Blur Theory is a drive through Dakar’s streets through the lossy lens of AI. This drive starts in real places and then takes a journey through various generative AI models, losing context and definition at each turn.
To create each image, I gather photographs of a specific place, while in motion - walking, on a bicycle, on a bike, and in a car. In parallel, I prompt a diffusion model for that specific place, which will usually give me a result that is very inaccurate. I then merge the 'real' images with the diffusion model ‘artificial’ images and apply a double exposure style to arrive at the final image. First, it’s about establishing the lack of context (or lossiness) of the model, and then using actual images to correct the model as best I can. Of course, the results are always fictional or speculative so it's really my positionality in relation to the real images and the model’s images that gives this speculation meaning.
I am interested in how we remember things, people, or environments. Memories are subjective, points of view, that relies on emotions to be entrenched in our minds. When I try to remember my city, Dakar, most of it is hazy but some of it feels clear – places where something interesting enough happened to warrant lingering in my consciousness. So the idea with taking pictures while moving was to convey the subjectivity of memories without manipulating an image, which would add another layer of subjectivity to the already subjective experience of seeing while in motion. Such that, even before confronting the inaccuracies of diffusion models, I can be transparent about my positionality and the biases it implies. That I am human, the citizen of a place in which I have memories, and that those memories aren't objective or factual. They are formed in the act of living and the temporality living implies. What I remember of my city is an amalgam of past, present, and future: being dropped off at school by my grandmother, dropping off my child at school, imagining how a facade would look better with different tiles, going to visit my grandmother's various friends, walking to my studio, watching my dad come home from work on his scooter from the balcony, driving around my block looking for parking, reimagining cailcedrat trees I remember from childhood on a roadside where they've been long cut down, walking to the bodega to buy some cheese with my grandfather, etc.
In the end, memory can be a kind of imagination that draws from sources that are real. This same understanding of memory can be applied to describe an image generated by machine intelligence. The difference though between me and AI is that most of my 'database' (the experiences that feed my memory) are highly grounded in the places I try to remember. If there was a way to quantify my 'database' of Dakar compared to a diffusion model's database of Dakar, I think I would come out as having more datapoints. Of course there is no way to actually prove that, but because I can tell whether a diffusion is right or wrong about my city, I assume that it is the case. The other difference between me and AI is that I would never attempt to 'remember' the streets of Hamburg (a city I have visited once for a total of 72 hours) with the same authority as I remember the streets of Dakar. A diffusion model doesn't have that discernment about positionality and context. It just does what it's told with the same authority, and it's up to prompter (so far) to determine how accurate the results might be.
A general definition of theory is that it is a supposition, a belief held without proof, to explain an observable phenomenon. It hypothesises the whys and hows that explain the whats. Some theories are more widely accepted than others, but all theories share a profound relationship with subjectivity which is necessary to their vulnerability to critique.
Theories are foundational to empirical studies and scientific discoveries, and their vulnerability to critique is their strengths. They can be torn apart, built on, adjusted, thrown away, etc. These activities are essential to how we grow what we know and how we know. It is on that basis that I decided to name this body of work 'blur theory'. It refers to my own set of suppositions about how certain places, people, environments, and ways of knowing come to be gradually erased and eventually forgotten.
In making this work, I propose my theory of erasure against the dominant theories of progress that bring it about. I am looking to expose the vulnerability and inherent subjectivity of my theory about how my context is systematically erased to the vulnerability and subjectivity of hegemonic theories of progress that have informed much of technological advancement and caused the erasure of my context.
AI is the first kind of technology that visibly confronts all of us with how our theories about progress have affected the way we see and make sense of the world. AI models are memory machines. They draw from data about the world to 'imagine'. However, the state of the 'data about the world' asks many questions about taxonomy and the political nature of knowing and recording: Who decides what things (objects/ideas/places/people) are worthy of naming? Who names them? Who looks for the named things to train models with? Who points the system to where the named things are? And why? And how? And when? And where?
The answer to these questions points to the highly subjective and therefore political nature of the 'data about the world', its incompleteness, its unfairness, its unevenness. Very much like my own personal memories of my world. AI then cannot be 'the' world, but 'a' world within many worlds. And theories about progress are at the root of how this immutable fact came to be overlooked today.
Ultimately, what I mean to propose is that all theories can (and should) be re-examined when they no longer serve humanity and the planet. Their vulnerability isn't always clear to us, especially when we've held them as true for so long. We often forget the reasons we hold them as true and that, for example, unspeakable violence over centuries might have been involved in their wide acceptance. Yet, their vulnerability is precisely what we need to remember and exploit in order to bring forth alternate theories that serve us better.