the running of the program. While creating these optimizations is ideal, if they do not work as expected they are not useful to the client. Because of this, I believe that a chatbot where one can input the goal of their workflow and the chatbot can give suggestions on how to accomplish the task would be very helpful to users. This would lead to less time debugging and also teach the user that there are many different ways to accomplish the same task using the same program. I personally would want this augmentation because it would allow me as the user to perform tasks faster and more accurately which would lead to more optimized designs in a shorter time period. One may argue that by using it, the user is not learning the specific inner workings of the program but this chatbot is only giving suggestions, the user still has to implement the suggested workflow in practice.
Part 3:
A company that is diving into the intersection of machine learning and artificial intelligence is Autodesk. They used generative design to design their office in Toronto and through this project developed the refinery. After this, they started Project Rediscover to figure out if the dynamo and refinery tools could be expanded to a full architectural scale. Through inputting both human centered metrics to the program as well as quantitative metrics and constraints, the program was able to create thousands of alternative floor layout options and evaluate through the metrics which ones were the most optimal for the desired uses of the client. This is interesting because it allows the architect to strictly design for the uses of the space rather than guessing. Both the human aspect of ideas like workstyle preference were able to be paired with ideas such as daylight hours to create the most optimized system. The user was also able to set constraints on certain areas of the design to show where windows were or label what areas should be corridors to further optimize the layout for the user’s needs. This allowed the user to ensure that they had the best floor layout possible for the company. I do not think this application would have changed the way that I worked this quarter in terms of workflow though it may have expanded my lens of possible evaluation metrics that I could have used for my work in modules 6 and 7. This showed that while building geometry is important, ideas such as comfortability of the employees are most often more important to clients.
The next example of AI and machine learning being implemented in the field today is through a physical application called Hadrian X. This is an automated blocklaying system which can lay masonry bricks to develop the interior or exterior load bearing walls of a home. Once a building footprint is given to the system, the technology develops the layout of the blocks and then is able to lay 360 blocks an hour through the system. This only requires 2 operators making it efficient and giving it high accuracy and precision as all work is being completed by the machine. This is interesting because there is a large housing crisis in the United States so if more structures are able to be built at a faster rate with less human labor required, this will allow more housing to be available to the population. This technology is still new meaning that it is expensive and not common practice but with the rise of technology there is a market in which this machine will be very valuable especially if it can be scaled to accomplish projects beyond single family homes. This technology relates to the panelization procedure that has been used throughout the course. Similar methods are most likely used to determine where the different blocks will be placed based on the building dimensions. I do not believe the knowledge of this technology would have changed how I worked this quarter but rather shows an application about how what I have learned this quarter can be implemented in the real world.
The last example of AI and Machine Learning that is being implemented today is Delve, an urban planning and real estate platform that focuses on generative design for large parcels. This tool allows users to make decisions faster when it comes to the early stages of site planning including related to zoning, environmental factors as well as financial cost and financial potential. This allows the user to evaluate many different options and see what quantitatively fits their metrics best. This is solving the problem that typically when city planning, the designer is just looking at geometry and zoning and does not consider the users of the space. This platform gives a human aspect to the optimization as well as financial in terms of cost input versus cost output. This software expands the physical extents to which optimization can be performed and executed well. When thinking about how this would have changed how I worked this quarter, this tool would have been useful to implement in module 6 if examining the effect of surrounding buildings on the new building that was being designed. It could inform more about how the different buildings influenced one other in terms of sun studies but also informed what size of building would be the most economically feasible and make the most sense based on the current layout of building in the area.
Sources for Part 3:
Hadrian X: https://www.fbr.com.au/view/hadrian