Artificial intelligence (AI) in Architecture. What are the practical applications?

Artificial intelligence, machine learning are buzz words in 2017. Countless articles detail the massive and promising changes it will bring.

And indeed the changes are very fast. AI is being used in more and more professional domains in very practical applications. Journalists have now AI based bot writers. Policemen have AI based tools to help them in their criminal investigation…


But what artificial intelligence applications are currently developed for Architects? And what are the possible applications can we imagine in the near future.

Designing a building is a complex operation, but is also made of many simple and repetitive tasks that make our daily work as an architect. Just naming a few example of this “dummy” tasks:

  • offsetting this array of posts of 50cm because a constraint changed
  • checking a roofing slopes so that water get evacuated properly
  • Checking that no elements overlap another

What if we were assisted at least for those tasks by a AI based bot?

Let’s imagine the possible uses of Artificial Intelligence in the field of architecture and specifically in the context of the generalisation of BIM-Building Information Model- in Architecture. Indeed BIM bring standardisation and colabaration to a new level. It provide an unified IFC export language and attach semantics to 3D models that help go further in what a machine can do.

What is Artificial Intelligence AI

Articial intelligence for Architecture Design

AI based model checker

BIM model checking is the operation of verifying a 3D BIM model to detect defects :

  • clash detection : geometric collision between incompatible elements of construction
  • fulfilment of legislation for example accessibility by disabled peoples
  • BIM Syntax checking : a wall should be categorised as wall…

Some existing softwares are specialised in those operations. they allow to periodically check a 3d model against a list of user defined rules.

As far as I know they are not yet powered by AI but it seems sensible that more of more model checker softwares take advantage of AI for their tasks.

What AI can do ?

AI can learn from a database of many BIM models to detect automatically good and bad patterns. How to tell the AI that a pattern is good? Well perhaps asking the experts when in doubt it is was is called supervised learning. If a false clash detection is detected, the system will improve to learn from its errors

Those quality checking tasks are quite good to be automated. It remove some unpleasing work that machine are good at. It guaranty a final better quality and less uncertainties and risks in the site work phase.

AI based construction solutions suggestions

AI based architectural design helper

Architects, designers are proud to be creative people. And when talking about artificial intelligence, one fear that often arise is that the machine can take that very part of the job they like most : the creative part.

But we can also take more positively the rapid advances in Artificial Intelligence in the fields of creativity. It will allow to create new smart design tools that will rapidly overpass the old “non natural” softwares like Autocad, Revit… with which we used to communicate to create our architecture design.

That is exactly the viewpoint that Patrick Hebron in the excellent article Rethinking Design Tools in the Age of Machine Learning explained in full details.

He explained why old existing profesional softwares for creative peoples force creators to understand a non natural way of thinking (multi level menu system, many complicated features that need to be learned…) to achieve what he want.And if he do not want such complexity he is forced to use general public softwares. This last type of softwares is simpler, but force the user to produce a predefined type of work and disallow exploring new possibilities.

AI based creative tools and architecture softwares will overcome these limitations by allowing to explore easily the infinite possibilities while keeping the user interface quite simple and natural.

He categorize 3 differents approach that AI based creative software could empower.

Design through exploration

When designing a space, an object, a web site… we often have 2-3 constraints that need to be defined and have major impact on the final design. For example the angle of the slope of a roof, the width / Heigh ratio of a glass (as illustrated in Patrick Hebron article), or the size of the screen in web design.

Often we use our intuition to preselect a sensible starting point for this dimensions…[Work in progress]

Design by description

Conversational interfaces

Artificial intelligence for Urbanism and Cities

Artificial intelligence for Construction

Drone base robotic

A team of MIT and more and more people interest in using drones to build architecture.

Managing a network of autonomous drones that can work together is challenging but thanks AI it could be possible. Also the drones need to carry more weight that today, to be able to be used in real construction construction site.

Artificial intelligence for Housing

It is probably the most know usage of Artificial intelligence for broad public. Everyone has heard about devices like Amazon Echo or Google Home. If not we recommend you read recent Wired article on artificial intelligence at home comparing the two devices.

All these IOT (Internet of Things) devices use AI to understand the house habitants and control the whole house system (sounds, heating…) with the voice. Of course when no order is given to them, they can optimize the system. For example Google Nest sensor or Netatmo french equivalent can monitor thanks to a cam, the presence of user and predict their next arrivals to switch off the heating when needed sparing lot’s of energy. By the way the energy and money saving is one of their key commercial argument.

The house is a complex system that everyone has an experience with. Many things can be optimised, many automations can be imagine, to increase even more the level of confort and interactivity of the space.

The domotic the IOT for house ancestor is dead, like minitel french internet ancestor, or like first mobile phones. Indeed its approach was too narrow, too technical and was lacking of the unifying power of the platforms and of the ease of use brought by Artificial Intelligence.

I have asked a quora question :
What are possible applications of AI (artificial Intelligence) in Architecture & Construction?
so people from Quora can also suggest idea to enrich this evolutive article

Artificial Intelligence change the way urbanists can design and think about cities.

An app to intelligently detect risk on site works thanks to the analysis of construction site images.

Thanks for creating this post. I’m getting interested in this question from the other side - the software engineering world. I’m curious if machine learning is currently being used in any architecture or building firms.

I am both architect and software engineer, but not specialized in AI. I think AI is not directly used by architecture firms currently, but by the mean of software. As I tried to highlight in this post there are a few apps that emerged that claim to use AI. But I think it is still emerging.

AI could be used for very practical repetitive tasks in architecture or construction. But also for more architecture design related tasks. Like testing possibilities automatically and find the best one, the one that fit most to program constraints. This last use could save a lot of time and end with much more performant construction. It could also change the shape of architecture and generate a new style.

1 Like

Ah, cool! I am a software engineer as well. I work for – our core strength is machine learning, although I don’t do that - I do frontend/backend (fullstack) work. My cousin is an architect and he was describing the process of manually going through the documentation for each product (fire alarm, piping, wiring, etc) on a big project and checking that it is to spec. That struck me as an ideal machine-learning use case, if that information is available digitally. So now I’m curious :wink:

Do you find yourself spending more time doing architecture or software engineering these days?

1 Like


I am spending my time mostly in software engineering (frontend mostly but also backend) and entrepreneurship. Though I have finished last year a modern and ecological house in Britanny.

Good idea actually the automatically scan specs. But practically how you would imagine such a AI job?

  • A database of specifications from the one hand.
  • A list of specification for the current project on the other

This type of specifications in France is grouped in a document named CCTP, where we describe all the qualities of the elements of construction that make the project. This document is legal and serves as the main document in case of conflicts on the site work (more important than maps and drawings actually). So it is of the higher importance for architects, despite it is really boring!

But what I do not understand is how to deal with the heterogeneity of cases?
How could it work?
The engine analyzes databases of CCTP’s and tries to find a match?
But those databases are a mix of good specs and bad ones. How to know the text is the best to describe the project?

Some perspective about AI and architect profession :
“the same technologies that might appear to be threats to our autonomy as architects could, jujitsu-style, be turned into opportunities.”
AI could resolve some pain point of our profession and we should imagine what kind of possibilities it open…

Thanks for getting back to me. The house you designed looks really cool - thank you for sharing it.

So, here’s what I was imagining, based on what I know from speaking with a good friend who worked at a large firm for several years here in the U.S. Forgive me for going over information you probably already know - this way you can see what I know (and more importantly don’t know) so we can establish a common ground :slight_smile:

The two critical documents for an architecture firm overseeing a project here are a “project book” (or project guide) – I think this is similar to your CCTP in France – and the actual blueprints/plans. At my friend’s firm, the project book was a written document, often a PDF, containing all the information about the project. Once plans had been finalized and work begun, changing it was a big deal.

The blueprints/plans could be in a program like Autodesk, and usually were, but there was always a copy (often printed) as a PDF or image document.

Over the course of the project, contractors would send the firm plans for parts (say, a fire extinguisher system), so the architecture firm could compare the proposed part to specification. These plans were often emailed and always in a digital form themselves, whether as a text document, PDF with images and/or text, or an image document.

So the machine learning system I am imagining would essentially extract information from the project book and plan diagrams, and then extract information from specifications as they came in (a user would probably copy the relevant file from the email into the system), and compare the two to ensure they matched. Machine learning applications can already do impressive text analysis and image analysis, so this is possible.

A trained person would still have to review the work of the software, but it would save them time.

From talking about this with my friend more, it sounds like Building Information Modeling (BIM) software may be starting to cover some or all of this ground - I am not sure.

Does that make sense? What are your thoughts?

I am going to check out the link you shared now!

Ok, I understand more now. So your application will be more useful at the construction stage when all the drawings and specifications are somewhat frozen. And will compare execution plans provided by construction companies with architects specifications. Compare text with text, images with images, drawings with drawings. And take note of differences.

Yes, indeed that could be useful, as sometimes some changes are too small to be noticed easily. Also for example, if an outdated source model has been used by the constructor it would detect it as well. And of course, if the specifications are not matched. But in this last case, it is not always obvious to tell if they are matched or not. A different drawing does not mean that the feature is not compatible with what the architect planned.

@fmz is BIM manager and work in the Bricks project with me. May be you have heard about anything similar in the BIM ecosystem?

New archdaily article on the impact of artificial intelligence on Architecture and the impact on it on the work market in AEC.

While I was reading this (good) article, I remembering about an old video that an profesor showed us some years ago. This is an exemple of automatised architecture in every step of the design process. Some purists of architecture would say that is not “Architecture” but I think this is a quite good exemple.

1 Like

I love this image that resume quite well the AI paradox that could be very good at hard intellectual game (like GO, chess…), and could fall short at game used by human babies.

Another application to compare different drawings with AI :