AI should be embraced by architects rather than met with scepticism or fear, writes Zaha Hadid Architects principal Patrik Schumacher.

Since Open AI's DALL-E 2 became available to early adopters – including Zaha Hadid Architects (ZHA) – about 18 months ago, this and similar artificial intelligence (AI) systems have taken the world of architecture by storm. The same has happened in many other design disciplines and artistic professions.

These recent AI systems enjoyed the fastest adoption growth rates in history, and the enthusiasm is justified. Everybody using these systems enjoys a momentous boost in productivity and creativity.

AI systems have taken the world of architecture by storm

To answer critics, such as The Guardian's architecture critic Oliver Wainwright, who think all this can be dismissed as superficial because it works via mere image generation, I would like to remind us that a picture can speak a thousand words. From these coherent images we can create a well-articulated spatial composition, and also get input about the plastic articulation and material expression of the building.

To be sure, the non-trivial art of prompting and selection must be learned and mastered to get convincing results. Those who do not understand the importance of articulate visual communication for architecture's social functioning have missed one of the most important lessons of the history and theory of architecture. Kevin Lynch's evergreen classic The Image of the City underpins my point, a point I have elaborated extensively in my writings on architectural phenomenology and semiology.

The image is very important. However, ZHA, and our discipline more generally, also embraces and actively invests in a broader conception of AI that goes beyond the technique of neural networks.

It was exactly 40 years ago that I started my architectural studies in Stuttgart University. At the time, artificial intelligence was still relatively young. However, generative computational processes were already being explored within architecture by George Stiny at UCLA/MIT, Horst Rittel and Frei Otto at Stuttgart University, Bill Hillier at the Bartlett, and John Frazer at the Architectural Association, to name just some of the pioneers I had the pleasure to be touched by or interact with personally.

Although neural networks were already around in those early days, the above-mentioned trailblazers developed, before big data, intelligent generative methods based on explicit theory-based programming (as well as evolutionary algorithms) and focused on organisational-functional optimisation.

At ZHA we are investing in long term research-and-development projects and teams in this broad tradition, with special focus on both generative and analytic space planning tools (including machine-learning tools) and agent-based occupancy simulations using gaming AI, writing original code in our effort to build potent design-optimisation software focused on social functionality.

I always welcome competition as a catalyst of self-improvement

More recently we have initiated a research project aiming to develop a game-based methodology and toolset meant to facilitate participatory urban development and community creation. All of these new tools not only increase our labour productivity and efficiency, but also enhance the quality of our professional services, enabling works and insights that simply could not be had before.

With respect to the most recent wave of big-data-based AI tools, the worry has been articulated that AI will make many jobs within architecture redundant, or altogether replace or devalue our profession.

I do not share this worry. The historical experience with earlier productivity leaps, i.e. when CAD and CGIs were introduced, was that the productivity gains went into higher quality work, into more options for clients, and thus contributed to better designs and decisions. I foresee the same now with the adoption of AI.

Architectural design fees are a relatively small part of the overall cost of creating a new building. It makes no sense to save here, but it makes a lot of sense to further increase creativity and the number of design options from which to develop solutions. In accordance with economic cost-benefit logic, design fees as a percentage of total project costs has been going up in recent decades and I expect this to continue, moreso the more we innovate and deliver qualitatively new capacities.

Going back to the current excitement about image-focused generative AI, as everybody's creative capacity is boosted by AI, the competition in creative design intensifies. Epigones proliferate and might be hard to distinguish from the original. To be sure, downstream quality execution on time and within budget is much harder to replicate, so that it is relatively easy for us to maintain our overall competitive edge.

But I am also not at all worried about facing the newly empowered competition enabled by AI in terms of front-end design. I always welcome competition as a catalyst of self-improvement.

I am not only concerned with the flourishing of ZHA, but with the flourishing of architecture

In any event, we have been embracing these new tools from the get-go, for our own further empowerment. ZHA has the unique advantage that these huge general purpose systems have gobbled up more images from ZHA than from any other architect, which in turn allows us to credibly and creatively explore and evolve our own oeuvre with these systems.

While we do this shamelessly, we are also developing our very own tailored neural-network systems, calibrating Stable Diffusion with our own project-image datasets, and adapting it to our own specific architectural needs. We are currently testing various training methodologies.

We are also building new interfacing features and functionalities. We integrate constraints and controls that allow us to steer the output through parameters and direct design inputs. We can also build the output cumulatively and vary some parts while keeping other parts invariant. We can fine-tune outputs with project specific training datasets. We run these models on our own server farm. Our next aim is to generate clusters of images (front/back, outside/inside) that together describe a coherent design.

These are just some hints indicating the direction of our AI research and development trajectory. It's oriented towards upgrading our capabilities and competitive edge, with a view to rapid results because this space is moving so fast. We are also collaborating with universities, e.g. trying to tackle the holy grail of 3D AI.

This is all just a beginning, although the new capabilities are already starting to become compelling. I think it is very important, not only for ZHA, but for the discipline of architecture as a whole, to not only consume generative AI tools as end-users but to get involved in developing discipline-specific versions of these tools.

As always, I am not only concerned with the flourishing of ZHA, but with the flourishing of architecture, the built environment, and its creative contribution to societal progress. It's all part of human flourishing. It's thrilling.

Patrik Schumacher is principal of Zaha Hadid Architects.

The image was created by ZHA using DALL-E 2.

Dezeen In Depth

If you enjoy reading Dezeen's interviews, opinions and features, subscribe to Dezeen In Depth. Sent on the last Friday of each month, this newsletter provides a single place to read about the design and architecture stories behind the headlines.

AItopia

This article is part of Dezeen's AItopia series, which explores the impact of artificial intelligence (AI) on design, architecture and humanity, both now and in the future.