The discussion around the likely impact of artificial intelligence on architectural practice is beginning to get genuinely interesting. In the period following the release of ChatGPT, much of the focus centred on image-generation tools such as DALL·E and Midjourney. More recently, however, attention has turned to the broader potential of emerging AI technologies in areas such as project scheduling, staffing, technical specification, performance evaluation and tendering.

‍

One particularly intriguing development emerging from these discussions is the role AI might play in the process of planning and development. While early, high-profile experiments—such as Sidewalk Labs’ controversial plan for Toronto – attracted considerable media attention [1], this piece focuses on a more routine aspect of the planning process: the basic application for planning permission.

‍

Before we get started, it is important to recognise that no two planning systems are exactly the same. In some municipalities, applications are assessed ‘in-the-round’ – characteristic of the “British/Irish planning family”, according to Newman and Thornley [2] – taking everything into account, including the architectural quality of the proposed design. Other systems (the Napoleonic, Germanic, Nordic and North American planning families) are more concerned with adherence to performance criteria and zoning regulations. So, when we speculate on the possible impact of AI on planning-application processes, we are not comparing like with like.

‍

That being said, popular opinion would have it that planning systems worldwide struggle with bureaucratic delays, inconsistent decision-making and difficult administrative procedures. To address these issues, jurisdictions in various locations have begun exploring AI-powered tools to accelerate and improve development approvals.

‍

One tool increasingly being adopted is ‘computer vision’, a powerful AI technology which interprets the information included in a planning application. Computer-vision tools are becoming highly sophisticated, capable of identifying where missing information in a drawing may be preventing either the AI model or the human planner from making a decision.

‍

The very simplicity of the technology which drives computer vision means that it is likely to find wide adoption. The technology has four key features:

‍

1. Neural Networks: These systems are trained on vast datasets of previously approved architectural drawings, allowing them to recognise patterns, standards, and recurring elements. This depth of training allows for planning reviews that are both accurate and consistent.

‍

2. Object Recognition and Classification: Modern computer vision can distinguish between architectural components—walls, doors, windows, mechanical systems—and assess issues such as corridor width or travel distances to escape routes.

‍

3. Semantic Segmentation: AI can now understand the spatial context of elements in relation to each other. For example, it can flag a bedroom placed beside a fire hazard as an error.

‍

4. Multimodal Communication: Advanced models can cross-reference written annotations with elements contained within a drawing, enabling checks for consistency between plans, sections, and specifications.

‍

In recent years, the City of Gainesville, Florida, reported that its proprietary AI review system reduced planning-review times from several weeks to just a few days [3, 4]. Similarly, Australian-based AI firm Archistar is gaining attention for its work with the cities of Austin and Vancouver [5, 6]. Most recently, the Department of Municipalities and Transport in Abu Dhabi claimed its AI-assisted system can deliver almost instant decisions for single-family home applications [7].

‍

As a measure of how quickly things are moving, the ‘Object Recognition and Classification’ technology cited in item no. 2 above is now being replicated in small academic environments, including the Department of Architecture at South East Technological University in Waterford. In recent months, fifth-year students at SETU examined the different ways AI is likely to affect architectural practice. One student, Conor Nolan, trained a basic computer-vision model to identify symbols and other information on architectural drawings. The experiment was limited in scope, but it clearly demonstrated how easy it would be to create an AI model capable of reading planning drawings and identifying missing information. (To get a sense of how Conor’s experiment works, scan the QR code below and, once the app is running, point your camera at the drawing beneath the code).

‍

‍

It seems inevitable, then, that AI will become a standard feature of planning processes everywhere, including here in Ireland. This raises a number of challenges, both social and technical. On the social side, planning holds a particular place in the Irish public consciousness and the idea of streamlining the process – potentially reducing the time available for public discourse – may require careful consideration.

‍

On the technical side, the variety of AI approaches already available may have a more profound impact on planning systems like Ireland’s than on those found in North America or other parts of Europe. In planning systems where outcomes are determined by strong mathematical parameters and performance metrics – such as those found in Gainesville, Austin and Vancouver – many of the outcomes can often be determined by conventional computing approaches. The addition of AI, while useful in many regards, represents more of an incremental improvement than a fundamental change.

‍

But in a system such as the one practised in Ireland, where applications are judged on a variety of sometimes very subtle metrics including quality of design, AI models trained on deep sets of historic data could prove transformative. These models have the capacity to examine previous applications in forensic detail, learning to recognise the complex factors that contribute to successful applications. This capability could enable more satisfactory outcomes on difficult planning applications while simultaneously guaranteeing fairness and consistency.

‍

The success of such a development naturally depends on the quality and consistency of the planning decisions that will form the training data for these new AI models. The consistency of the Irish planning system has been questioned over the years, which could limit the effectiveness of a heavily AI-informed planning regime. However, this challenge also presents an opportunity: the process of preparing data for AI training could, in itself, drive improvements in planning consistency and transparency.

‍

The integration of AI into architectural and planning practice represents more than just an advance in technology – it marks a fundamental shift in how we approach building design as well as how we plan our urban areas. As the tools evolve, their role will likely expand from basic compliance checks to assisting in achieving the optimum design response to any given set of conditions [8, 9].

‍

AI may also help us get beyond traditional divisions between planning families. Rather than maintaining the current distinction between Irish/British systems focused on ‘in the round’ assessment and Germanic/American systems emphasising adherence to preferred geometric arrangements or performance criteria, AI could enable all jurisdictions to implement planning systems that offer sophisticated solutions to complex urban problems. The technology's capacity to handle multiple variables simultaneously – from technical compliance to aesthetic considerations – suggests a future where planning systems can be both rigorous as well as satisfying.

The recent exhibition The Reason of Towns [1], along with the associated publication Approximate Formality [2] by Valerie Mulvin, are an appraisal of the inimitability and potential of our towns and villages across Ireland. They highlight the distinctive layout of the Irish town, characterised by a strictly structured composition and a foundational assemblage of public buildings. This has provided our towns, even with the most modest populations, with a rich compilation of fine churches, market houses, libraries, and courthouses often constructed from cut limestone and granite, establishing the foundation for a well-defined urban landscape.

‍

This formal configuration around market squares has provided the backdrop for the theatre of domestic life for centuries. However, many such squares currently stand devoid of vitality, plagued by neglect and dereliction, and burdened by excessive traffic and parking congestion. Any pride or affection we feel for them is inevitably tainted by the knowledge that they are imprints of a colonial past, which lingers in the configuration of streets and squares viewed as not entirely our own. Traces of the past still quietly inform how we move through and relate to them today.

‍

Over the past one-hundred years since independence, Ireland has struggled in navigating the postcolonial landscape and in addressing buildings with a residual colonial legacy. To date, a considerable portion of this discourse has primarily focused on the city of Dublin. The deliberate destruction and subsequent preservation of its characteristic Georgian terraces over the past century has been well debated and documented, and the value it adds to the urban fabric of the city has generally been accepted within the consensus.

‍

‍

The capital city assumed a symbolic role in negotiating the relationship with these buildings, determining which of them would be permitted to become emblematic of the emerging nation. This, coupled with the fact that the private market dictates that we develop urban areas faster, compelled the city to engage with its colonial built heritage earlier than its rural counterparts. Notwithstanding the triumphant role that economic priorities play in our evolving relationship with these buildings, this pressurised and hastened response to negotiating their legacy gives insight into the process involved to fully assimilate these buildings into the nation’s psyche. As this process is not as precipitated in a rural setting, an additional dimension of time is added to the dynamic. This passage of time hasn’t healed our relationship with these buildings; it has merely dulled it, leaving behind a quiet, unresolved ambivalence.

‍

Irish society within the twenty-six counties underwent a discernible shift at the beginning of the latter half of the twentieth century, transitioning away from a predominant fixation on resistance against British imperialism towards a heightened focus on contemporary economic realities. Consequently, the enduring colonial legacy of many of these buildings has made meaningful engagement with them increasingly difficult. Thus, many of them effectively became ignored and abandoned, and in being "tombstones of a departed ascendency – they are of no use" [3]. The collective memory deemed them too innocuous to warrant eradication, yet too historically complex to facilitate meaningful engagement. This brings us to today, wherein our rural towns and villages exhibit a uniquely strong sense of communal pride, yet often remain markedly detached from the very built environment in which they sit.

‍

Efforts certainly have been made over recent decades to challenge and question these prevailing narratives by various agencies promoting the conservation of our built heritage. However, there tends to be an emphasis on architectural features and artistic characteristics over the social aspects of the built environment. By focusing predominantly on technical and material issues, the broader socio-cultural significance embedded within historic structures can be overlooked, thereby neglecting narratives that contribute to a more holistic understanding of heritage [4].

‍

‍

This contributes to a significant portion of the Irish population lacking a sense of connection or ownership towards these colonial buildings, perceiving them as outside the scope of the nation's shared heritage. This disconnect does not stem from ignorance regarding the architectural significance of these structures. Instead, it arises from a residual colonial sentiment and collective memory of historical events. Without acknowledgement, this disconnect nurtures estrangement; an estrangement which cannot be overcome by simply celebrating a building's merits and architectural significance, but must invoke an architectural praxis built on social engagement.

‍

This is what Michael D. Higgins refers as "a feigned amnesia around the uncomfortable aspects of our shared history" [which] "will not help us to forge a better future together" [5]. He explains how the Decade of Centenaries has provided for a period of ethical remembering, which helps to understand the reverberations of the past for today’s society. It has necessitated uncomfortable inquiries into the events and influences that have shaped Ireland and continue to influence its contemporary landscape. The fruit of this enterprise, however, is a resilient society that fosters a "hospitality of narratives" [6], enabling it to effectively address the complexities of contemporary challenges. Through an architectural lens, this empowers communities to reclaim pride in their town and village centres, while critically engaging with and acknowledging the complex histories often embedded within these spaces.

‍

This revalorisation of built heritage in our towns and villages should not be understood as a finite project, but rather as a continuous process in the ongoing effort to unravel the enduring structures of the colonial condition. Nor should it be seen as unattainable, as many communities have already transformed these buildings to produce socially engaged spaces befitting of the communities in which they serve. It would be disingenuous to suggest that colonialism alone causes dereliction and decay across our urban spaces. There are many active elements within our own creation that drive this process. However, it is important to recognise the significance of this unique dynamic and the complexities it introduces, particularly when compared to our European neighbours with which Ireland is frequently and, at times, too readily compared.

‍

‍

It is not an exaggeration when Mulvin says, “the conservation and sustainable development of Irish towns [...] could be Ireland’s significant contribution to world culture for the next number of years” [7]. The conservation of these spaces can be understood as an act of decolonial cultural agency that reclaims architectural narratives suppressed by imperial paradigms. Our built heritage identity can extend beyond modest thatched cottages and traditional cozy pubs to encompass structures such as market houses, Carnegie Libraries, bridewells, railway stations, and workhouses, which are often integral to the fabric of our towns. This, in turn, plays an essential role in confronting the monumental housing and climate crises that imperatively shape the trajectory of our future. By acknowledging and confronting the contemporary forces of colonialism, Ireland can move towards a future built on a foundation of ethical remembering, reconciliation, and celebrating of its built heritage. The proof of this will be thriving towns and villages that promote sustainable ways of living; a built heritage of which we can all be proud.

Here is your edited text, following the provided editorial guidelines:

When OpenAI launched ChatGPT-3.5 two years ago, there was a rush of speculation about the impact the new technology would have on the building professions. A collective sigh of relief soon followed as architects everywhere realised that prompting text-to-image tools produced far less convincing results than they could possibly have hoped for. For the time being, at least, human participation in the practice of architecture seemed assured.

Nevertheless, various other strands of AI-based technology will inevitably bear upon the way in which architects approach their work. This article suggests how a few of these advances are likely to influence the practice of the profession in the near, intermediate, and longer terms.

‍

The near term: less tedium

As things stand, according to findings published earlier this year by the RIBA and elsewhere, AI is already starting to make its mark on everyday architectural practice. Text-based large language models like ChatGPT, Claude, and Gemini – easily accessible to anyone with a laptop and an internet connection – are already being employed to automate tedious and repetitive tasks. PDFs and Word documents containing data relevant to routine office operations, such as planning legislation and building regulations, can be uploaded directly to an LLM and searched with impressive speed for relevant and precise information. LLMs are also being used effectively to contribute to performance specifications and equipment schedules, as well as to assist in the preparation of technical reports.

In the sphere of visualisation, the limitations of text-to-image tools like DALL·E, Midjourney, and Stable Diffusion, as well as Photoshop’s native AI tool, Firefly, have already been alluded to. These limitations mainly relate to the technology underpinning the generative process. Text-to-image generation is mostly performed using a technique called ‘diffusion’, where thousands, and sometimes millions, of images of existing buildings are scrutinised at the level of the individual pixel, and the data harvested from this process is then recombined to create a plausible representation of a building that does not actually exist (Fig. 2). The resulting image may, at first glance, appear to depict an architect-designed building but, as the technology that produces the image does not correspond to any of the human processes involved in architectural design, the resemblance is little more than superficial (at least, at the time of writing).

Much work in this area has been carried out by the French architect and machine learning engineer Stanislas Chaillou. For a fuller explanation of how text-to-image technology works, as well as a more favourable opinion of its capabilities, his book Artificial Intelligence and Architecture: From Research to Practice is a good place to start.

While text-to-image technology has a long way to go, image-to-image tools can sometimes prove useful. The elegant line drawing to the left in Fig. 1, by Shay Cleary Architects, appeared quite recently on a cover of Architecture Ireland. A phone picture of this image was uploaded to an app called LookX, which very quickly translated it into something approaching a conventional 3D rendering. The text prompt used to generate the rendering was: ‘create a contemporary living room in a carefully restored Georgian house’. The image that appears on the right was a second iteration. It is easy to see how such a tool might come in handy in the tense few minutes before a client meeting.

‍

‍

The medium term: agents and advanced CAD

If we characterise near-term AI as something akin to an office assistant, medium-term AI is more likely to resemble an office administrator.

For about a year now, the tech press has been heralding the arrival of what it refers to as AI’s ‘agentic era’, a phase that roughly corresponds to stage three of the framework for AI adoption famously hypothesised by OpenAI’s CEO, Sam Altman. In this approaching phase, we can expect AI tools to be specifically tailored to the ways in which architects, as individuals as well as in teams, actually work. A company called Anthropic recently gave us a flavour of how agentic AI might work. They have developed a feature where an AI app can be trained to, in a sense, take control of a computer and independently perform complex tasks such as booking flights, arranging project team meetings, or filing documents into rationally organised directories.

Over the coming months, we are also likely to see medium and larger practices migrate to increasingly sophisticated AI-enhanced CAD and BIM applications. Although officially launched in May 2023, Autodesk’s Forma is continuing to evolve in interesting ways. Forma allows architects to quickly model complex site conditions and then immediately simulate the impact that the local environment will have on a variety of site proposals.

An application called ARCHITEChTURES covers similar ground to Forma. However, in addition to providing massing solutions for complicated, multiple-use developments, ARCHITEChTURES does a credible job of proposing apartment plans from preliminary site layouts. The apartment plans it suggests are updated in real time as different site options are examined.

While Forma, ARCHITEChTURES, and a host of other very impressive apps are not purely AI-based, they combine advanced machine learning techniques with long-established parametric design approaches to simplify the most laborious and least rewarding aspects of the design process. Development in this area is advancing at pace, and we can expect a wide range of applications of this type to become part of the working environment in the coming months.

‍

The long term: area planning and design assistance

The longer-term impact of AI on the practice of architecture is, of course, far less predictable. Of the many areas where advances can be expected, two seem particularly ripe for consideration.

First, stepping momentarily outside the strict confines of the architect’s office, it is worth reflecting on the impact AI-related technologies will have on the broader building regulatory environment.

As an indication of what might be expected, the City of Los Angeles is using AI tools to speed up the process of reviewing planning applications. Its system uses algorithms to evaluate development proposals against the city’s zoning and land-use regulations. Cities like Austin, Toronto, Melbourne, and Singapore are innovating in broadly similar ways. Allowing for cultural and other differences, Irish planning authorities are likely to advance in a comparable manner, using AI technologies to assist with the preparation of development plans and design guidelines, as well as with the review of planning applications. Such a development would bring about a significant transformation in the design and construction environment; however, while it might suggest a more efficient and transparent process, it is easy to see how other complications could arise.

In the longer term, we can also realistically expect that AI will eventually start to reshape the creative process itself. As LLMs and other AI systems become more sophisticated and move beyond the diffusion technologies mentioned above towards more vector-based models, architects will be able to develop tools that examine and mimic their own personal design methodologies. In this scenario, an architect might upload drawings from previously completed projects to a specially trained AI model. This model would then scrutinise the uploaded material and assist the architect in refining and developing concepts for future projects in a manner complementary to the architect’s own design technique. It is a tantalising idea and well within the bounds of possibility.

‍

Concluding comments

In preparing an introductory piece about its emergence in the architectural profession, many interesting aspects of artificial intelligence, including some innovative applications to which the technology is already being applied, have, of necessity, been overlooked. Similarly, some important developments in the regulation of the technology have been ignored as well. These include what some might see as California’s emerging role as the world’s AI regulator-in-chief; recent EU legislation concerning the use of AI, as well as interesting critiques of this legislation by Stripe’s Patrick Collison, among others; the massive amounts of energy required to run LLMs, as well as the knock-on effect these requirements will have for the natural environment; the AI-related re-emergence of nuclear technology in the production of energy; and ethical issues around the training of LLMs. All these topics will have an important impact on how the practice of architecture develops in this country. We would do well to get the conversation going on these issues sooner rather than later.

Finally, while some in the profession remain fearful of the impact that AI will have on job security, others cleave to the more hopeful and, perhaps, more likely view that AI will actually put more control over the design process into the architect’s hands. The truth is that it is too early to say what the overall impact of the technology will be. All we can realistically do for now is stay ahead of developments and take advantage of their benefits.