Yep. The "optimized" floorplans would be an order of magnitude more expensive to build than the original. I mean, drywall alone would take a lot more time with all the cuts, mudding and sanding more joints. The thought of framing out these walls makes my head hurt.
I'd least envy the people trying to arrange shelving in the library...
Still, interesting. I'd be curious to see results as he adds more practical constrains.
(I'm not even sure if it's real or feasible; I just remember a bunch of pop news articles with a video of a large CNC machine depositing concrete to construct a building.)
Building design is older than humankind, and there's tens of millenia worth of reasons why they are the way they are. But those reasons aren't always clear. Essentially modern floor plan design is the output of a black-box machine (human) learning algorithm.
The flawed output here, and the comments, help us point out what some of the missing variables are: egress, airflow, construction cost, etc. Do enough of this and you can get a vastly improved model.
...or maybe illustrate that some of the existing external constraints are stifling architecture as a useful art form?
Modern structures have to be fully designed before they are built, whereas most traditional structures were designed and built at the same time, and the design could evolve as needed to fit the environment. Now, we do everything with straight lines because it's easier to make plans and estimate materials and communicate with builders and file for permits and verify code compliance and so on, but you can do much more interesting and complicated designs if you don't have to communicate the design with a human at every step.
I think sometime in the not-too-distant future we'll have practical machines that can construct buildings designed by software to conform to the features and limitations of the building site and the desires of the future owners. If there's no need to communicate the design to a human other than "does this rendering look good?" then we don't need right angles everywhere just to make life easy for the draftsmen and carpenters.
This seems kind of related to something Christopher Alexander said about twenty years ago (which was linked from HN recently [1]), that current design and construction methods and business models have basically made good architecture well-nigh impossible to achieve on a wide scale. I don't know what he would think of using his architectural pattern language as a set of algorithm heuristics, but it's one possible way forward.
Genetic algorithms, indeed.
Eh, not quite: all of these "reasons" are adaptations to a specific condition. Many of which only exist since recently (for example, reinforced concrete enforces certain adaptations).
Evolution is not working towards one static best solution. It is always a process of adaptation of an existing solution to the current context.
Elegantly put. Thanks. I was struggling without success to make a similar observation.
Qualitative items such as the ones listed are important for customer focused environments, however, I'm not sure if AI can account for such factors.
This post is quite timely.
One of the nice things you can do with optimization problems is plug humans into the loop as oracles. Often, 'we know it when we see it', and we can do pairwise comparisons of 2 possibilities. So you can train a ML model based on win/loss comparisons and it'll learn to take into account the softer qualitative aspects via preference learning.
A recent example you might remember from the press: "Deep reinforcement learning from human preferences" https://arxiv.org/abs/1706.03741 , Christiano et al 2017 https://deepmind.com/blog/learning-through-human-feedback/ https://blog.openai.com/deep-reinforcement-learning-from-hum...
But also "Deep TAMER: Interactive Agent Shaping in High-Dimensional State Spaces" https://arxiv.org/abs/1709.10163 , Warnell et al 2017.
You can even pair it with EEG or brain scans for implicit ranking: "Towards personalized human AI interaction - adapting the behavior of AI agents using neural signatures of subjective interest" https://arxiv.org/abs/1709.04574 , Shih et al 2017.
There are solutions that I think show potential out there. I don't think our future AI designed world necessarily need to ignore difficult-to-quanitify dimensions like aesthetics. (Though amazon turk is expensive, especially in developer man-hours, so I can understand if that won't always be done.)
There have been recent studies about AI powered shirt design - the original input uses existing designs in terms of color and shape rather than the basic naive description of requirements that an engineer would give. Then the designs can be assessed by a review board or put up on a site and not produced until some n quantity of purchases.
You wouldn't try to detect cats in images without labelled data why would you try something MUCH harder without labelled data?!?!?!?!?!
It's a bit like saying that poetry will mostly be written by computers at some point in the future, or humans will mostly have sex with robots in the future. There's going to be more to this than whether the aesthetic experience is acceptable.
Also, the generated rooms look much like the sort of rooms one would build with cob. A cob schoolhouse could be awesome.
You need good sightlines for the kids, room for desks, and need to ensure kids aren't isolated in a corner. Plus light (not just to a small courtyard), ability to escape in case of fire or other emergency!!!!!!! and for minimal noise intrusion from other classrooms.
Minimal space and building material are just not the right constraints and speak to absolute ignorance of the actual problem on the part of the designer.
He even clearly states his goals: The creative goal is to approach floor plan design solely from the perspective of optimization and without regard for convention, constructability, etc. The research goal is to see how a combination of explicit, implicit and emergent methods allow floor plans of high complexity to evolve.
Nowhere does he claim this is an implementable solution, or meant to be one... just an interesting one, given a set of priorities.
As pointed out by commentator from the r/ML subreddit, the plans resemble biological arteries — which intuitively make sense given that the objective was to minimize distance to nearest fire exit. The analog is that arteries try to maximize surface area to volume ratio to ensure blood flood.
I too would love to see a 3D environment built from these, so we could get a feel for them.
Of course, if you designed and built such a thing all at once using a giant 3D printer, I guess you'd be building in an entirely new way, which would be completely different :-)
You'll notice as you scroll down that this article _does_ vary the parameters and weights. For instance, the one that optimizes for windows tends to create interior courtyards.
I could imagine adding parameters and weights for thresholds, areas for crafts, light on two sides of rooms, rectilinearity, water shedding, etc. etc. etc.
But I love the idea that what comes out the other end is wildly weird and yet functional, and feels organic and nest-like rather than square.
Perhaps there's wisdom in allowing cities to grow organically/chaotically after all.
It seems like a grid system is superior if you have no knowledge of the city, but a "organic" european layout is functionally superior assuming people always choose perfect routes.
I'd love to see these visualized in first-person, so I could walk through them and look around.
The main focus of my solution was to make the school a torus. Like many of my youth projects (writing a text adventure game...), I did not have the knowledge to solve the problems scientifically.
I'd like to make my point that any kind of over-optimization of one or more aspects will put one of the biggest virtues of truly good and functional architecture at risk: being adaptive for a wide variety of future changes (use, program, technology, climate, energy resources, partitioning, shrinking, expanding, etc.).
The actual task of designing a building is to find the right balance in a myriad of parameters, which sometimes create synergies (think sunlight and heating), but often are just one step away from undesirable impacts (think sunlight and overheating). Flexibility in architecture always results out of generous tolerances and robustness - which is always in danger to be eliminated by optimization for limited scenarios.
Also many aspects of the design logically derive from each other: if I plan a school with natural ventilation, it'd be probably a good idea to have windows in two sides of the room, that can exchange the whole air of the room in a 5min break. If I opt for a mechanical ventilation this advantage would be gone and the disadvantages of not having a more compact cubature would override and result in a completely different layout.
While I appreciate all kind of tools that give me an insight into complex interdependencies (how do floorplans with optimized A,B,C look like?) i think that good architectural solutions need humans to make a tailor-made decision based on a bigger picture of our society that has the chance to be valid for some decades(centuries?). Good architects choose to rely on typologies that evolved from history for this difficult task and transform them when needed.
I'd be curious if the approach of OP could be used backwards as a software based analysis what details make successful typologies actually successful.
On a side note: It's pretty interesting that the resulting floorplans of OP are somewhat similar to the traditional arabic city structure (google traditional damascus city center and zoom into the still intact quartiers).
Houses will go the same way as the tools become available. In the yacht business a new design took a room full of men a couple years to design and draw, so they were slow, serious business. Now it takes three guys at computers six months. So people expect it: a new design every year at the boat show. Go two years without and bad rumors circulate.
The idea was to have a function, for each office worker in the office that would grade the place they are sitting in (distance to toilet, printer, lighting, space around desk, etc.) and have algorithm evolve open space plans to have best overall satisfaction and also make sure that there are no places that have very low score.
Now, this algorithm could only take existing floor plan, it could only place furniture on existing floor with existing walls but it was still nice excercise.
https://scholar.google.com/scholar?hl=en&as_sdt=0%2C39&q=sha...
Chairs, tables, desks, bookcases, shelving systems, and all the things that go on top of them. Paper. Boxes. Books.
