SPACE/ Mapping AI

The premise of this essay is situating AI, through discussing media technologies more generally, in space. I depart from media studies, specifically media archaeology. This domain stresses the importance of both the machines and their imaginaries for analysis; equal attention to a combination of the material and discursive infrastructure that constitutes media technologies in daily life. Further I bring in Arakawa and Gins’ notion of ‘procedural architecture’, which calls for a materialism of time (2002). Considering that media technologies are often time-based, understanding ways in which their time dimension can be explored is integral to their analysis. The view of Arakawa and Gins is a materialist one in quite the literal sense; taking seriously the spatial metaphors employed in the language of time, they argue that time is (merely) a derivation of space and that the hierarchy between the two concepts should be reversed. The philosophical implications of their work, expounded upon by Lecercle (2010), bring me to focus on the real-time aspect of media technologies, as well as their structures and procedures. Some of the analytical tools I take away from and for this are spatial exploration and mapping. In other words, time-based media should be explored spatially, whereby maps can be a suitable way of documenting findings along the way.

At first the image of a nation-bound map of AI, where AI is pinpointed down to concrete addresses seemed absurd to me. How can an algorithm be contained in one place? What does geographical location mean to a media technology that promises to be ubiquitous? What are the implications for the ‘intelligence’ in AI if it can be mapped to a point?

— ‘Artificial intelligence’ filter applied to the CzechInvest Web App —

The area with the highest density of AI is Prague, the capital.

— Zooming into Prague on the CzechInvest map showing AI in the Czech Republic —

Combining the available filters, some interesting spatial relations can be drawn between sites of AI and other forms of infrastructure. Some first-glance associations: subjects working with AI appear to be accessible and connected by highways and first class roads. For a large part, their occurrence coincides with the presence of universities.

— The CzechInvest only map showing the artificial intelligence filter along with (left) highways and first class roads and (right) universities. —

I introduce this map here as a tentative argument for the relevance of understanding AI as a media technology in space. Although the bulk of the writing below is primarily theoretically oriented and therefore not directly concerned with an analysis of this case, parallels could be drawn with the arguments presented along the way. I return to the Map of AI in the Czech Republic briefly as a means of conclusion.

Thinking space through media archaeology

To begin to analyse algorithms like neural networks, I find the theoretical position of the media archaeological approach formulated in the field of media studies a useful point of departure. Given their contemporary uses, algorithms have become subjects of analysis of media studies. Aiming to give equal priority to material apparatuses as well as discursive formations, media archaeology strives to understand the history of media technologies (Kluitenberg, 2008). It focuses both on the machines and their imaginaries.

With media technologies like algorithms, these two dimensions have arguably become more tightly intertwined (the opposition here is to more traditional media technologies like archived documents, televisions or radios, which were some of the first objects of scrutiny of the field). As code operates remotely, and across different machines and platforms, the machine – or material apparatus – in the equation is somewhat more difficult to put a finger on. Likewise, the code is an executable discursive formation, an imaginary of the developer(s) formalised through conventions. The machine simply allows its self-realisation, given compatible infrastructure – both hardware and software, that is.

My point here is that today’s media technologies necessitate a media archaeological approach. Taking the metaphor quite literally, we need to borrow from archaeology to understand media like algorithms (or do we still need to qualify viewing algorithms as media?). One thing I find important to borrow is a focus on and in space. As an inherently spatial activity, this grants algorithms the spatial dimension they occupy.

Indeed some theorists have stressed the importance of this spatial dimension. Kluitenberg (2008) describes the ‘ecological’ view of technology, whereby media technologies are often interpreted as forming landscapes, or being constitutive of our ecology – living space. Haraway’s Cyborg Manifesto (1985) similarly pointed to a breakdown of the boundaries between organisms and machines, thereby implicating machines in a shared ecology with humans. Kitchin and Dodge have gone to the extent of defining code/space; occurring when ‘software and the spatiality of everyday life become mutually constituted’ (2011: 16). Allowing that code and spatiality do not always exist in an exclusively dyadic relationship, Kitchin and Dodge (2011) view some interactions as occurring in ‘coded spaces’. In such spaces code and space are not produced through one another, but code does influence the way spatiality is formed. They give the example of a public presentation being given with the aid of PowerPoint slides; their projection shapes the space, performance and its understanding, however the talk could (most likely) be given even if the computer were to crash.

Whether algorithms behind AI constitute code/space or coded spaces in their interactions differs. In some situations they are aids – informing decisions made by humans, for instance – in others they have full control over/ are indispensable to processes. The distinction is not clear cut. Would the digital space of a customer service line exist without the bots employed to respond?

New technologies tend to erase older domains of knowledge and skill, sometimes beyond recovery (Kluitenberg, 2008). This statement would suggest that even coded spaces are likely to become code/space over the course of time; a technology seen as an embellishment or aid today may become a necessity tomorrow as doing without it is forgotten. The implicit issues of recovery, displacement and forgetting bring us back to the language of a media archaeological expedition.

While I mentioned the need to borrow orientation in space from archaeology, media archaeology departs from a criticism of the practice. Rather than an objective insight into the past through observation of its material remnants, Foucault argued that archaeology is (or should be) primarily an analysis of discourse in its archival form; he considers the practice distinct from geology or genealogy (Kluitenberg, 2008; Foucault, 1989). Such an archaeological approach has been criticised for having the tendency to freeze its object of study in rigid, preconceived discursive systems. To emphasize this, theorists like Zielinski (2006) have found it necessary to shift terminology from archaeology to anarchaeology, and eventually ‘variontology’ to abandon the original association altogether. This points once again to the importance of considering both machines and imaginaries in order to begin to understand media technologies. A solely material analysis that disregards the influence of discursive contexts is in danger of becoming an ‘apparatus history’ (Kluitenberg, 2008). A focus on the invisible work of software that overlooks the material terms involved ends up treating space as a neutral backdrop, rather than a context-constituting layer (Kitchin and Dodge, 2011).

Another take on space: the materialism of time

Language can only say things in terms of space. If we think space seriously, ‘space, not time, is the site of meaning (the meaning of life as well as the meaning of the proposition)’ (Lecercle, 2010: 22).

Lecercle’s article that I am citing here is based on a discussion of Arakawa and Gins’ ‘procedural architecture’ elaborated in their Architectural Body (2002) manifesto. Having revisited this text many times, I find their proposition of ‘a materialism of space, as opposed to the idealism of time’ (2010: PP) a relevant theoretical exercise for grasping the inherent spatiality of time-based media, algorithms etc.

Arakawa and Gins work from a position of paradox; they deny the centrality of time by using its own vocabulary. Their spatial materialism relies on the notion that bodies are situated in space. Each body is a source of metaphors; metaphors used to explain the body’s metamorphoses. Time is therefore seen as a derivation of (a bodily) space. It is implied then, that it makes sense to make sense of our lives through space (as opposed to: through time). In space though, there is no guarantee of commonplace concepts like transcendence or history. Space, viewed timelessly, shifts our attention to praxis and series of procedures. From any point in space, there is an unlimited number of paths to other places. With the Borges-like image of a garden of forking paths in mind, ideals of linear time and destiny are easily undermined. By rejecting this ‘idealism’ of time (Lecercle, 2010), Arakawa and Gins propose the idea of ‘reversible destiny’ playing out in an open present. Their notion of materialism is a form of thought in which space is central and time is merely an essence of it.

Space is only real-time

There are two main implications of this line of thought that I would like to link to the topic of time-based media and algorithms that I probed earlier; (1) a collapse of the linearity needed to stitch together a continuum from past to future through the present moment, and (2) redirecting analytical focus towards structures and procedures.

Firstly, in this materialist position, an open present replaces the past/future categorisation of time. Conceptually, this complements the way media archaeology and its diversely termed offshoots analyse current media ecologies. Zielinski’s excavations focus on fluctuating ‘heterogeneous energy fields’, stressing the open dimension of development. In the 'Deep Time of the Media' he articulates a rejection of historical linearity, or notions of progress (Kluitenberg 2008, Zielinksi 2006). Parikka opens 'What is Media Archaeology?' stating that it is a book ‘on the pasts and futures, the past-futures and future-pasts as well as parallel sidelines of media archaeology’ (2012: 5), which to me suggests that these temporalities might as well be left out. In this framework devoid of determinacy, it is consistent to view the metamorphoses of bodies as transformations, rather than tropes of innovation – the notion left out to make this shift is progress.

Today’s focus on real-time access and availability further undermines the relevance of understanding the world in the temporal terms of past/ present/ future. In both popular culture and professional contexts, the increasingly accurate remote access to the here-and-now through instant telecommunication has come to shape everyday realities. As Virilio has put it, the tele-presence of an object paradoxically replaces its own existence; its virtuality ‘dominates the actuality’ (Kluitenberg, 2008: 223). With the latest media technologies, we’re getting closer and closer to the present moment from farther and farther away. Think: live streams, instant messenger, surveillance systems, the immediacy of news coverage...

Real-time paces both technologies consumed for amusement and those employed in systems of observation and control. Departing from uses of information technology in military strategy, Kluitenberg articulates the implications of the real-time on time eloquently:

‘The three time forms of the decided action – past, present and future – were secretly replaced by dominance of real-time communications. The future has disappeared partly in computer programs (that predict and simulate things to come) and through this notion of real time that collapses temporal distinctions; when one identifies on the radar or video screen a threatening weapon in real-time, then this mediated present already contains the immanent impact of the projectile (the future).’ (2008: 224)

This brings us once again to the space of an open present. Much like Arakawa and Gins’ materialism, Kluitenberg’s argument points to an invalidity of temporal distinctions. The reason I find it interesting to make this verbose loop, is that we arrive in this same space through a different path. This time, it is through an analysis of the ways in which media technologies are used; the ways they structure our (tele-)communication renders time other than real-time obsolete.

Structures, construction and procedures

With time collapsed to the present and our understanding of the world situated in space, how can we begin to analyse time-based media? The materialist view of time of Arakawa and Gins puts emphasis on construction and procedure. From this position, the centrality of transcendental notions that rely on a linear temporality, like origin and destiny, is replaced with a focus on praxis in the present moment in space – a plane of immanence. Praxis, is seen as a series of procedures. From a subject’s perspective, praxis can be seen as a body moving through an ‘ever-changing sequence of domains, associating herself with some more closely than with others’ (Lecercle, 2010: 32). The subject is not strictly individual, and is engaged in processes, rather than being their source. Meaning, arising from these associations made in space is therefore both situated and in flux. The necessity of spatial exploration to grasp meaning that Arakawa and Gins call for is satisfyingly compatible with Haraway’s often-cited proposition of the situatedness of knowledge.

Put another way, the creation of meaning is seen as the outcome of a collective architectural practice. Now that the past is off bounds, architecture might be a more suitable alternative to archaeology anyway (could we speculate that something like ‘media architecture’ is the real-time version of its past-oriented archaeological precedent?). If we take the reversal of hierarchy between time and space seriously, time-based media are fundamentally space based. As such, we can view them as architectural bodies engaged in processes of self-construction and self-creation. Remembering that spatial metaphors used by time can be literally understood as movements in space, spatial exploration can be a way into these processes of self-construction, to gain an understanding. The practice of mapping – a method of documenting spatial exploration – is then a mechanism of creating meaning.

Mapping AI: time-based media are inherently based in space

The CzechInvest interactive AI map layer with which I opened, is an example of a project that could resemble one form of such a spatial exploration. In the straightforward activity of bluntly putting artificial intelligence on the map, associations with other activities that happen on a daily basis in neighbouring areas can be made. The two I mentioned above are road connectivity and proximity to universities; I will not go into an analysis of how these factors may be implicated, but they point to starting points of daily realities that shape the practices and imaginaries that make up what we understand as AI. Indeed Arakawa and Gins (2002) describe spatial exploration and mapping as a means of ‘layering of landing-sites’. Such landing-sites not only overlap with others, but are also subject to boundaries. This map, for instance, cuts off at the national border.

For now I’d like to leave it with the idea of viewing algorithms as architectural components operating in real-time, the meaning of which is created through structures and procedures. According to Zielinski (2006), media are spaces of action; this holds both from the perspective of the external human and nonhuman bodies that use them, as well as for the machines’ self-realising actions themselves. Drawing on the materialist position discussed here, metaphors that constitute the ‘algorithmic imaginary’ (Bucher, 2016), should likewise be understood spatially. ‘To run’, ‘to load’ or ‘to train’ are verbs that take time, but primarily describe activities in space. When used as labels for the praxis of an algorithm, their execution becomes the way in which it self-creates and through which others understand it.