Reflecting on the NLA Built Environment Technology Expert Panel’s third cycle, Chair Camilla Siggaard Andersen highlights the challenges with Large Language Models.

Monday October 21 marked the end of the third cycle of the NLA’s Built Environment Technology Expert Panel. Since the very first meeting three years ago, we have become progressively more specific in our recommendations and focus areas. This year, we set out to work on three specific challenges, responding to new technological opportunities afforded by the rise in Large Language Models (LLMs), while addressing areas of interest identified in the previous cycles. Each challenge was tackled by a different working group, developing their proposition alongside their daily practice.

Challenge 1: How might the vast amount of data that is made available through online data platforms be curated and communicated into useful community insights using LLMs?

The working group initially set out to investigate the potential for using an LLM to draw insights from the London Datastore. In theory, it should be possible to create an overlay to this service that might help individuals and organisations locate the most relevant datasets and interpret their meaning. With hundreds of datasets to query, an AI-powered open-text search function could significantly improve the user experience of the Datastore and enhance inclusivity across the platform. However, due to the timing of this challenge in relation to the wider Data for London programme, the working group was not able to make further progress.

Instead, drawing on expertise from their own practices, the working group investigated the opportunity for hosting parametric models within an open-source gaming environment. The benefits of this approach would be increased community engagement in construction projects, particularly among younger audiences. Many architecture practices and academic institutions will have some experience experimenting with these tools, but their efforts have not yet galvanised into a common standard. Therefore, the working group recommended creating an online community of knowledge exchange, united by a belief in open-source technology. Next cycle, the group might look back at the London Datastore to investigate the potential for integrating context-specific insights with accessible 3D models.

Challenge 2: How might we increase the adoption of circular economy software in architecture by using artificial intelligence (AI) in combination with an upskilling programme?

Last cycle, the expert panel suggested that the GLA should launch a skills exchange programme like Public Practice, focused on the placement of people from the technology sector in architecture and construction, and vice versa.

This year, the working group investigated the potential for upskilling professionals in circular economy practices. The team initially identified a wealth of existing resources to support this goal. However, they did discover a gap in understanding client barriers to adopting circular economy solutions. Ultimately, developers and investors must be comfortable with a degree of uncertainty in design and construction processes, as circularity often involves reusing materials with fluctuating availability. Currently, no single system can process all circular economy materials, let alone predict incoming resource pipelines. However, in the future, large language models (LLMs) may help by making sense of vast amounts of unstructured data, identifying potential regulatory issues, and developing due diligence plans. This, in turn, might help developers and investors trust the process more, even if it is more iterative and agile than traditional methods.

In the next cycle, the working group recommends presenting a future-focused scenario to a selection of developers and investors to gauge their interest in a common AI-powered platform supporting circular economy practices.

Challenge 3: How might we develop new infrastructure business models to encourage the implementation of micro data centres with better community integration?

While it’s easy to get excited about digital technology’s potential to compute complex information and improve urban systems, it’s equally important to consider the physical impact of these processes. This was made clear by last year’s investigations into the ‘hardware of the software’. This year, the working group delved into the potential for micro-data centres to address the most common concerns: energy usage, waste heat, environmental quality, and monopolisation.

Inspired by the recent emergence of community-owned micro energy grids and local heat networks, the group explored the benefits of communities controlling essential infrastructure on a small scale. They also examined examples of data centres integrated with other urban systems, particularly for heat exchange. Three central questions emerged: 

1) What would incentivise communities to prefer a local micro-data centre over a distant “super scaler”? 

2) How could common data storage and processing services be efficiently delivered in smaller, distributed networks?

3) And, what would happen if we considered digital infrastructure a civic amenity rather than a commercial commodity?

The panel reiterated last year’s suggestion: to establish an Infrastructure for London governing body to oversee pilots, coordinate efforts, and promote utilities as a common good. The panel also proposed summarising the key enablers and barriers identified to inform further explorations and support continued advocacy. 

In the next cycle, we will continue to investigate the impact of digital technology on London’s built environment. We may take a step back to examine the broader digital disruption of traditional architectural professions, while also taking a step forward to advance this year’s propositions toward implementation.

As always, we will be guided by the three shared objectives defined at the outset of this panel: to discover ways technology can unlock a more democratic, sustainable, and human-centered city.