AI Generated Historical Maps : Complexities of Borders
In my last article "ML and History : Trust and Its Implications", I discussed the question of trust as an arbiter of history: how can we make sure that AI models that work with sources from the past do not come to fictitious or spurious conclusions? As a working project – developing new technological approaches, but also sharing those results along the way – one of the key tasks identified so far has been to develop maps that can accurately demarcate borders. The most obvious risk is that AI-generated maps might show the “wrong” border(s) for a particular time; given the use and misuse of historical “borders” in contemporary political discourse, such a possibility could have especially grave implications. At the same time, there is a more fundamental issue lurking beneath this threat: how do maps fix certain pre-conceptions of territory and political organisation? And in its ability to cycle through information at ever greater speeds, might AI help us avoid some of those traps?
Mapping Complexity: How Borders Function Beyond Mere Lines
What is a border?
At first glance, this seems like a question with an obvious answer: in a world divided into nation-states that claim sovereignty over their own territory, and so delineate it from areas beyond that claim, the border is simply what divides one state from another. Yet in recent years, specialists have developed increasingly nuanced ways of thinking about borders that encompass multiple disciplines (anthropology, geography, history, sociology). The operative issue is less the line on a map, but what that line can be made to do. That is true even in the modern world, albeit “national” borders are usually treated identically in mapping software like Google Maps, regardless of their status in practice. Maps have the great benefit of clarity, but must necessarily compress or eliminate complexity in conveying information.
Understanding Border Dynamics and Its Scenarios
At one end of the spectrum, the border between North and South Korea is demilitarised and fortified on both sides; at the other, the border between France and Belgium – both Schengen members – is essentially open to travel without restrictions or checks. What this does not entail, however, is that it is meaningless: despite the ease of movement, the Franco–Belgian border remains a jurisdictional divide that applies only insofar as both countries have agreed to harmonise their policies, in this case through the EU. What borders should do was likewise the crux of the thorniest, and longest-lasting, negotiations around Brexit: which controls should apply between Northern Ireland and Ireland, and which between Northern Ireland and Great Britain?
The complexity of those discussions is testament to the emotive, affective capacity of borders as something perceived in thought as much as marked in the ground. The self-sustaining interplay between the modern nation-state and its spatial extent – its territoriality – makes the external border not merely one aspect of sovereignty, but its defining feature. It posits “here” against “there”, “us” against “them”, even for people who have never physically interacted with a border (witness the Trump administration’s long-running saga around its “wall” at the US border with Mexico). To see a border as merely a line is to underplay the power it can hold as a cultural yardstick, and its potential as a form of unchanging self-definition.
Can this be represented historically?
This excursus into the experience of modern-day borders may initially seem to have little to do with AI or machine learning. Two competing pressures, however, emerge in terms of using such technology to develop new directions in historical thinking. On the one hand, any representation of the past is necessarily shaped by our present-day conceptions, and the difficulty of escaping them; given Historica’s ambition to create tools that can be used widely, and not just by specialists, there is seemingly only so far that this question can be problematised without making the output ill-defined or unclear.
On the other hand, these contemporary notions of the border – which become ever more prescriptive in their use of technology – do not parallel their role fifteen or twenty years ago, let alone further back. The Roman Empire, for instance, used limes as way of quite literally delimiting its borders; its successors across Europe, by contrast, had much more varied, and often more fluid, ideas of how their polities functioned as territorial units. The point is that postulating where those borders can be legitimately drawn relies on sources, both written and material, that were fashioned in those differing contexts.
As I suggested in my last article, significant parts of the past are “blank space” to the extent that any “positive” attestation is lacking about – to take a particularly acute example in creating maps – political control over a certain place. Alternatively, a historical map might seem to contradict a chronicle record. But there are many possible reasons for that disagreement, ranging from propagandistic claims to differing conceptions of the border, and from aesthetic choices to scribal error. However such sources are eventually reconciled, those reasons are less important than the fact that the reconciliation becomes fixed when it appears on a map.
What can AI do differently? Redefining Historical Representation
Given these challenges, the issue is what AI can do differently in bridging the gap between past conceptions and the questions Historica’s different audiences might pose of sources. Outside of academia, those questions may well be highly localised. For someone wanting to explore their family history, for example, it could be: which polity was a particular village part of in a given year? Printed historical atlases may not provide that information, so it can only be surmised. Yet to a large extent, digital mapping has already solved this problem; a user can change the scale at will, zooming in and out to show different information. While each layer necessarily still requires a particular fixity – depending on what the software chooses to display, and the importance it attributes to certain names – it allows information to be much more immediately sifted.
Nonetheless, I would also suggest that the flexibility of AI extends beyond shifting the presentation of information. Earlier I highlighted the apparent balance to be struck between the problematisation of a particular model and its utility. But what if the integrative capacities of AI allow such problematisation to be situated in the uses – or styles – of maps? As the Historica project progresses, those styles will become more varied, extending beyond political boundaries to include different distinctions (such as religious demographics), or overlaying AI-generated maps on to pre-existing software like OpenStreetMap. The real promise of both the digital humanities and AI, however, lies in not having to be restricted by “traditional” ideas of representation inherited from print (e.g. lines, block shading).
Insights from Innovative Cartography Projects
Recent digital cartography projects make this plain: for example, “Mapping the Enlightenment”, a joint initiative between UCL, the UK National Archives and the University of Athens, combined spatial signifiers with “flow lines” to trace the movement of travelling scholars in early modern Europe. In a different vein, Harvard’s “Imperia” project seeks to build a spatial history of the Russian Empire that pushes back against the apparent simplicity of “space”, to interrogate how sources of all kinds fit into the shifting parameters of that space across time.
My point is that the generative capability of AI models – coupled with how their outputs can be digitally rendered – changes what is possible in representing history. This is especially true if the model can respond as necessary to new information that disrupts its previous interpretation of a particular situation. But it also allows space to be changed through time, such as through blurring boundaries or using moving images. The popularity of YouTube videos that track borders through time is testament to how this shift can radically alter people’s perceptions of the past, away from merely static “snapshots” to a more dynamic understanding of historical evolutions. To go back to my remarks at the start of this article, however, only by attending to the flexibility and fluidity of the source material can we establish a sense of how not only the “objects” of history changed over time, but how those objects’ framing concepts (“the border”) did as well.