The Image Research Network

Architecture and the Built Environment Seen Through the Eye of the Lens and AI: Does It Matter Which?

Every day, we encounter photographs that have been manipulated via computer software. However, there is an argument that deceit has always been the case with photography, and audiences of such pictures are expected to be passive in acceptance of their controlled and constructed content, regardless of provenance. However, until recently, human agency has been central to physical creative processes. Using architecture as subject focus, this article questions whether creative AI software programs can mimic known built environments effectively and if AI-generative images embodying “photographic” accuracy may be aligned with the actual places and spaces they depict. Audiences of architectural photographs should be able to “read” buildings and are generally less concerned with the method of image creation. This suggest that AI-generated works pose no greater threat to the perception of “reality” than signifiers of the subject produced by conventional cameras. Using the software program Midjourney, this study employs Methodological Pragmatism, including image analysis, interviews, and qualitative research to question if creative photorealistic AI text-to-image outputs may provide a trusted view of the world. Viewer cognition and interpretation are discussed, along with the type of authority exposing architecture as subject, asking if audience “faith” has been taken away since the emergence of AI images? The research concludes that if we could understand what we are looking at, signifier or signified, it rarely matters whether a picture was created by a camera or by a series of text-to-image prompts. Education and trust are vital. As always in photography, context is king.

With a lifelong career in architectural photography, understanding the wider contexts of technological change through the introduction of generative AI imaging has become empirical to my practice. As a university lecturer, the implications of production moving away from solely lens-based media has the potential not only to affect me professionally, but also my students. Such significant shifts in methodology and by proxy, audience interpretation of this flux of new images into the public sphere, has become central to my professional understanding and academic criticality.

As a practitioner whose pathway from initial client brief to image publication also encompasses significant post-production, I felt it important to interrogate the intersection between a real photograph of a known place or space, and that of AI generative software’s attempts to create, or indeed, re-create a visual outcome in an authentic manner.

I took the opportunity to seek the advice of a respected architect and Midjourney user, Damon Leverett in Arizona. This relationship facilitated the creation of several vital visual outcomes. Together with another architect and neuroscientist Clarissa Machado in Italy, the ensuing dynamic of conversations surrounding Leverett’s images, coupled with Midjourney’s inaccurate representations of an authentic address, became key to my findings.

In summary, my research discovered that with the use of a creative AI tool such as Midjourney (which works from a substantial and ever-expanding image dataset) that the outcomes returned provided strong responses to architectural style-driven requests. However, they were disarmingly inaccurate with regard to replicating geolocated structures that exist in built form.

How does this affect me, my students and my professional clients? In the short term, we are rewarded. Because of AI generative software’s unverifiable responses, there is an enduring requirement for lens-based, on-site capture. More worryingly however, human cognition is significantly challenged by the plethora of ‘fantastical’ views which are infiltrating our conscience via social and mainstream media. We cannot always understand what we see, we are building uncertainty and query into our engagement with ‘photographs’. Our visual literacy is challenged and until we learn to critique and evaluate more robustly what we are presented with, we are in danger of blindly trusting what cannot be proven.

Since the publication of my article, I have begun to take its findings outside academic circles and into the public sphere. As an author of a substantial professional book: Photography for Architects, Routledge, (2024), I am able to canvas a reception for my insights at grass-roots level, through my talks across the UK for photographic clubs and associations such as Britain’s Royal Photographic Society (RPS). In these sectors the response has been powerful. Audiences are keen to understand how AI generative imaging programmes function and why they cannot be trusted in their visual outcomes. They are shocked to learn of hallucinations and misrepresentation of built environments and are keen to act as advocates in advising other audiences to question the veracity of the images they see. My agency in empowering others is absolutely central to my own line of continued enquiry.

—Martine Hamliton Knight