BIM Pirat: Can ChatGPT be an Architect

Lately, I've seen more and more posts along this line, and they all miss the point of understanding architecture as producing plans/models. See which topics LLM's can support and which not.

As you see, I started a new chapter - It’s called BIM Pirat’s Message in a Bottle. In it I explore more wider topics than in BIM Busters. The main reason to change is:

1. It's not enough to bust some myths to build better buildings; it's about embodying a new role to further this goal - The BIM Pirate, which challenges the status quo.

2. Social media is like throwing a bottle in the vast sea to hope that somebody (who is not a fish) finds and reads it.

3. We take ourselves much too seriously and don't connect emotionally to the topic of better buildings. Therefore, forget that we are still humans.

---

To answer this questions seriously, we need to answer the following:

What is an architect?

What does an architect do?

Some people will say an architect is somebody who studies architecture, practices long enough to get a certification, and pays the yearly fees. In other countries like Switzerland, "Architect" is not a protected job description. Moreover, nowadays, when you Google for Architectural jobs, you will find more software than classical architectural jobs. This reminds me of the joke about being a doctor, not a real one who helps people.

So far, ChatGPT can't become certified, and even more importantly, as far as I know, no LLM is intrinsically motivated to shape its environment and to enjoy suffering through cruel all-nighters finishing a design presentation :-). 

In my thinking, self-motivation is a clear line for "intelligent life," and I guess to get to real general intelligence, we need to introduce some kind of evolutionary pressure that has consequences for the algorithm. I'm just not sure we should go in this direction, but that's a discussion for another time.

So the "What is" question is not so helpful. Let's try what does an architect do:

There is the Vitruvian definition of an Architect who produces stable, useful, and beautiful structures. This definition is still valid but does not encompass the full complexity of the job, and I prefer the following: 

1. The architect translates wishes into requirements and solutions - a good concept.

2. The architect enables clients to make decisions. 

3. The architect serves the construction site so that the approved concept can be realized frictionless.

Most architects fail at all three jobs or at least can improve. I hear the outcry, so what I mean:

1. In many projects, we see late-stage design stages, some due to not listening enough to the client's needs.

2. A common complaint from architects, engineers, and BIM professionals is that clients don't make decisions. That may be true, but it just means that they are in the role of a victim. To get out of this victim's position, it's better to look at what you can do so clients can make decisions.

3. Serving the construction site is about delivering the correct information at the right time to the right people so that they can work as streamlined as possible.

Most buildings will be built in the end, but I seldom hear people say the last years of designing and building were a pleasure and that the building completely satisfied the user. 

So where can ChatGPT or any LLM Machine Learning application support to become a better architect?

1. They can help formulate better material requirements documents in a language the user and the specialist understand.

2. They can explore a wider solution space faster. All the applications of creating renderings or different urban design configurations come to mind. Client support for decision-making becomes even more critical when using these generated solutions. Just imagine the decision paralysis when having to choose between 1000 similarly well-presented solutions without clear criteria.

3. Supporting the client's decision is still limited. Clients can try to ask for a summary of the often too long, too technical, and too complicated documents that are widely handed over to clients for approval and foundation for decision-making. But I would be afraid of hallucinations with these documents.

4. Our work with the Building Copilot enables very granular metrics and benchmarks for new designs. These can become tools to evaluate objective designs and find gaps. But it's still only a tool, and it won't (should not) make the decisions.

5. Creating buildable models with ML is still an unexplored space, and I know no company that is tackling this challenge. (It's a very complex one, so it is not a good starting point for any startup.)

So, I hope to read less about ChatGPT becoming an architect and more about how it is being used to solve one of the industry's many problems. What's your take on it?