Machine Learning, access to justice and the T72B3

Much of the use of artificial intelligence in legaltech may be exciting in its implications but is pretty mundane in its actuality. It is hard for the wonders of e-discovery and document review to get the pulses racing of all but the most dedicated techie. But, just occasionally, you come across a genuinely gob-smacking use of machine-learning in collecting evidence in an access to justice context. As recent examples, check out three videos published by London-based Forensic Architecture on Russian involvement in Ukraine, ‘Triple Chasers’, tear gas grenades and herbicidal warfare in Gaza.

Forensic Achitecture is based at Goldsmiths, University of London. It takes its name from what it describes as ‘an emergent academic field … the production and presentation of architectural evidence—relating to buildings, urban environments—within legal and political processes.’ That might seem a bit dry but it really isn’t. ‘We investigate state and corporate violence, human rights violations and environmental destruction all over the world. Our work often involves open-source investigation, the construction of digital and physical models, 3D animations, virtual reality environments and cartographic platforms. Within these environments we locate and analyse photographs, videos, audio files and testimonies to reconstruct and analyse violent events.’

So, what we have is a very 21st century human rights organisation highly oriented toward modern media. ‘Contemporary conflicts and human rights violations increasingly take place in urban areas, amongst homes and civilian neighbourhoods. The nature of urban war is such that parties in conflict wilfully blur the line between civilians and combatants. At the same time, those environments have become densely media-rich. The proliferation of smartphones has meant that human rights violations in conflict have never been so thoroughly documented. However, such cases can be complex, and understanding what has taken place can be challenging. Architectural analysis and digital modelling techniques enable us to unravel that complexity, and to present information in a convincing, precise, and accessible manner–qualities which are crucial for the pursuit of accountability.’

The result is an organisation that interfaces with media at both ends of the process – the collection of evidence and its presentation. Forensic Architecture (FA) has received recognition for both. It was nominated for the most prestigious British modern art award, the Turner Prize, in 2018. FA prepared a video explaining its work that is also with looking at. FA’s belief is that ‘art is not just functional’ but that its techniques can help to reveal the true meaning of events as mediated by observers who film, recall or record them.

FA’s most recent report is that on the Battle of Iloviask in Ukraine which was produced for the European Human Rights Advocacy Centre to use before the European Court of Human Rights. ‘Using machine learning and computer vision, and building on the work of dozens of open source investigators, reporters and media outlets, Forensic Architecture gathered together the most comprehensive collection of evidence for the presence of Russian military personnel and hardware throughout the battlefield. In total, we gathered evidence of almost 300 Russian military vehicles around the Ukranian towns of Ilovaisk and Luhansk. We presented that evidence in an interactive cartographic platform.’

The smoking gun for Russian involvement was proof that a particular tank, a T72B3, was present in Ukraine at a time when it was only in service with the Russian army but where any Russian military engagement was denied. The evidence was distilled by machine learning that allowed search of thousands of videos and photographs.  FA narrowed its parameters in the process of evidence collection from looking for military vehicles, particularly tanks, to this particular one. Identified frames of interest were then analysed by human experts. 

A similar process was used for machine learning in relation to proof that an American company, Safariland was using tear gas unlawfully on the US border with Mexico. FA had only a limited number of actual pictures of the ‘triple chaser’ grenade but it was able to generate artificial pictures to create ‘a synthetic training set’. The investigation led to the resignation of Safariland’s owner, Warren B. Kanders, as vice chair of the board of trustees of the Whitney Museum of American Art which had commissioned the investigation.

So, mid-August provided a very dramatic demonstration of the different ways in which AI can impact on the law.  Over 3000 delegates poured into Florida’s Disneyland for the annual ILTACON (International Legal Technology Association) extravaganza to hear the latest commercial applications of technology. Meanwhile, in the humbler surroundings of London’s Greenwich, Forensic Architecture was pushing the boundaries of what machine learning can show us in the context of some of the hardest access to justice and human rights cases. 

Leave a Reply