Columbia GSAPP: Horizontal MSCDP: Lorenzo Villaggi

So thanks for coming. Um welcome to the fourth and final uh lecture of this CDP 2026 um lecture series. We call it conversations with practitioners. So our lecture series serves as a kind of introduction to diverse practitioners from across computational design practice. um offering a range of perspectives on design, technology, engineering and architecture. So far um our series we've had Conan Olsen, Noah Toyonaga, Luchia Rabelino, and Tiffany Sang. And to introduce um this week's guest, this is Lorenzo Veli. Um he's an architect researcher applying emerging technology for lowcarbon and habitable futures. He is an adjunct assistant professor at GSAP um and principal research scientist at the living. He specializes in novel algorithmic and datadriven design approaches, reusable design, intelligence, advanced visualization and sustainability. He graduated with a masters in architecture from Colombia GSAP earning the Charles McKim and Avery 6 awards. And recent projects include the NIS engine factory for Airbus in Hamburg, the Elkmar affordable housing district for Van Veain in the Netherlands. Um the Autodesk Mars office in Toronto and the embodied computation lab for Princeton University. Um and Lorenzo's work has also been exhibited at numerous international venues including the Venice Bonale MoMA and the center Pompadu. and he's offered several scientific papers on datadriven techniques and thought workshops on computational design. And additionally, we're able to announce today that um Lorenzo is also going to co-e the fall colloquium with Laura Kiran. So for the design for the CDP students here, you'll be spending a lot more time with Lorenzo next semester. Okay. >> Thank you so much, Katherine. Super excited to be here folks and getting to meet you at least briefly today. Um and I'll start with a little provocation or framing um of the talk. So as Katherine mentioned my work aims at developing scalable application of emerging technology but specifically in the context of sustainability. So why why is it important for us or for me uh personally um as designers and architects uh there's a lot at stake uh the built environment as we all know very well is responsible for around 40% of global carbon emissions and at the same time we need to accommodate the largest wave of building growth in the next 30 20 to 30 years. So recent statistics say that we need to add almost 300 billion square feet to the global building stock um uh in the next 20 30 years which kind of amounts to one large or entire New York City every year for the next 40 years. So how are these buildings going to get built? And more importantly what is the role as designers in all of this? what is our capacity to take on this challenges and what tools, systems and processes uh can we leverage to make an impact. So today I'm going to talk about a few use uh case studies or projects and I'll use them to give my attempt at picturing um this new kind of design intelligence uh that could help us tackle this challenge in new kinds of ways. And um this is a kind of design intelligence that I argue uh involves both natural intelligence, one that is local, lowcarbon and material driven as well as the best of artificial intelligence. One that is scalable and that can help artificial agents and human agents to design with complexity as well as manage and handle complexity. So start with this uh first project called living bricks. Um this is a firstofits-kind installation made of living and lowcarbon materials as part of an exhibition at the Pompidu Center in in Paris. So here we asked ourselves um what if architecture could not only be grown but also self assemble without the typical blueprints or drawings and model delivery that architects are accustomed to. So the material involved in this project is mcelium which uh I'm sure a lot of you folks are familiar with. Uh mycelium is the root-like structure of mushrooms and when combined with archicultural byproducts like uh corn stalks in this case they can be placed in any mold of any shapes and become a solid object for different kind of use cases. So this is considered a lowcarbon material and an alternative to carbon intensive materials used in construction like concrete or steel because it involves uh little to no energy to grow and it is made oops doesn't loop and it is made uh almost entirely of biomass waste which in of itself is a carbon sync component because it accumulated and stored carbon during its own life cycle. So in this project we tested two new approaches uh that leverage the unique properties of mcelium. So here we're also asking what what is the material intelligence with this material? What can we do that we wouldn't be able to do with other materials. So here we explore two new things. One is boweling and the other one is jamming. And their combination can open up new opportunities for new kinds of design and new kinds of rapid construction methodologies. So bi welding is the process that occurs when two or more mycelium units while they're still in their growth phase um and are placed together next to one another would would continue to grow and graft onto each other effectively creating one single continuous uh object. So this is a close-up photograph showing the outer skin of the mcelium uh of the of the skin of the of the mushroom that is grafting and connecting on to neighboring units. The other component as I mentioned is jamming. And jamming is a simple but yet very complex phenomenon uh that occurs when granular matter or microscopic uh parts like cryst salt crystals, rice or even sand transition from a state of a liquid-like state like things flow as if like it was water and they transition to a static or solid like state. And an everyday example that you might have experienced is when you try to pour salt or sugar uh into something coffee or your dishes. Sometimes those crystals are stuck and that is due to many different reasons like temperature, humidity, but especially certain kinds of compressive forces and the interesting thing is that the these compressive forces can be controlled and manipulated to induce certain kind of jamming conditions. So instead of being an unwanted condition like in the soul shaker, what if we could use this phenomenon, this physical property as a um form of design intelligence that we can manipulate for our own purposes. There was also another challenge with this exhibition which is the the sheer scale of the final installation which amounted to 4 m to 4x 4 m by 2 m which is like 13 by 13 by 6 ft. Um and in addition the the overseas location of the exhibit precluded any kind of uh full-scale prototyping here in New York City. So to uh to address this challenges we devised a set of a sequence of smaller scale experiments that we conducted here in New York. So we started with a smallcale jamming experiment where we poured our medium and opened the funnel of a sample container to see the resulting emerging structures. So here we deliberately controlled a few parameters the opening size of the container at the bottom the slope and angles of the side panels and also the way we're pouring the medium. So as you can see you know at the beginning as we pour it it behaves as a liquid but then as we control the angles of the slopes and the opening container. So indirectly controlling this compressive forces we can start inducing certain kind of jamming conditions. And I think this last one is probably the one that we were uh loosely satisfied with. So meanwhile we also started testing different components of the actual shapes of the parts and identified that slightly tapered cylinders uh were working fine for our purposes. Uh and then we started scaling up our experiments and instead of using little wooden components, we started using actual mycelium components and kept calibrating the container geometry and pouring procedure. Some of these were failing in uh miserably and others started giving hope that this experiment was actually going to work. And uh eventually start started performing very analog structural simulations just by sheer human force with my very weak biceps and it seemed like things were going to work. Um and at the same time and this is kind of also speaking to this nonlinear process of testing and experimenting and prototyping. We're also testing full scale mycelium units and trying to see if our assumptions of bow welding and growth would scale up as the components would scale up. So this was like a I don't know a 5 foot tall uh column made of like a dozen uh full scale parts and uh also this experiment started giving us a little bit more confidence that this um uh these ideas would work. So finally in Paris we scaled up everything without ever having done this before in New York and uh we built our full container/formwork poured all the units released uh opened up the funnel released the units a little arch started emerging we covered the entire formwork to let the units grow further and allow for the parts to bow weld on top of one another and eventually um started removing all the formwork and revealing the actual arch structure. So ultimately this was an experiment in a new kind of living architecture, one that is low carbon and that involve collaborating with a local and lowfi sort of material intelligence and computation. So the second project is uh called the embody computation lab that we um completed in 2017 for Princeton University. It's a robotics and fabrication facility for the school of architecture there. And in this project specifically, we've been exploring early AI applications for circular economy in the context of design. So I won't talk extensively about the whole building design and the concept behind it but I will focus on the facade which is where we tested um these kind of new workflows and ideas for circularity. So to support circular economy and design we looked at salvaged scaffolding boards in New York City construction sites that would otherwise end up in a landfill. And we were particularly interested in the materials natural variation and the inherent non-standard um features that come with when you work with natural materials, but especially when you work with salvaged materials. And we wanted to find ways to highlight these kind of natural variations and celebrate it. And one way of doing this is by sand blasting these boards. So sand blasting is the process of like shooting at high pressure microscopic medium against a surface. Uh and if you do it against wood, it removes the hard the soft part of wood and leaves the hard part of the wood intact. Um as you can see from the photo on the right. So that reveals intricate contours and textures. And the use of this material was also supported by a few hypotheses. uh one is thermal performance. So um we're thinking about an invisible jacket or pockets of air trapped uh in the surface in these micro contours and then two hydroperformance to shed water. So instead of sand blasting entirely uh the surface of the boards, we focused on the knots which is where the grain was most interesting and unique. And to do this, we leveraged u the state-of-the-art at the time in 2017 of computer vision models and trained a convolutional neural net to automatically recognized knots on woods in scaffolding boards. So after some initial training, the algorithms got pretty good at seeing where the knots were. And then after we created a custom robotic sand blasting machine to create these microcontours based on where the algorithm told us the knots were located. So unlike a typical CNC milling uh machine procedure uh milling procedure that indiscriminately removes material uh this is a process that selectively blast and removes material uh based on specific reasons and purposes. So in this video, I'm summarizing a little bit the whole process for the facade. We gathered data, photographed all 900 boards, developed a custom model to see where the knots are located. Uh sand blasted selective parts of the board, in this case only the knots. Um sorted, trimmed and um uh organized all the boards that we were going to work with uh based on tone and other data. uh came up with different kinds of logic and and strategies to understand how to place the boards on the building uh based on again tone and the amount of sand blasting and the patterns that the boards were revealing after we sandlasted them and uh and then finally um yeah installed all the boards on the facade. So ultimately this facade registers uh a story of the material from resource extraction to initial use as scaffolding reading from left to right to computational analysis of unique features to selective fabrication and to reuse as a building facade. And more fundamentally, this allowed us to develop a new perspective on buildings as [music] this kind of temporary formation of material uh materials, energy and labor and also connected to other formations before and after the life of a building. So the design registers at different scales. Um you see a closeup and then a little bit further back. And here are the boards installed on the building. Further back again and finally within the context of uh Princeton campus. I'll take a sip of water between one. So shifting gears as we transition you know in the as we transition from material intelligence to a combination of artificial and material intelligence here I want to start start to talk about AI for design exploration and trade-off navigation. Um so I've been working in this space for several years and a version of this workflow um is generally called generative design or multi-objective optimization which is a kind of methodology that combines different kind of technologies. One is design automation. It combines simulation. It combines search algorithms or evolutionary computation and all of this to semi-automatically explore design options while also navigating tradeoffs between conflicting metrics. And one recent example of this application out there in the field is a recent collaboration with airplane manufacturer Airbus. And this involved the design of a groundup building uh a factory for the Nassels of the A320 airplane and in Hamburg, Germany. And the cells are the enclosures of the airplane engines. So this project was located in one of the last remaining sites of the Airbus uh uh campus in Hamburg. And uh so the first challenge of this project was how to work with the site's unusual triangular shape which was um kind of made it very hard or impossible to fit one of their uh normative uh rectangular large buildings that they've been designing up until then. So we explored two potential strategies. Uh one was creating a large triangular building with a courtyard. So taking advantage of the triangular shape of the site and the other one was creating a very tight and efficient and smaller rectangular building. The second challenge was capturing um talking to all the engineering department heads at Airbus involved with this factoring capturing the complex production processes rules and constraints and adjacencies and flows between all the production stations. and encode them into a workable design model. So here you can see a summary of some of the programs and production stations and sequences of operations that are involved in the factory of the cells. And as a result of these challenges, we developed a procedural design model that uh could create both types of design families, a triangular one and rectangular one, but also include all the different production stations and rules that Airbus provided us with. Then we created 10 measurable goals uh for a good factory. Three of these goals were operational uh which meant measuring three types of production flows. Uh three goals were financial which meant measuring things like construction cost. Two goals were environmental. And then finally there were two social goals that meant and involved um trying to capture the experience of u employee work conditions and the experience of visitors in the factory. So for the operational goals, we measured the flow of logistic tasks uh that are shown in green, the flow of paint tasks that are shown in red and the flow of assembly tasks which are shown in blue. And using this kind of automated datadriven workflow that I mentioned earlier, we explored thousands of design options where each design was valid, meaning they all fit on the site and included all the required workstations and rules that Airbus provided us with. But of course, some were better than others at meeting this conflicated and complex 10-dimensional set of uh design metrics [music] or goals. And uh so we leverage different kinds of data visualization techniques to reduce the the complexity and the high dimensionality of this design problem um and map them. So map this 10-dimensional output space to a 2D uh more legible uh map like and in this case we used um self-organizing maps and identified clusters of designs that were similar based on certain [music] features and started using these kind of techniques to discuss with all the stakeholders involved with the process and uh trying to surface the types of design both from a qualitative aspect and from a quantitative and performance aspect. which ones were the ones that we would want to move forward with. It eventually narrowed down the selection process to three high performing solutions and identified the final candidate design which was great for production efficiency, embodied energy, daylight, uh employee work condition and average for construction cost. And all of this unlike its rectangular counterpart which was really good for construction cost but uh average or poor um on the other metrics. And again we worked then with all the other project stakeholders to validate eventually the final design selection for the best triangular design and further refine the floor plans. And in conclusion, uh, in addition to having greater production efficiency and being comparable to other to construction cost of other more typical Airbus buildings, these were, uh, we argued that these were better buildings um, better designs to work in and that are extremely sustainable uh, with features like natural ventilation. One goal of this project was can we run these factories throughout the whole summer without mechanical cooling. So natural ventilation was a key um priority. They also came with great daylight renewable materials uh including net zero concrete. And finally here are some rendering shots of the final design. Um I'll conclude with this final project. Uh this is uh inside walls uh sorry a little bit recently. It's still an active research project that involves uh using limited and multimodal data uh together with AI to see inside walls and promote circularity in design. So again we we've been focusing quite a bit on existing buildings for the past few years and the reason for that is again as mentioned before um the built environment accounts for roughly 40% of global carbon emissions. Um but the in other interesting uh thing to note is that reusing existing buildings is is u a scalable path to decarbonization and uh highly effective because it can reduce up to 50 to 75% um the carbon emissions that would otherwise occur if you were to do new construction. And just for context, uh here I'm reporting a recent McKenzie report that states that the global retrofit market will and could amount to $4 trillion by 2050. But reusing existing buildings uh can be complex and time consuming um and it comes with way more challenges and unknowns than working with uh new construction. This is another example from our one of our collaborators Hapold where um uh you know this is a building in London that shows that it has been renovated over six times over the course of a 100 years and you can imagine then the complexity and unknowns that that may come with. So in the most general sense, this project involves an exploration of new AI technologies uh that can help make it easier and faster to reuse existing buildings. Uh or in other words, you know, everybody knows that AI is really good at helping us generating uh new things. Uh but this research specifically is is an exploration on how AI can help us leverage existing things. So um let's consider a typical scenario of an office renovation. Um here the architect and owner want to achieve a lowcarbon footprint by avoiding new construction and reusing most of the existing materials and systems. So most renovation projects start with a laser scan uh which gives you great information about the visible aspects of a building but it cannot tell you about the invisible aspects of a building including what's inside the walls and the condition of the materials and systems inside the wall. So by combining limited and multimodal data like floor plans if available, photographs, GIS information and thermal imaging, our work enables architects and owners to see in sidewalls um just like a X-ray like vision. Uh and in other words, we can predict hidden systems and materials including their potential damage and the financial value of the material inventory. So in this case for example we can identify 48 linear feet of um of 2x6 metal studs and if any of them have kinks or defects, nine linear feet of PVC vent pipe, seven linear feet of galvanized steel for electrical conduits as well as 12 cubic feet of mineral wool for sound insulation and the presence of humidity or mold. So our case study involves a uh industry collaboration with a u large uh engineering and design firm that is working on a conversion of an office space to a data center that will include new social and workspaces. And one of their goal has been to reuse as much as possible. And for this case study, we we started by focusing on a very small part uh this this wall section um which we also made a replica of u that you can see with the photograph on the right. So the whole workflow um involves three stages. One is gathering the limited multimodal data. uh one is feeding this data to a large multimodal model and then finally generating an inventory of the predicted materials of the assembly of the wall through a knowledge graph uh kind of data representation. So to start we gathered limited but generally available data about the existing building uh and its context and this included stuff like GIS information. Um and this is um very powerful because it's generally publicly available and it gives pretty good context about what the building could be made of just by itself. It can give information like uh when the the year that the building was built, the number of floors, the building use and renovation history. So, in a way, one way that I describe this project is a a seasoned contractor that goes in a Brooklyn neighborhood and just by knowing where the building is located can probably already tell what kind of issues that building might have, what the facade is made of, and what the interior partitions are made of. So, this is loosely what this model is doing by just giving it GS information. We can also start adding historical floor plans if available, if the owner has retained these uh that can showcase overall layouts and additional details. So adding gradually adding more context clues about what could be inside the walls. We can start adding point cloud models from lighter scans uh that can reveal visible details of asbuilt conditions. And we can also augment this set of data with sensorbased readings such as sparse thermal scans and radio frequency readings. These kind of data sets can help our model um improve uh predictions uh especially in respect to things like MEP systems um and again yes improving overall accuracy of the of the predictions. So the pred the prediction output is uh represented as a computer readable knowledge graph as well as a human readable 3D BIM model. Um maybe some of you are already familiar with knowledge graphs but very quickly um a knowledge graph is a structured form of uh uh data representation uh that involves nodes and edges and these in our case are directly mapped to the components of the 3D model. So more specifically, each node represents a material or a group of materials or systems of the wall, whereas the edges represent hierarchical connections and functional connections between the different materials or systems. For example, the top node represents the full assembly. One level down, we have the stud frame as a whole. One level deeper, we can find the individual vertical metal studs, and so on and so forth. Um again very briefly how does the LLM part work? Uh this involve the orchestration of a few agents [music] that work in sequence um and have all the necessary background information to produce structured outputs. With each step the generated assembly graph is further refined based on the insights and the type of modalities of data that we provide the system with. The work involved performing a series of ablation studies to understand the sensitivity of our predictions in relation to the types of uh modalities of data that we were providing the system with. And this is a very quick brief summary of our studies that kind of hints at how low quality but diverse modalities of data instead of high quality data of the same type can improve overall model prediction accuracy. So for example uh if we include only JS information uh the model has very limited context clues about the wall and will return a typical exterior wall made of uh drywall CMU blocks and brick cladding. But if we start adding other kinds of data like floor plans, photos and thermal imaging, this helps the model grounding its analysis and starting fairing uh things like stud location and even outlet boxes. uh more expensive data like lighter scans and radio frequency readings can help the model find additional elements uh like electrical conduits and duct work and of course is if we add thermal imaging the confidence and accuracy of the pred this prediction starts increasing as well. Um as mentioned before this area of research is still very active and we are testing the workflow across uh different cities. uh we started with Boston uh and moved to other location uh in North America and uh overseas. And this table is showing again an upward trends um upward trends in accuracy predictions as we mix different modalities of data. And then finally um our workflow um from multimodal data to graphs to 3D models is being tested and showcased through this interactive applications where visitors can select different combinations of multimodal data and see the type of output that our model generates whether it's a knowledge graph or the 3D representation on the right and additionally this is how it can be experienced on the field uh using a tablet loaded with our prototype we can start viewing in real time what may be uh inside a wall. Um our prototype eventually culminated also into a physical exhibit where we showcased um the capabilities of of this idea and this prototype. Uh we designed this exhibit to showcase um more uh more generally our approach uh to uh AI for net zero buildings involving all these steps from prediction to inventory to designing lowcarbon assembly using existing components. And here are a few pictures of the installation uh that was part of a big tech uh conference. And uh yeah, I think this is my last slide. End it here. Thank you so much. [applause] Okay. Everybody, anyone have any questions ready? Renzo, >> uh, I was wondering if you could talk about how you came to the the tapered cylinder form for the mcelium bricks. Like why choose that shape over, I don't know, maybe standard brick? >> Yeah, that's a good question. Um there were many factors involved in that decision that I can't fully recall but one of them was cost uh is always a driving variable and the tapered cylinders come from plastic cups. So that was easy to prototype with. It was relatively cheap and um and then it turned out uh but besides the cost component it turned out that softer and smoother edges were better than harder edges. Harder edges as you poured all the components together they could break. They would become more fragile and brittle whether something as close as possible to a sphere would perform better. The additional um advantage of using micylin for that kind of purpose it was because can be squishy. So unlike jamming structures with hard parts uh which a lot of other researchers have for um using other materials the advantage of my was that it would compress and would adapt to the compressive forces providing further structural stability. So uh the cylinders slightly tapered cylinders worked really well for that that use. >> And uh do you see like more room for optimization in the form? Like if you were to do it again, would you try to change the form if you had the budget? >> Oh yeah. Yeah. Yeah, for sure. Especially if there's a way to combine uh to make mix materials, you overcome some kind. For example, if you start making uh shapes that were deliberately designed to connect with each other, um certain kind of hooks kind of designs or uh star- shaped designs. Um maybe a mix of wood and my uh I could see that being a very scale and make this probably more applicable for real world. I actually have two questions. One uh probably follows codis. Uh I just realized that. So uh when looking at that project, would you say that like however you throw those pieces, you end up with a cutenary curve kind of structure no matter what or like do you ever get a different result? >> That's interesting. Um, in this specific case, we always ended up with arches, different kinds of arches all the time, but there were always these kind of like >> um arch-like structures. >> Have you tried like measuring them to figure out what kind of arches are like? >> Trying what? Sorry. >> Like measuring them like >> measuring >> measuring like to figure out like what formula would fit best for that one. >> Oh, no. But that >> is it like the same or like different actually? >> Yeah. Sorry, could you repeat the >> I I mean like you know how you go you have like y equals x uh squared. >> Yeah. >> So you have like that kind of formula for catinary curve and then you can like figure out if every time you end up with a different version of a catinary curve where you simply zoom in or zoom out. So that's why it looks different, but you're basically just zooming in and grabbing like a a smaller portion of it. So that's why it looks distorted. >> Yeah, that's Yeah, that could be a way. I mean, there's definitely a lot of space to introduce a a digital component >> where there is um the aid of computation to start predicting maybe how these materials may >> um may behave. Mhm. >> Uh we took a very analog route because there was just so much uncertainty with how these thisian material would work. So even trying to even if we wanted to rationalize the whole >> arch with functions or kind of geometry uh equations like you're mentioning there's still the real unknown of will the material actually behave in the way we uh think the model visually will suggest and >> um >> for us the first step was figuring out how it would work physically first but I think the next step would How could this be encoded into software? And what would be the variables that we need to take into consideration when we work with a material that is squish squishable and soft and might change a little bit the >> the kind of expected outcomes of an arch. I can I can imagine if we used a different kind of material we would get cleaner >> arching >> but um this would not the case and even keeping the same kind of conditions >> and not changing anything and repeating the same experiment didn't yield always the same kind of results. So there was always a a component of unpredictability that made us extremely nervous every time we did it. >> Um till the very end. But that was part of the experimental approach with with new materials. >> I mean, honestly, like the jump from the small scale to the large one, like it felt like such a jump. >> Yeah, [laughter] it was. >> It's crazy. Yeah. >> Yeah. Yeah. Yeah. It was uh Yeah. I was I was uh nervous. >> Very brave. [laughter] Thank you. And the second one, so I'm doing a project about um point clouds and how you can identify objects from the point cloud. So like here's a point clown and here's a start and here's a window. So have you looked into that kind of approach versus scanning the space uh with a term? There's a it is a very active area of research >> extracting intelligence from >> so there's plenty of research papers rec so yeah happy >> yeah thank you >> hi thank you so much for the presentation um I think in all of your projects. One thing that I really appreciate um is the way you represent your workflows and um like even the predictive workflows was super interesting. Um like transitioning I mean uh starting this program this summer I think that's something that I'm trying to familiarize myself with um kind of thinking in a more uh um in a way like that. Um I guess my question is can you speak a little bit more about you know creating a workflows in the beginning and like you know planning versus experimenting and like you know like how important it is to stick to your workflow or like finding things along the way. I don't know it's like you start with something and you don't like in a lot of the classes I feel like you just start something and you don't really know what's going to happen next and yeah so I don't know if you can speak a little bit more about that. >> Yeah thank you. That's that's a great question and I have no answer to that. Um except for um yeah the workflow that I've been showing they're more of a they're both a guiding map as well as a a a device to tell a story. And for me personally workflows are a way to start devising a plan at the beginning. Um, and what is useful to think about when you start thinking about a plan is to not um be too attached to it and be open to things being surprising to you or things changing over the course of your research because what you know at the beginning may not be the same of what you know throughout your research project or at the end. So the workflow is really just like a your best guesstimate of what you'd like things to play out, but then being flexible enough to like change and modify the the map or the steps that you want to go through as you investigate more your your topic of um the other advantage of laying out the plan is um sharability with other people. Can you what is the best way to share information and research and knowledge to other folks so that they can replicate what you're doing and build off of what you're doing? So that's also another reason why uh making very explicit your steps in a very research and scientific way I would say um is is valuable so that other folks can interpret what you're doing and build off of what you're not sure if this helps but that's how I work. Yeah. Sorry, my question kind of like feeds into Akquila's question. There's there's seems like there's a lot of uncertainty like in the process of a project like that and sometimes uncertainty can be like not motivating. So how do you just like convince yourself to move forward even when uncertainty is like very prevalent in a project like that? Like is there any >> um target that you set that you want to reach and is that target um based on form or based on like um a conclusion. Yeah, that's a that's a interesting question. If I if I understood correctly, how you're asking how how do we how do we engage with uncertainty and how do we move beyond the moment where or all the different moments where we don't really know what's happening or Yeah, I mean that's that's a good question. Um there is something exciting I think about uncertainty and the unknown which is um which is I think it's an important ingredient for novelty and for innovation maybe uh I would say and and research uh the best research is the one I think that is exploring uncharted territory and often u some of the work that we've been doing here. It does involve taking some some risks and um trying to come up with our best guesstimates of what could happen and learning one step at a time what uh is unfolding before us and taking that thing into account to make one additional step. Um it's yeah it's like [music] walking a room with no light and you have just a little flashlight that you slowly carve out your path knowing that that is just one of the many possible paths that you could have taken. But as collectively more people are investigating this kind of like uncharted territory start illuminating more areas of room and start building more confidence for more people to jump on research which is why I think you know like we've been doing work with my for some time but there are whole programs now dedicated to like uh architecture and mycelia that it's a very excit it's still a very uncharted territory but we're seeing like big momentum in biomeaterials and specifically Again, I'm not sure if I answered your question. >> No, that that was great. Thank you. But that's how I deal with a certain >> Thanks so much. Cool. >> Thank you. [applause]