Generative AI and participatory urban design: the prospects and pitfalls

Nobody will miss it: ChatGPT. Midjourney. All the exponential development in generative AI. It seems like the boom is endless, and urban design is naturally a part of the discussion on use cases. But is AI a boon or a danger to participatory urban design processes? And how can we make use of it the right way?

Here, I lay out my perspectives on the matter and consider what’s at stake for participatory processes, and what it means for Metvibee and the urban planning and design community in general.

Drawing by hand; in this case, on a mobile phone.
Drawing by hand; in this case, on a mobile phone. AI can create professional drawings, but is it a good replacement for hand-drawn designs like this? (Coimbra, Portugal)

Generative AI: a brief overview

It seems amazing that AI can generate almost everything possible for humans, and might even be on its way to take away jobs from writers, artists and designers.

On a closer look, it might not be what it seems. Generative AI relies first on machine learning of a large number of design samples through neural networks, and then to generate a design from what was learnt and according to the prompt given by the user.

In other words, it can only “think” and “design” inasmuch as it was fed, and is highly susceptible to trends and biases of everything that human beings create.

Why do we do participatory urban design?

We must look back on the raison d’etre of participatory urban design to see what the future holds for it.

Arnsteins ladder of participation
Arnstein’s (1969) ladder of participation (as always)

The point of Sherry Arnstein’s famous ladder of participation, which I believe most urban practitioners are familiar with, is that important decisions are often being made with little regard for the people most affected by these decisions, and that we must enable more control by citizens at the local level in order for them to manage resources, make decisions, and create futures to their benefit and thriving.

Participatory design is one facet of increasing citizen control in urban design processes by actively partaking in the creation of visions AND retain control over the outcome. The keyword “control” is key here, both in terms of process and outcome.

Can generative AI be participatory?

Generative AI holds great promises in fueling the imagination of humans that make use of it to create designs they don’t normally have the time and capability to do so. This, unfortunately, is also its pitfall in terms of participation and control.

There is an oft-repeated phrase in computer science: garbage in, garbage out. The generativeness of AI is at the mercy of not only how the algorithm is designed, but also the quality of the sample being learnt by the machine. The output of AI will be biased as long as humanity as a whole still holds deeply held biases and stereotypes, and we face great challenges in addressing biases in AI. Relying on generative AI can, unfortunately, potentially perpetuate racial and gender biases which we’re fighting against.

More dangerous, I would say, are the less obvious biases that might be seen in the urban planning and design field if we put too much hope on generative AI. Besides gender and racial biases, there are also great biases against disabilities, people of low socioeconomic status, people of a non-Western culture, etc. which are at risk of being amplified by algorithms. In addition, visions of urban futures risk being dominated by the handful of key opinion leaders in urban design based in the Global North, which has more access to these types of experimental technologies compared to the Global South.

And most importantly, even if we’ve somewhat found a way to uproot all these biases and stereotypes completely, we still can’t escape a basic pitfall of AI: the voices of the vulnerable minorities and the disadvantaged would be drowned in the mighty ocean of big data, generic designs, and copycat solutions that pays no attention to these voices that urban practitioners need to advocate the most for.

This is a design photo I “created” with Midjourney for my Stockholm Centralbron High Line project. While “professional” design images might be attractive, we must be careful as to its pitfalls in losing citizen control to biased algorithms and machine learning samples.

On the optimistic side of things, however, generative AI may give citizens the possibility to create professional designs that they don’t have the tools and expertise for, and thus have an idea on how things would look like in the future according to their visions. And of course, machines have the ability to learn, generalize and possibly make suggestions from vast amounts of data, which is useful in urban planning (but ultimately it’s still human interpretation and decision that matters).

Here I must advise caution: while generative AI would be helpful in creating imaginative visions and points of reference for urban futures, we risk resigning our capabilities and control of the design process to the seemingly omnipotent power of AI, and thereby losing control of not just the outcome but even the process itself. And even risk falling back down to the bottom rung on the participation ladder by way of manipulative algorithms, sometimes uncannily.

This is not to deny the potential of generative AI as applied to participatory urban design processes. Generative AI is as good as a tool as extended reality (which we at Metvibee are passionate about), for example, when it comes to participatory processes. Participants in a workshop can of course try to envision the outcome by themselves through prompts they’ve come up with in a collaborative design workshop, and feel they’ve achieved a lot. AI-generated images can find good use as promotion material to gain political support for a design project. But we must bear in mind the dangerous pitfalls of AI, and use it in caution.

It’d actually be great if I could turn this design into a professional image using generative AI.

So what’s in store for participatory urban design? And for us at Metvibee?

Here at Metvibee, we chose active, collaborative design in extended reality as a tool for end-users to gain control in the creation of their urban visions. Participants are actively creating urban designs, while responding to their fellow participant’s imaginations as well.

We see generative AI as a tool for participants to create designs in a more efficient manner, with the proviso that the process and outcome must not dominate the design process. The participants must retain control on the input, process and outcome. Generative AI must be used in the right manner to ensure the stakeholders that are most affected by design decisions can benefit the most.

Generative AI is powerful but has many issues. These issues (especially privacy) are being exposed and hotly debated, and while these debates will slow down the current hype on AI, I see this as an excellent opportunity to explore the ethical challenges that would accompany its adoption in participatory urban design processes.

Metvibee will essentially continue in the direction of active collaborative design in extended reality, but we’d also like to explore generative AI as an aid in developing these designs. If you’re working on it, I invite you to connect with me on LinkedIn to work together in using AI the right way in furthering the adoption of participatory urban design processes.

Raphael Mak, founder and managing director @ Metvibee