‘AI’ Isn’t What You Think It Is
Artificial Intelligence is now everywhere, in conversation, in headlines, and increasingly, in architecture.
But in most cases, the term is being used loosely. We are told AI is transforming our tools, our workflows, even our creativity. But if you look more closely, what is being marketed as intelligent is often no more than a probabilistic engine, a system trained to predict the most statistically likely response to a given input.
This is not intelligence in any meaningful sense. It is not reasoning. It is not autonomous. We must draw that distinction, not to resist change, but to engage with it more critically.
Working with the Tools, Knowing Their Limits
At Kevin Kelly Architects, we have always embraced advanced digital tools, not for novelty’s sake, but for what they enable: clarity, speed, and agility in a complex design landscape.
We have explored a wide range of platforms marketed as AI, from Midjourney and Runway to the OpenAI suite. For our purposes, the most effective workflow to date has involved daisy-chaining Stable Diffusion into our modelling environments, Grasshopper, Rhino, and V-Ray, via ComfyUI. This allows us to move quickly from early concept models to visually rich renders, even before detailed geometry is in place.
It is a powerful pipeline, fast, elegant, and efficient. But even here, the underlying mechanism remains a statistical algorithm. It does not understand, prioritise, or interpret. It returns the most likely outcome, not the most appropriate one.
That is a key distinction, and one that matters.
What Real AI Would Look Like
If we are to call something intelligent, it should be able to do more than mirror its training data. True artificial intelligence would be able to:
Understand complex contexts
Set goals and pursue them
Adapt to new information
Reason logically
Make autonomous decisions
No system we have encountered, including the most advanced large language models (LLMs), does this. They simulate fluency, but they do not possess insight. They generate outputs, but they do not know why.
And as these models are increasingly trained on AI-generated content, the risk of recursion grows. The knowledge base becomes circular, systems trained on systems trained on systems, flattening the scope of originality and reinforcing the average.
Precision Over Probability
Much of our work involves scripting and modelling environments in Grasshopper and Python, platforms with a high barrier to entry, but with precision and controllability far beyond any generative AI tools we have tested.
These environments demand a different kind of thinking, logical, structured, and intentional. They do not guess. They act as extensions of thought, transparent, traceable, and modifiable. That makes them, in our view, vastly more reliable.
This does not mean we resist new tools. It means we apply them deliberately. We know where they add value. And we know when deeper understanding is required.
Language and Hype
There is a growing disconnect between how AI is being discussed in architecture and how it actually functions. This mirrors what happened with BIM, another term that was widely adopted before it was properly understood.
We believe this knowledge gap matters. Clients, collaborators, and even design teams deserve clarity, not just functionality. It is not enough for a tool to work. We need to understand how, why, and where its limits lie.
That is where we position ourselves, not as early adopters for the sake of it, but as critical navigators in a noisy field. The goal is not to chase novelty. It is to use what is available, with care, to serve complexity with clarity.
We are safe, for now
The tools are improving, and we welcome that. But Artificial intelligence, in its truest form, is much more complex than a high fluency probabilistic sequence of algorithms producing the most likely response. It involves judgment. Discretion. Ethics. Interpretation. It involves where the limits are and what is really going on with LLMs.
At Kevin Kelly Architects, we bring digital fluency not as a selling point, but as a means of doing the work well, blending speed with thoughtfulness, clarity with complexity. The difference lies not just in what we use, but in how we use it.
And that, in the end, is where real intelligence still matters most.