The Global Archicad Community

[Jim Allen]

On AI and coding literacy:
"State of the art machine intelligence today is usually provided as libraries of code, written by large-scale influential companies, (for example Google’s TensorFlow or Facebook’s Pytorch) structured by professional engineers such that the involved architectonics and procedures conform and reinforce the engineering problem-solving mindset. Anyone not able or willing to adhere to this mode of operation—for example architecture, which is neither a discipline, nor it is strictly about problem solving—is kept at a safe distance, and offered tools and tutorials. The knowledge about a computational concept, for example a much-hyped GAN, is offered to the peers as a complex technical paper, and to the rest, simply as a library of code to be played with. If an architect, willing to reinvent her field in today’s novel and significant technological context wishes to acquire the necessary literacy to navigate the space where the knowledge is created and negotiated, she faces considerable difficulties. Prerequisites to enter the field are the same as for future engineers, along with pedagogic principles. If, on the other hand, we are unwilling to pursue this literacy, we are once again in a situation to simply accept the tools and let them write our legacy. However, this time, the shortcut that we would be taking might have far greater consequences than before. It could, in fact, do a great honour to computer science by allowing it to turn a three-thousand-year-old legacy of architecture into one of its particular specialisations."
Nikola Marinçic

I'm sorry, the words are in English, and they are all in the right order, but I'm struggling to understand this.
The bits that I do understand I think are completely wrong.

The only people who tell us that AI will replace architects, are those who don't actually properly understand what we really do.

[Jim Allen]

I had a quick play with how this might work. I mocked up an interface proposal, which I have attached below. It is based on the principle of making individual components using just the slab tool. It allows quite a lot of flexibility, but anything that is more three-dimensional can be modelled with the morph tool and loaded.

It keeps things simple.

There is not too much wrong with the current GDL editor. It's arguably too complex, but the key issue is that you have to script geometry creation as opposed to drawing it. This is the single biggest obstacle to making new parametric objects.

The idea is that you draw named geometrical elements, configure the materials, fill, line types and colours etc, and then you can change the scale, rotation and location of these relative to one another, and you can use the parameters in one element to drive those in another.

It also separates the specific elements of the Object Properties box, so they can be configured in isolation.
There are lots of things that Objects can include, and configuring these in discrete dialogs is not a bad approach.

You will notice that the Appearance in 2D and 3D is based on changing to the specific view in the preview box, and then clicking an edit button, so the preview acts as the way of configuring the views.

This is intended to be a starting point for discussion, not a fully-formed proposal.

Comments are encouraged!

Screenshot 2020-04-03 at 15.59.08.png (394.64 KiB) Viewed 955 times

[leceta]

I'm sorry, the words are in English, and they are all in the right order, but I'm struggling to understand this.
The bits that I do understand I think are completely wrong.

Understanding just the bits is not enough. This (could) lead you to decontextualize the words.

In my understanding, what the author wants to say is that architects using tools developed by people from other disciplines (software engineers) are ( arguably) limiting their "design space" to foraneous mindsets. So the interest of having architects literate in coding.

BTW, one of the lemmas of Graphisoft has been that ArchiCAD is "designed by architects for architects"

[TMA_80]

Sorry to ask this, but regarding the visual interface, Is "the library part maker", a potentiel take ( vs the family editor in Revit)?
Or, is the nodal approach: Marionnette, Grasshopper, Dynamo. the real next step and take?

Maybe, both...

[mikas]

I think that if there ever was an easy way to implement a general visual gdl-editor, it would have been done already. By Graphisoft itself, or by a third party developer. GDL, it’s an easy to comprehend language. I by myself have done a couple of objects from the ground up. A coordinate tool, and a water slide object. The latter one is still in use today, and it works just like it is intended to. Both of them were scripted as of late 90’s if I remember it right.

For me it’s been a while though with scripting, and it’s more complicated today than it was a couple of decades ago, I think. I would love a visual gdl editor. It would be magical for me and for everybody who are here today working with ArchiCAD.

Having done a couple of gdl-objects myself, I don’t think it’s an easy job to make and realize that kind of an editor. There are much too many variables in it.

I believe strongly in grasshopper connection myself. It’s visual enough, and mathematically more powerful and versatile than what gdl in itself is. GDL, as old as it is, does fulfill it’s purpose unbelievably well still today - as a geometry engine. I am amazed what they did back then at ´80s in GS Hungary laboratories. Great Job, thank you.

And then were are these other languages too to interact with ArchiCAD, like Python, which I am not familiar with, unfortunately.