AI BIM. Can ArchiCAD be ready for that?

[Podolsky]

Hi there!

I have decided to start a separate thread about Artificial Intelligence in AEC. I hope here we will able to collect as much information as what happens today in the AI world, show examples of using AI in BIM programs and programs driven by AI, and discuss what architects and other AEC professionals would like to expect from it.

I have a very strong opinion, that these technologies will come to us very, very soon (because AI already exists in many things we are using every day - like search engines or digital photo applications) and this might be a very interesting subject to review.

Please find below a scheme I have prepared, showing how AI for BIM might look like.

Building elements (as we know tools in ArchiCAD) are controlled by placement algorithms, that coming from building classification databases). For example - placing partition walls in the office with the right chosen sound insulation, fire ratings, correct corridors lengths, fire escapes, etc.

I/O engine responsible for Input / Output - but in architectural terms - automatic drawings generation and publishing, remote communication, including communication via e-mails, teamwork, IFC exchange. It is something like a secretary-robot, that supervising the BIM project.

Language analyzer transforming languages into commands, understandable by the system (software). It can understand human language and communicate with him as a chatbot, it’s also can understand different languages - like Python, JavaScript, AutoLisp, etc. Even read IFC (because IFC is a script).

Physical simulations help to improve correct element placement. Simulations shall be 100% on physics (more like physical engines in 3D animation software). Includes loads, earthquakes, heat distribution and loss, fire spread, wind load, radiosity, and photon tracing, similar to Monte Carlo.

Additional block called construction simulations helps to represent the construction process, including delivery, animation of cranes and installation process, construction timeline, and similar.

Each building element has two additional layers - assembly (if it’s a wall, then it might be studs, cover, insulation, and brackets) and behavior. Behaviour is connected to both physical and construction simulations.

[Bricklyne Clarence]

Podolsky wrote:
No. I'm 1000% sure - errors I mentioned not because everything needs to be done yesterday. Because workers I've been observing just [censored] stupid. I cannot find another word for that.
And the worst - after picking up such an error - people are lying in your eyes, that everything fine. Well, maybe because of London... People lost their mind here. Seriously - what I have seen in past 6 years - shocking.
So first they are producing a lot of really-really bad construction documentation, after there is coming conflict with clients, then coming lawyers and solving conflict, and after another team receiving this terrible corrupted documentation and fixing errors. Most of big buildings (all this glass skyscrapers in the city) are built this way in London. Through the scandals with architectural firms.

Definitely computer program can do much better, than most of architects do projects today. C'mon - there are examples appearing already of AI driven architectural programs. In next few years we will see more and more - just necessary to look into right direction.

Literally everything you've pointed out here is down to human error(s).
Not down to some shortcoming in the software they use (be it the lack of AI or AI-assisted algorithm).
And not even problems that are likely to be resolved by "replacing" those humans entirely or otherwise by AI software.
Who's going to control the software and drive it? Humans or machines? Who codes the software? Humans? Do you see where I'm going with this?

At the end of the day it seems as if you or the people who work with have more of a problem with the people they hire rather than the software they use, and that the hiring process is where the solution to like, 90% of your problems, might actually lay.

"AI" can't cure or solve what are at the end of the day human flaws and shortcomings, and we are by nature (and some would say by design) a flawed and imperfect species.
"AI" (and AI-assisted software) can mitigate or perhaps reduce the impact of some of those flaws, but we'll never live in a world where software algorithms completely eliminate the human factor and human error from things we do.

As an aside and an example (and to make the point), Facial-recognition software is now known to have severe limitations when it comes to distinguishing and recognizing distinguishing traits in non-Caucasian/non-white faces.
Not because the software is racist. But rather because the people who wrote the software (mostly white people), unbeknownst to even themselves, wrote into the code their own implicit biases and their shortcomings in recognizing features in other races,...into their software.
(And to be fair we ALL have these shortcomings when it comes to recognizing and scrutinizing distinguishing features in faces of people of races different from our own. That's just how nature and evolution works.)
Which then becomes apparent in how the software runs,....less than perfectly in a situation where it's supposed to,....as you would have it.... "eliminate human error" in doing the same - when all it ends up doing is not only perpetuating and replaying those same human errors (written into its code), but only more efficiently and possibly faster than humans can.
That's not a solution to a problem.
That's just making the problem more efficacious and streamlined.
The facial-recognition software and algorithm problem is an easy one to fix. And it's a fix, ironically that involves the addition of more (read : diverse) humans into the software writing process.

If you're incapable of seeing that the limitations of the software will always be delineated by our own limitations as human beings (whether as the ones who write the software or the ones who drive and use it), then you'll always be taken in by the enticement and temptation of the fantasy of a world in which software and AI solves all our problems.

As an(other) aside, you can't possibly be always sure (...like,..."1000% sure") that everything you say it right or correct, and that what everyone else says or recounts as a rejoinder is wrong - even when their recounting things that happen in their own experiences or from their experience and that you have no notion of.
Life just doesn't work that way.
None of us is ever correct 100% (let alone "1000%") of the time, while others are always wrong.
And not being open to seeing things from others' perspective or points of view, isn't conducive to dialogue or discussion.

Again, just my opinion.
You may see it differently,...and you probably will and do,...and that's okay.

I'm just saying.

[Podolsky]

Okey, why I'm so sure about I'm writing here and what I'm expecting from the future.

Currently in the most countries in the past 40-30 years AEC industry changed so much, that we actually lost the meaning who architect is. It still exists in education (as a theory) and in different qualification descriptions - that architect is actually becoming Chef Builder through some lifespan, but in reality nobody cares. Design and Build - this is the top and a dream of most architects of older generations (like was my father). Today architect will say about lawyers, that he cannot take responsibility for structure, the future building is sliced like a pie and each part of it is responsibility and insurance of someone else. Like windows - responsibility of manufacturer and if you decided to use two different windows company in one building - most likely non of them will take responsibility or provide warranty.
In BIM world happens the same - different disciplines care only about their parts - ownership and copyright and responsibility of their modelled walls of placed steel beams. We read nice articles and see videos about successful collaboration between contractor, engineers and architect, but in reality this successful collaboration does not happen - because nobody really wants to collaborate. At the end contractor never trust drawings he receives and built it using his knowledge, checking errors, making changes, hiring engineers to check and re-design various parts of the project and finally tells to architect, what to show in the final set of drawings - just because by the law you cannot build without architect involved.
In most cases BIM (even when it's a great technology with a lot of possibilities) becoming fiction, that made life of most AEC professionals more complicated, and again - they are using it (or trying to use) just because law and government programs forced him to do it.
I think they whole AEC structure is seriously corrupted today. Every time when I think that all subcontractors of Ladenhall Building lost money - I think, then what all that about? What kind of dystopia it is - to pay money for work? And of course corrupted systems do not last long.
My expectation that AI will start to re-organise the AEC industry (because BIM didn't) - because computer algorithms will be able to make buildings as it suppose to be. So if computer will show better and faster result then human driven company - then developers and contractors will start using AI instead of inviting architectural company on board - just to avoid headaches - and will solve legal part of architect to be involved into project different ways. Hopefully lawyers are always creative.
Very possible that architects - who will like to stay on the market - will move into design and build business concept (because this is what architect must do).

I also would like to add - people are working on AI for AEC already, in past few years several AI driven software already appeared and will appear more in upcoming 5-10 years. It will be mind blowing for the most, but we will get use to it. Just now very right time to think about all that and prepare to big changes.

[Podolsky]

Interesting development (maybe video is not so presentable). AI driven MEP:

[Podolsky]

In presentation they mention, that they are using AutoCAD and Revit only as input / output tool. The AI algorithms are running independently. Also they are using servers to process - that mean the whole process can be completely autonomous.
I'm sure ArchiCAD also can be used for such tools. Just the weakest point of ArchiCAD at the moment - it's more UI oriented and requires human input. There is currently no set of commands, that all can be entered in terminal to make the models, as it can happen in AutoCAD using only command line. So, for human being ArchiCAD is much better to work with, but for server based robots software that has similar structure like AutoCAD - it's the best tool to generate CAD / BIM information.
Of course exists ArchiCAD Grasshopper connection, and I guess Rhino and Grasshopper can run autonomously, just a question - can ArchiCAD handle completely autonomous mode, will Grasshopper-ArchiCAD live connection stay stable during hours (or maybe days) of work.

[Laszlo Nagy]

Can I ask and can someone explain to me what is AI about anything in this video?
This is simply an algorithm with many parameters and variables. If this is AI, then most GDL objects are AI, as they generate countless variations of geometry based on user input and parameters.
My understanding is that AI makes decisions on its own which looks like human intelligence.
I don't see anything in this video that does that, anything that is not determined by a human programmer.

I think this would be AI if the program would place the variant it thinks is the best without asking a human and it would be confident in its decision. So, the AI would have to be able to have its own viewpoint and would have to be able to form its own opinion.
Also, my understanding is that AI involves Machine Learning. I don't see any learning on the part of the program in this video.

Emre wrote:
Just to add to what Podolsky is saying, here's a quick video of the tool in action:


Whether or not this is well implemented in Revit I don't know, but from a first look, it seems like an interesting tool that would be certainly of use to bigger practices, and perhaps even small ones. Not saying this is the most important tool needed in AC, but it is a very good example of AI implementation.

[Miha Nahtigal]

It is not an algorithm with many parameters and variables.

With GDL you specify your parameters and an object is generated. Eg. for workdesks arrangement... you specify your desired grid (4 desks in two rows), rotation and insertion point in a room and that is it.

Generative design works differently. You specify initial input (I want 8 desks in a room), constraints (room space, distances,..) and scoring points (lightning, distance to exit, views...). Scoring is the most important part of algorithm.

Algorithm than places desks (starts semi randomly) and scores the room layout. Than starts a new iteration based on previous highest scored design (tree structure) and than tries to improve its score.

After a few million iterations it displays final variants from selected number of branches.

[Laszlo Nagy]

Yes, that is Generative Design.

My question is about AI.

[Miha Nahtigal]

It is more or less the same in this case.
Machine learning (generative design) is a part of AI.

[jl_lt]

i think what mr Nagy is asking is that, no matter how complex the algorithn is and how many millions of iterations the computers does, its just that, an algorithm

computer algorithns are normally a blackbox process for non programmers (we as users dont see the internal workings), but as i understand algorithms, they are just a series of steps to acomplish something. So, if we could see the steps we could determine the same output as the computer following the steps (it just would take us much much longer).

so in that sense, a patrick schumacher blob is not more parametric that an ArchiCAD window object. and BOTH requiere user defined input.

as stated in another thread, the problems mr podolsky describes are not inherent to computers but more with human processes and input. as of today, an AI that can substitute these aforementioned human processes and inputs and decide whats useful and what not in a non deterministic way doesnt exist yet (and if or when it does, beware of the technological singularity )

[Anonymous]

LaszloNagy wrote:
This is simply an algorithm with many parameters and variables. If this is AI, then most GDL objects are AI, as they generate countless variations of geometry based on user input and parameters.

I think Laszlo has touched the "Achilles' heel" of this whole AI hype...
The AI acronym is being used to describe what it is not. Artificial Intelligence, in my understanding, presume the whole process of: Observe > Gather Information > Process Information > Describe a Problem > Elaborate Solutions > Make a decision. All the examples shown in this Thread don't cover all these aspects. I mean: Like the example, the user must input how many desks he want, the constrains, what are the hight score point, and finally make a decision.
Perhaps we have to change the concept from Artificial Intelligence to Assisted Intelligence.
My 2cts.