Designer–Initiated Design vs. AI/BIM–Initiated Design

December 15, 2024

I have a problem with the terms ‘top-down’ design and ‘bottom-up’ design despite us all knowing what they mean and accepting the hierarchy they imply but, as long as design remains something that happens in stages in some kind of sequence, there’s nothing wrong with thinking of it as a process. However, top, bottom, left and right are just names and don’t necessarily denote a hierarchy or a level within one. I know architects like the idea of the design process being a hierarchy as long as they’re at the top but flowcharts have START at the top, bottom or side and it makes no difference. Buildings are constructed from the bottom up and no-one has a problem with the reality or the semantics of that. Even then term Design-led implies a leader so I’ll use the more neutral termsDesigner–Initiated and Information-Initiated instead. The best we can do for now is not further embed notions of hierarchy but, even so, it remains difficult to think of design as something not initiated by a designer and those perhaps subconsciously pre-processed ideas we call inspiration or creativity but on conclusions and patterns drawn from identifiable building information.

Designer-initiated Design 1

Many years ago now, I had an idea for a tower building with balcony-like shading devices organized according to a pattern that had nothing to do with the building’s internal configuration. I just thought it might be nice to have a building with balconies as a sundial with shadows on and around the building’s surface changing with the season as well as time of day.

There were other reasons.

This idea predated me living in Dubai but was always on my mind once I did as there were many towers around and keeping the sun off the walls is always a good idea in hot sunny climates. For a few days in the middle of Summer, the sun in the UAE is directly overhead. Any type of vertical grid sun screening such as that on the approach walkways to Abu Dhabi Louvre is 100% ineffective.

The diagonal spiraling arrangement of balconies is visually stronger than their horizontal arrangement by floor and this confounds the perception of scale. Any elevational articulation other than the tyranny of floor levels makes a building look more monumental – larger than it is.
In my initial study, identical balconies are equally spaced but incrementally displaced on each successive floor. These two levels of regularity generate interference patterns at the edges and (for me at least) that satisfying balance between complexity and simplicity that’s been taken as truth since the 1920’s despite no-one ever saying between complexity and simplicity of what.
Finally, when seen at speed from a passing vehicle, there is a momentary illusion that the building is rotating. This is of absolutely no use whatsoever but costs nothing and is something you don’t see everyday.

In the Fall 2019 semester I asked my students to design a 25-storey tower for a central Dubai site with 360° views. Interchange 2. You’ve seen the one. The existing 15-storey building is hard up against the interchange and visible from all roads approaching it. In other words, I was selecting the site to suit the project I had in mind and you can’t get any more top-down than that.

That semester’s project was the last complete one before Covid. It went well and teaching and learning objectives were met. I was pleased with how my project turned out, and I’ve written in detail about the internal configuration and layouts before. [c.f. Carbon Offset, Spiral Binding] As designed, the three unit types spiral around and up the building but their distribution can be optimized for ROI because view and its desirability will both vary with orientation and height. The building did everything I wanted. Scale was duly confounded and then confounded again by the camouflage and the irregular horizontal spacing and type of windows.

The trees are an affection shamelessly borrowing from Stefano Boeri but, and back to questions of scale, how big is a tree? For me, the important thing was to make a 25-storey building look monumental next to its 50-storey+ neighbors and, on that count, I think it was successful. As a cherry on top, I imagined the cylindrical daylight-visible LED signboard creating the illusion of rotating anti-clockwise. This was my nod to the famous neon the existing building once had.

There are plenty of interesting rabbit holes here, but this post is about top-down design vs. bottom-up design. True, I made this problem very easy for myself here as the idea for this building had little to do with planning, structure and servicing. The physical context was also simple in being one of roads and other towers but, now I think of it, my idea relates directly to both. I could say it came about from rigorous site research but it was merely an opportunity to use an idea from long before and, in my defense, it was for this site that I decided to use it. The distribution of units and their internal layouts have been designed to accommodate the shifting of balconies with surprisingly few variations, but all of this could have been solved by an architect-of-record and possibly better. Moreover, were this an actual project, the construction and detail drawings would then be created using building BIM primitives that were loosely assigned at the design stage but were now fixed. The consultants would use them for their own work, possibly even converting the BI into more specialized, dedicated design packages. Conversion errors would have crept in but would hopefully have been anticipated and compensated for. Finally, a small army of construction workers would have built the thing, starting unapologetically from the bottom.

Designer-initiated Design 2

BIM has been used to close the gap between design and detailed drawing packages. New York’s 8 Spruce Street building that was once known as New York by Gehry [c.f. Moneymaking Machines #2: New York by Gehry] is a good example. Because of the gloomy economic outlook 2006-2008, the developer decided to switch from condominiums for sale to apartments for rent and this suggested a floor-to-floor height reduction. A one-foot reduction over 76 floors gave another six or seven floors of apartments for rent (about 150) but meant the facade had to be redesigned to suit the new window positions. However, all this entailed was resetting a few parameters in the facade algorithm and sending it off to the fabricator once again. The design process never changed but the process of generating revised building information was significantly simplified and streamlined. All maybe Gehry had to do was approve the changes.

We’re no longer wringing our hands about the difference between algorithmic design and parametric design, possibly because both are used in much the same way, even if algorithmic seems to imply more sense of purpose. The left and middle sketches below are Zaha Hadid showing us her disdain for the tyranny of the grid, and on the right is Patrik Schumacher devising an “irritant” to kick-start the design process much like, one imagines he likes to think, a grain of sand can become a pearl. Whatever happens afterwards is someone else’s problem. I’m told that what usually happens is a bunch of senior designers pondering what packages and work sequence to use. The designer’s position at the top of the tree is maintained. .

Designer-initiated Design 3

In the past couple of years we’ve seen text-to-image and sketch-to-image AI significantly simplify and streamline the process of generating what are called design options for clients as if that’s all clients ever wanted. Here’s a selection of AI-generated tower images prompted, I’m told, by the ‘natural beauty’ of pine cones. Design may be reframed as navigating solution spaces but how to navigate that solution space and where to drop anchor are still decisions made by the designer. Rather than revolutionize the design process, we’re given these new tools to ensure the status quo is firmly maintained. “For things to remain the same, everything must change” goes the famous line from Giuseppe Tomasi di Lampedusa’s 1958 novel Il Gattopardo [The Leopard].

Information-initiated design 1

We’re no strangers to building information as the basis for building design. Many a design begins with either a development envelope or a massing model to ensure the project will actually fit some permissible building volume. Sometimes the role of the architect is limited to this and perhaps some checks for code compliance. Both tasks are easily automated. The image on the left below is what the architects gave to the visualizer. The image in the middle is the basics in place and the one on the right is the final product and all done within 48 hours using a “workflow” of Sketchup, VRay and Photoshop, apparently. This process isn’t even a collaboration – it’s just consultants doing their jobs, sending files back and forth along with the occasional email. Each of these images contains an aspect of what might be called architectural design but none can be said to actually be architectural design. On the left is development gain and on the right is perception management. This is fast-turnaround and possibly speculative work that was mostly rote, if not automatic anyway.

We teach students to work out a distribution of functions, areas and volumes first, and (at worst) then add some elevational treatment to make it look nice, or (at best) rethink the entire thing with all that in mind. We teach students about the Dessau Bauhaus building and imply that architectural design is about articulating functions with different masses and windows. Building program is one type of building information we continue to use to generate building designs, probably because articulating program easily provides us with a representational meaning we find satisfying. Or perhaps with one we easily find ourselves satisfied with. Maybe that’s all it need be. It’s done architects well over the years.

Other types of building information are equally capable of generating meaningful aesthetic options. The following is an example where program meets structure. It’s another teaching example that, although from a decade earlier than the one above, is a more useful illustration of a different way of initiating a proposal for how a building should look. The grey-green building to the right of Burj Khalifa in Dubai is the 2010 Rolex Tower, by SOM. It’s a mixed-use tower with about 30 floors of office space and another 30 of residences. It’s perhaps the most handsome building along Sheikh Zayed Road and the project I set my third year students was to design a tower for the adjacent site. The first thirty of so floors are office space serviced by the core on the right., The remainder of the building is four residential units per floor, with penthouses on the last two floors, one with its own swimming pool. Residential doesn’t require as deep a floor plan so six transfer beams are used to reduce the depth at the rear by about four meters. This was a basic design decision partially responsible for the form of the building. The building has a frontality to the street and, for whatever reason it’s there, the facade void adds a hint of anthropomorphization.

Faced with essentially the same problem but not wanting to do the same, I put the cores in the middle and truncated the corners, reducing the floor plate area by the same amount. I designed the units so that a 1-bed unit and a studio unit could be substituted for a 2-bed unit depending on the view in any particular direction and level. The mixed uses, the apartment layouts, structure, access and fire escape are all solved and, if you want, a price can be put on truncating the corners instead of doing the same as SOM. I prefer my solution with the living rooms in the corners because views along Sheikh Zayed Road are far more interesting than views across it.

All of this can be optimized by algorithm. There’s only one thing that can’t be – at least not yet. By truncating the corners I produced a ‘hipped’ shape I thought resembled the queen to Rolex Tower’s king, together dominating the chess board. I strengthened this association by making my building a bit shorter, exposing the structure as decoration, and making it rose pink to complement the grey-green of the SOM. These are all “poetic” decisions if you like, and we’re still a long way off being given a selection of them to choose from. There has always been a gap between development gain in the sense of maximizing the buildable volume, and perception management in the sense of making the thing look nice, if not meaningful. Mostly, the two are separate but in this case, floor area optimization and external appearance are linked. We just don’t know how.

The link between building information and building aesthetics is there. It done us well in the past, even if we’ve never been able to isolate and explain it. Discovering exactly where and how to link building information algorithms to ones that can extract and offer aesthetic options based on it is more likely to approximate how design actually works than the AI hocus-pocus we have now.

• • •