Module 8 Questions:
I chose to report on the talk by Bill Allen on “The Future of BIM Is NOT BIM, And It's Coming Faster Than You Think - The Sequel” (option B). In this talk, Allen mainly talks about how, although BIM has now been adopted, it is still too passive and manual and has room for improvement. Even though firms now have detailed digital models, a lot of the valuable information is only viewed through rows and columns rather than in a more intuitive and digestible way, such as through EvolveLAB. Bill Allen argues that the industry is in a “messy middle” between traditional Building Information Modeling and a future of Building Information Optimization/Automation, where algorithms and computational methods take on a larger role. Instead of the designer manually drawing elements like walls or columns, a new approach is to define minimums, maximums, and desired outputs, then let the computer generate, test, and compare many possible options. This will explore many more options than a single designer can manually create. New methods include evolutionary problem solving and optioneering. Computational programs can also be used in areas other than design, like MEP and construction. In order to adopt these new methods on a bigger scale, developers must reduce “friction” with better UIs and more integration.
One thing that surprised me was the example where Allen’s friend, Ryan Cameron, used an Alexa to filter through design options. It was interesting to see a technology that is commonly used in daily life applied to design. I had never thought of that as a possibility, and now I wonder what other everyday technologies could be integrated into design programs. I was also surprised by the company M2.XAI, whose algorithm reportedly went through 1.2 million design and coordination iterations in 12 hours. That feels like a huge jump from the Dynamo scripts I have been implementing, even though parametric modeling is already much faster than manual modeling. Now I wonder how much more design capability I could have had using the power of AI.
Given the scale possible, I would like to try AI and optimization methods on smaller, low-risk parts of a project first, such as layouts or quantity takeoffs. I can imagine architectural design and building shapes being generated in ways that could lead to exploring really interesting results. However, as a structural engineering student, I would be hesitant to use AI for structural engineering components such as truss designs or column layouts because it could lead to catastrophic results if done wrong. Perhaps it would be possible if the inputs and constraints were extremely detailed and clear, but I would still be hesitant to use it until it has been proven reliable through real-life case studies and until exact details and procedures are implemented into design codes.
One moment where I felt the limits of the current tools was the amount of time computation took. For the assignment with two customizable inputs and twelve outputs, each run took my computer over half an hour. This made debugging difficult because even though I tested parts of the script separately and made sure they worked first, joining everything together could still create unexpected errors. An AI tool could have helped by looking at the overall layout and warning me about potential issues before the program was run, rather than only showing the individual node-based errors that Dynamo currently has.
Another limit I felt was with the mundane tasks that seemed like they could be much faster with AI. For example, I would test what would become custom nodes first in a normal script, but then when I converted it into a custom node, I had to rewrite all the inputs and outputs again as input and output nodes. With the power of AI and optimization, more possible designs could also be explored faster, so I could test a larger range of inputs instead of limiting the study because of time constraints.
I would want this kind of augmentation for debugging and expanding the design search, but I still think some of the friction was useful because it forced me to understand how the model was actually working. I don’t think AI should be used to generate the entire node layout, but it would be helpful to have a built-in assistant that could help answer questions because often my errors were due to a lack of understanding in using the program, such as where to save certain files and what nodes no longer exist or have changed since tutorial videos were recorded.
Genia: https://www.genia.design/
Genia is a AI program that generate structural design options from architectural drawings. According to its product page, users can upload CAD or BIM drawings and the AI program will identify architectural elements like walls and windows. Then, the program will generate multiple structural layout options based on the drawings. The generations are checked through structural calculations and relevant building codes. The program can even export these calculations and material takedowns. What makes this interesting is that it brings AI directly into structural engineering and claims it to be backed by physics and math rather than just being a visualization or concept. For my work this quarter, Genia could have changed how I thought about the structural side of parametric design by helping compare different structural layouts more quickly for assignment 7 when I was investigating structural layouts as an evaluation.
Smarthopper: https://smarthopper.xyz/#demo
SmartHopper is an AI assistant for Grasshopper that can read and interact with a user’s Grasshopper interface. Instead of being a separate chatbot where the user has to explain the whole problem, SmartHopper can look at the canvas directly and help with errors, organization, and other parametric modeling tasks. It can also connect to the McNeel Forum to find relevant community answers. SmartHopper is interesting because the AI is integrated directly into the modeling environment and is tailored to modeling instead of only giving general advice from outside the program. For my work this quarter, a Dynamo version of SmartHopper would have been extremely helpful when I was debugging my files. It would not have replaced the design process, but it could have reduced the time spent on technical confusion and let me focus more on the design.
Maket: https://www.maket.ai/
Maket is an AI floor plan generator focused on residential design. It lets the user describe what they want such as the number of rooms or square footage. Tthen the AI generates editable floor plan options that have accurate dimensions.The user can also specified desired finishes and renders for the floor plan design. Maket is interesting because it allows users without architectural or interior design experience to spatially plan potential spaces and designs. It lowers the barrier for DIY designs. For my work this quarter, specifically Assignment 6, Maket would have been especially helpful because it could have allowed me to use a more accurate estimate of usable or habitable floor area when estimating profitable space, rather than relying only on gross floor area. This would have made the evaluation more realistic because not all gross floor area is actually usable or rentable.