Skip to main content
en
Close

Memorializing IHL violations in the age of AI: opportunities and risks for transitional justice

, Accountability / Analysis / Artificial Intelligence and Armed Conflict / IHL / New Technologies / Technology in Humanitarian Action 12 mins read

Paola Forgione  Former Legal Adviser, ICRC
Megah Bintang Mahadinah  MAS Candidate, Geneva Academy of International Humanitarian Law and Human Rights

Memorializing IHL violations in the age of AI: opportunities and risks for transitional justice

Memorialization, one of the five pillars of transitional justice, plays a vital role in preserving the memory of international humanitarian law (IHL) violations and gross human rights violations, supporting accountability, and preventing recurrence. Building an accurate and inclusive collective memory of past atrocities is a resource-intensive process, fraught with challenges ranging from limited access to archives and conflict zones to the risk that dominant narratives overshadow marginalized voices. Artificial intelligence (AI) and data-driven technologies are increasingly used to ease these burdens, from cross-referencing dispersed records to expanding public access to memorialization projects.

In this post, Paola Forgione and Megah Bintang Mahadinah examine the opportunities and risks that AI presents for memorialization processes. The authors argue that while AI can accelerate documentation and broaden the reach of memorialization efforts, it cannot substitute for the human judgement, empathy, and contextual understanding that transitional justice ultimately depends on.

In the aftermath of conflict, societies must confront the wrongdoings of the past while also navigating profound political change and a weakened rule of law. Delivering accountability and reparations for serious violations of international humanitarian law and other international crimes, ascertaining the truth about the circumstances in which they occurred, and securing a lasting peace, all at once, is a formidable undertaking.

Memory preservation is essential to transitional justice. Collecting and publicly honouring accurate and comprehensive accounts of the harm inflicted on all victims and survivors can restore their dignity, promote societal healing, and prevent the recurrence of violations, especially when that memory is transmitted to future generations.  Memorialization projects and campaigns – such as historic sites, monuments, museums, artworks, exhibitions, performances, commemorating events, and documentary films – protect collective memory from revisionism and denialism, enabling society to learn and regain ownership of its own history.

Memorialization helps fulfil specific IHL obligations, such as investigating and punishing war crimes, as the information preserved through memorialization can be relevant in criminal trials to punish those responsible for war crimes, even decades after they occurred, as such crimes are not subject to statutes of limitation. Memorialization also includes commemorations, public acknowledgement of grave violations, formal apologies and expression of regret towards the victims, which are all recognized forms of  reparations for IHL violations. Because of its central role in transitional justice, memorialization is considered one of the five pillars of transitional justice, together with truth, justice, reparations and non-recurrence.

Challenges of the memorialization processes

Memorialization sits at the intersection of two processes: rigorous documentation of violations and their shaping it into a collective narrative. Memorialization entails establishing past facts through in-depth analysis of official and unofficial records, such as reports, letters, diaries or photos, which are often treasured in public and private archives. A mapping exercise then allows researchers to identify the locations of these relevant archives, which may sometimes be scattered across different countries, as well as the access methods required, such as special authorizations or payment of fees.

While documenting facts establishes the foundation of memorialization processes, making that record into something a society can carry forward requires a different kind of work. Those engaged in memorialization projects must adopt dialogic approach and dialogic truth, where a public memory is not only a passive reception of facts but an active societal confrontation of the past legacies and responsibilities. Memorialization projects demands listening to survivors to capture and convey their experiences, as “the voices of the victims (…) must play a key role in the construction of memory, thereby avoiding the distortions that the perpetrators may attempt to impose”, while at the same time inviting viewers and audiences to join a conversation on the causes and legacies of past crimes.

All these activities present practical, psycho-social and political challenges. First, they demand a lot of human time and financial resources, for example to examine thousands of documents or to meet the victims in different countries, which can be especially burdensome for community-based projects. Second, regardless of the resources available, access to certain areas may be impossible, whether because of the security situation or because the authorities in charge do not authorize it. Third, sustained exposure to first-hand accounts of violations may result in vicarious trauma for those collecting the information. Fourth, the project’s outreach may be limited to those living in or travelling to the area where the memorialization site is located. Finally, experts have highlighted the risks of the politics of memory, whereby competing narratives over victimhood mean that the memories of some groups thrive while others are marginalized. This is the case with state-sponsored memorialization projects that adopt a “top-down” approach, leaving some communities underrepresented or even “weaponizing” memory against certain groups. Allowing different communities to express their stories freely is essential to the success of transitional justice projects.

How can AI help?

One of the challenges of memorialization process is the need to analyze and cross-check a large amount of information. Artificial intelligence (AI) and Data Driven Technologies (DDTs) promise speed and scale, a chance to accompany the laborious and emotionally exhausting process of data analysis towards a more efficient process of truth-seeking and evidence-gathering, both for formal accountability measures and for bridging that to public consciousness. In doing so, they help in collecting, organizing, and analyzing records spread across various archives. They tackle the challenge of sifting through vast collections, identifying patterns, and triangulating open-source material.

This process helps reconstruct incidents that might otherwise remain hidden within fragile or inaccessible records, reducing the risk of vicarious trauma for practitioners. The work of agencies and civil societies organizations such as Forensic Architecture, Bellingcat and Mnemonic, utilizes AI and DDTs for photogrammetry, geospatial analysis, multimedia synchronization, and open-source investigation, in order to translate that complex and intricate evidence of atrocities into public-facing forms. Their work also promotes accountability, cross-checking claims of IHL violations and gross human rights violations, combating denialism and contributing to preventing impunity.

In addition to effective investigation and documentation, these technologies advance the representational aspect of memorialization. Where access to memorials or archives is limited both for practitioners and the general public, digital interfaces expand outreach and engagement, enabling memory to go beyond specific sites or institutions. Aforementioned agencies and civil society organizations, among others, adhere to open-source principles, which may foster transparency in their methodologies, but also help aggregate dispersed and fragmented records and make them more widely accessible despite geographic and political constraints. However, the role can never replace the actual access to places and to people that practitioners – such as human rights and forensics experts, journalists, historians, and archivists – need in order to document IHL violations and other gross human rights violations. An exclusive use of AI techniques erodes this human work.

Finally, by leveraging digital tools, the memories and experiences of those affected by past atrocities remain visible and accessible to people across the globe. This can further help to bring out marginalized stories to the surface. AI and DDTs enabled content increases the visibility of transitional justice processes, broadens participation and awareness around trials, events, and relevant justice measures. Interactive digital installations, such as Colombia’s Truth Commission digitalization and chatbox, Chile’s AI-supported archival initiatives Nuestra MemorIA, and DDTs-powered Phoenix of Gaza XR, show how technology can increase visibility by enhancing access and preserving memory beyond the limits of physical sites.

Despite these benefits, the question of whose voices are amplified, silenced, or overlooked remains relevant to digitalized memorialization projects. AI and DDTs serve as valuable tools that can alleviate the material challenges of memorialization, but they cannot replace the contextual understanding on which meaningful memorialization ultimately depends, and they are not a perfect fix to achieve justice.

The risks of using AI in memorialization projects

While AI may help address some challenges of memorialization, like in the examples above, its use could also lead to new ones. AI may undermine the “do no harm” principle, exposing victims to emotional distress or confusion. For example, reconstructing missing persons’ faces through AI may be a shocking experience for families, especially since the reconstructed features may be merely speculative and hypothetical. AI is also trained predominantly on datasets that overrepresent white faces, potentially misrepresenting or marginalizing other groups. Victims may also suffer the consequences of lack of transparency or clarity about personal data use. For example, in contexts where sexual violence-related service providers are required to refer all cases to the police, victims’ accounts within a memorialization project might reach third parties without prior information and consent, and expose them to retaliation. Similarly, there is not enough transparency on the way AI companies manage requests from government authorities to access their datasets and what impact it can have on victims.

To mitigate these risks, memorialization projects must be grounded in comprehensive policies on data protection, informed consent and cybersecurity before using AI, and update these policies as needed, even in the absence of specific legal requirements to do so. These policies must focus on the “do no harm” principle and rely on consultations with local victims’ groups, in order to adapt to the social, economic and political context. They must also include a human rights and ethical impact assessment, to be conducted prior to the use of AI. From a purely legal perspective, some memorialization projects must also comply with the 2024 EU Regulation on Artificial Intelligence, which inter alia tries to strike a balance between the risks of AI and its use in support of victims. Its scope includes AI systems whose providers and deployers are based outside the EU, if the output is used in the latter. Furthermore,  the privacy rights of individuals and the protection of personal data are regulated by the  Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, the first legally binding treaty of this kind for member states of the Council of Europe. Experts have also proposed a framework for a global treaty on AI and human rights.

Memorialization projects supported by AI are often visually appealing, engaging and easily accessible. They raise awareness about violations that could otherwise be forgotten or erased by state narratives. The downside is the risk of treating memorialization as a standalone response to violations rather than part of  a broader, complementary transitional justice, which also requires states to investigate and prosecute in order to clarify facts, punish, and allow victims to seek reparations. Memorialization offers instant recognition when legal routes are slow or unavailable. However, it works best alongside other measures and not as a substitute.

Instead of replacing accountability, the practitioners of memorialization should envisage its use as AI-affected evidence in future trials: Will it be admissible in court? Is the information credible? Notorious problematic scenarios include the “black box” issue, which is generally defined as the impossibility for humans to properly understand the decision-making process of an AI system (e.g., what is the model’s internal logic for identifying a munition in a specific video?) and the chain of custody of digital evidence (e.g., how has a video been protected from manipulation?). Furthermore, evidence obtained through violations of privacy or other human rights may be inadmissible in court. While a universal legal framework is lacking, some initiatives aim to establish standards for AI use in trials, such as the 2025 UNESCO Guidelines on AI in the Courtroom, the 2022 Leiden Guidelines on Digitally Derived Evidence, and the 2020 Berkeley Protocol on Digital Open Source Investigation.

Finally, a general problem with using AI in memorialization is its inability to convey empathy to victims or to make culturally and religiously sensitive decisions. Whatever gains AI may offer in accuracy and efficiency, they do not outweigh the importance of tact and emotional understanding. Removing the human factor affects not only the victims, but also those supporting memorialization, such as humanitarians and human rights experts, for whom the relationship with victims is not one-way, but also a source of humanity and purpose.

Conclusion

The use of AI in projects aimed at preserving the memory of IHL violations and gross human rights violations might present  opportunities, such as potentially reducing the amount of human time needed to gather information, reaching global audiences, and reducing the risk that some groups’ experiences are overshadowed by state policies. However, using AI could just as easily lead to unintended consequences, from violating human rights to treating victims as mere items.

Meaningful human control throughout the memorialization process mitigates these risks. Unlike AI, humans can assess the particularities of each context and engage in open and confidential communication with victims, ensuring that existing data protection, human rights and cybersecurity policies remain effective. Humans can also verify the accuracy and credibility of AI outputs, limiting the impact of issues such as the “black box” problem. Finally, and fundamentally, human engagement is what makes the entire transitional justice process meaningful for everyone involved in it.

 

Author’s note

The authors would like to thank Sophia Devlin, Alexander Kriebitz, Thomas Unger and Anna Myriam Roccatello for their precious advice and for generously sharing their expertise with us.

 

Related articles

Share this article
You should also read these articles