Getting There
First of all many thanks to John Hill who alerted me to Brooklyn-based practice SO-IL and three of their projects. One of them you can see here. https://archidose.substack.com/p/architecture-books-week-22025 The next day I came across another article here, https://www.theguardian.com/artanddesign/2025/jan/28/so-il-tankhouse-housing-new-york-big-apple, announcing the same book (which I’ve ordered) and another link to another project underway. SO-IL seem hot at the moment. The project I’ll go into in detail about here has an external access corridor as a kind of threshold space and this is a good development. All floors of this building aren’t typical but opening onto the access corridor on this level are doors to three units and these doors have sidelights. We don’t know whether these are clear or frosted glass or whether the entrance doors are glazed. Let’s assume they are. Two of the units have kitchen windows overlooking this corridor. One of the units even has a bathroom window overlooking this access corridor yet nominally separated from it by a balcony inset.
Granted, the narrowness of the site meant a dual-loaded access corridor site was never going to be an option so the non-rentable to rentable ratio will be high. It’s around 18% which is nowhere near the ruthless 11% of Lake Shore Drive. The architects are rightly proud that their new approach to housing is finding favor but admit that these developments are expensive. It’s just a fact that repetition lowers cost, as does shaving unnecessary area from wherever it can be. Any kind of wobbliness, misalignment or re-entrant corner is going to add cost and there are plenty of those. If we find all that beautiful or interesting then we need to remember it doesn’t come for nothing. True, incremental setbacks in line with the zoning envelope can add terraces and balconies can be enclosed or not to add elevational variety. SO-IL seem to be doing all this but whether it was entirely driven by code compliance and exploring minute loopholes, or whether to create a handsome building (which they have) we don’t know. Probably a bit of both.
As with any high-rise, the void between the public realm and the private realm on the sides facing the street is airspace. The Western norm has habitable rooms such as bedrooms and living rooms given the best illumination and, in tight urban locations, this means they face streets as the only open space likely to provide that illumination and that, also because of that, is the most viewable view. This is just how it is. In terms of its internal life, the building will be expressionless for all of the day and most of the night. An active facade of kitchens, bathrooms and access corridors simply isn’t going to happen on a street side in Brooklyn or most other places. Instead, the active side of the building with access, kitchens and bathrooms is on the other side of the building where, looking at the site plan, it is probably going to face something similar across the boundary sooner or later. For now, I’d like to not focus on building massing and finishes or even cost, but on the layout and the relationships the private realm of these units have with the public realm. On the street sides, that relationship is conventional in having bedrooms, living rooms and balconies. This could simply be the result of marketability considerations meeting building illumination codes.
The typical floor layout above shows two gates divIding the access corridor into three parts. I expect these to be full-height security gates with some sort of see-through grille that isn’t bars. One of those gates makes the northern third of the access corridor into a kind of front garden (with a table outside the kitchen window), while the south third does the same for the other two units. The middle third of the access corridor is for the elevators and fire escape stairs. The kitchen of the middle unit isn’t given a window, presumably because it faces the blank wall of the fire stairs but also because it would be passed at close quarters by the residents of the unit at the south end. Who might be looking into these kitchen windows has clearly been thought about. Situations such as these Japanese examples below where an access corridor passes by a bedroom window or even a kitchen window are avoided. London also has significant housing stock (of council/ex-council) flats with deck access passing by kitchen windows and secondary bedroom windows so, while this distaste of The Other passing by one’s windows has a cultural component, it’s not the complete truth.
In a managed residential building the middle third of the access corridor with the elevators is a kind of pseudo-public realm but still public enough to warrant security gates between it and the private units. All the same, the security gates indicate that it’s the space of The Other. Although this access corridor could have served as a physical threshold, the security gates make two thirds of it into a visual one in much the same way as the grassy area outside the kitchen window of Yamamoto’s Dragon Lily’s House tells people not to walk there. We have is the northern unit accessed by what’s essentially a gated private garden, and in doing so effectively replicating an aspect of a detached worker house of the 19th century. The same goes for the gated south portion of the access corridor. What we don’t know is whether these portions of the access corridor are managed space or whether they were sold as threshold like spaces for the units fronting them. It could be the case that the security gates were added at a later stage to increase the marketability (and the price) of the units but the scissor stair with separated accesses for north and south apartments seems to contradict this. If this “threshold” space was sold as part of the units (subject to certain conditions such as the fire escape not be blocked, no laundry be hung to dry there, and the space not be enclosed) then the non-sellable to sellable ratio is 17.5%. If not, then it becomes close to 25% which is very high. It’s a problem but it’s an interesting problem and it’s good to see some architects bumping up against it.
The part of the access corridor immediately to the south of the fire escape stair has two indents. Both will make the elevation more exciting I suppose, but the downside is that there’s now no space for a table like the north unit has. Enlarging the corridor would have made made that west elevation less eventful and increased the non-sellable/sellable ratio even more, so this decision not to have it makes me wonder what the priorities were. Is it a case of why bother making the space for a table if it has to be shared with another unit? Narkomfin this is not.
• • •
The other big piece of news this week was the release of DeepSeek and its now verified claim to have equivalent performance to ChagtGPT but using less expensive and less sophisticated and less energy hungry computational resources. This is good news, firstly for contradicting the existing mindset that throwing more resources at something will automatically make it better. “Scaling-up”, I think it’s called. If the fundamental logic is correct, then scaling-up might being proportionate or even exponentially greater benefits. However, if the fundamental logic isn’t correct, then it’s simply more of something flawed. Achieving equivalent performance with fewer or less expensive resources is something that industry, if not architecture, is good at. The history of industrial production is an endless sequence of things being done better and with fewer or less expensive resources. As with many things AI, it’s unclear (at least to me) how this equivalent performance was achieved without an equivalent expenditure. My guess is that it has a better model linking the data with the questions it might be asked.
Consider. As the saying “Like trying to find a needle in a haystack” implies, the example of a haystack is often used to imply a large mass of unstructured elements. Suppose, for some reason, our task was to find a piece of straw that was as close to horizontal as possible. If we don’t have a model for how a haystack is organized, then the alignment of every piece of straw will have to individually investigated. But suppose, from observation, we noticed that pieces of straw at the bottom of the haystack are more likely to be horizontal because of gravity and the jostling that occurs when a haystack is built up, and that pieces of straw at the top are more likely to have a more even distribution of alignments. A simple working model such as this is sufficient to focus our attention on the place we are most likely to succeed. This doesn’t need to be scientifically proven. The point of a working model is that it works. I suspect the real innovation of DeepSeek is something like this. It did have its beginnings as an algorithm for predicting trends in the financial markets. This is one reality that some clever people have been modeling as algorithms for at least thirty years now. “Computer trading” it used to be called.
Another example. Suppose you wanted to find out somebody’s password for something. You could assemble some vast amount of computing power and initiate what’s known as a brute force attack in that hope that, after a certain huge number of attempts, you will arrive at the correct password. This happens and has been known to work, but it is stupid. But suppose that, after observing the person whose password you’re after, you form a model him as “the kind of person who uses his birthday as his password” then you will probably be able to guess his password with far fewer attempts.
Okay – a third example. There was once a child who was to become a maths genius. It was probably Leibnitz because it usually is in stories like this. Anyway, he was at school and the class was being unruly so, before the teacher had to leave the class for a while, the teacher set the class the task of adding up the numbers from 1 to 100. I suppose there’s a version of the story whee Leibniz solved the problem before the teacher even left the room, but Leibnitz was definitely the first to finish.The other students began by adding 1+2=3, 3+3=6, 6+4=10, 10+5=15 and so on. Their approach was brute force number crunching. Using this approach, whoever could do it faster arrived at the answer sooner. What Leibnitz did was make a model for how the numbers were structured. It went 1+99=100, 2+99=100 … 49+51=100. There were 49 groups of 100 (4,900) plus the 100 (that was not used at the beginning) plus the 50 that remained at the end (after 49+51=100). Answer = 5,150. Because of the presence of this simple mathematical model for how numbers are related, the computation involved only 6 (max.) additions instead of 100.
Natural intelligence seems to be about finding better and easier ways of doing things. We might have thought what we have now is artificial intelligence, but it seems like it was artificial intelligence that wasn’t based on how intelligence really works. An artificial intelligence that uses a model of natural intelligence seems a better way to go.
• • •
Share this:
