Ця стаття також доступна українською мовою тут. This post is also available in Ukrainian here.
This summer, evidence collection at the International Criminal Court (ICC) took a major step forward with the introduction of OTPLink. The new online platform provides a more accessible avenue for witnesses of international crimes – including rape, torture, murder, and enslavement – to directly submit information to prosecutors in real-time. OTPLink complements the ICC’s larger digitalization initiative, Project Harmony, which uses cloud computing capabilities and artificial intelligence to review large quantities of complex information and evidence. Together, these initiatives promise more accessible information submission, streamlined and secure information processing, storage, and preservation, as well as more efficient analysis of complex information. Ultimately, the tools should aid in the international criminal justice project by granting victims and witnesses greater agency and accelerating the Court’s truth-seeking function.
For all their promise, the new systems also have drawbacks and present potential pitfalls. Illiteracy, language barriers, and lack of digital connectivity limit who can access OTPLink in ways that may be mitigated through dedicated ICC outreach and translation software. Project Harmony’s machine learning tools will assist in identifying patterns of violence and individuals, so reducing issues of algorithmic bias is crucial for the program to succeed. Finally, there is a real risk of individuals or organizations submitting false or misleading information to the database as evidence, which means any machine learning models should be subjected to rigorous digital and manual oversight, and models should be trained against such misleading input data.
OTPLink
On May 24, the ICC’s Office of the Prosecutor (OTP) launched OTPLink, a new online platform that enables victims, witnesses, civil society organizations, and governmental bodies to directly submit information or evidence related to international crimes to the OTP. The new interface is a centralized system with a “single-access point” that replaces and consolidates various older information-gathering systems including OTP Pathway.
OTPLink was created primarily for receipt of Article 15 submissions, though it also accepts and processes submissions related to Situations under investigation. Article 15 communications are information that individuals or organizations send to the OTP either before or after the opening of a Preliminary Examination. Arguably, the platform may have a greater significance for Situations under Preliminary Examination, the stage during which the OTP determines whether a full investigation is warranted, or investigations where the OTP investigators are unable to travel to the country – such as in today’s Afghanistan, where security and cooperation issues hinder information and evidence access. In such instances, the success of the OTP examination or investigation may depend on the information and evidence submitted by third parties, and OTPLink may become a vital source – and perhaps the only information and evidence sharing tool – connecting the OTP to witnesses, victims, and other stakeholders. In the past, the OTP created separate and temporary online forms to receive Article 15 information, but OTPLink unifies that disjointed process into a single system, using “a digital chain of custody that collects and preserves information” and produces “a dependable and tamper-proof record of the collection and handling process.”
OTPLink is simple and user-friendly. The platform allows users to submit their information or evidence anonymously or with their name and contact information. When making a submission, users can provide details about the incident, including its name and the date it occurred. The platform also provides an interactive online map allowing users to pinpoint the incident’s exact location. Finally, OTPLink allows users to upload up to 1000 files with nearly 4GB earmarked for each submission. The only language of the online interface is English.
Project Harmony
Though an innovation in and of itself, OTPLink represents just one part of a larger effort by ICC Prosecutor Karim Khan to modernize how the Court collects and processes potential evidence. In February, Khan announced Project Harmony, the OTP’s ambitious evidence management platform that harnesses artificial intelligence and cloud-based capabilities. Project Harmony touts features including rapid pattern identification, automatic translations, facial identification, image enrichment, translations of media files, targeted searches of source material, automated transcription, and video and image analytics. While none of these innovations are novel on their own, when combined in this way they could prove invaluable for the OTP’s efficiency in collecting, storing, preserving, analyzing, and reviewing evidence. For example, facial identification tools could assist investigators to develop leads, making it possible for them to more quickly compare multiple images that may depict the same person.
While scholars and practitioners have called for the creation of a streamlined data repository for information and evidence for several years, budget challenges stymied progress. Both Project Harmony and OTPLink were largely funded through a European Union grant to build “a modern, efficient evidence management system” and the “development of artificial intelligence and machine learning tools to analyse and process data.” Microsoft, Accenture, and their joint venture Avanade have supported the OTP to develop the tools encompassed by Project Harmony, including that of OTPLink.
Issues to Monitor
The Need for Accessibility and Outreach
Despite OTPLink’s improvements in accessibility, some of the most important stakeholders, including victims and witnesses, may face challenges in accessing the platform due to illiteracy, language barriers, or lack of internet access. They may also lack trust in the platform itself, having doubts about the safety and security of their submission or the ultimate adequate evaluation of their information. The ICC Outreach Unit and the OTP should prioritize informing stakeholders about OTPLink in their existing outreach activities and consider conducting new awareness raising campaigns on social media, TV, over the radio, and in magazines, in local languages targeting impacted communities.
Unfortunately, the ICC’s outreach efforts are a mixed bag across its Preliminary Examinations and investigations. For example, in jurisdictions like Côte d’Ivoire, Mali, Sudan, and Uganda the ICC Outreach Unit has conducted positive activities including holding information sessions about the Court’s mandate and jurisdiction and kept impacted communities and victims informed about the progress of specific cases under the ICC’s consideration through radio and television programs in local languages.However, in other in other jurisdictions like Bangladesh/Myanmar, Burundi, the Central African Republic, the Democratic Republic of the Congo, Georgia, Kenya, Libya, Palestine, and the Philippines, the Outreach Unit’s activities have been limited to conducting online activities, working with academia, journalists, civil society and legal professionals, and producing information sheets and videos.
In the remaining jurisdictions like Afghanistan, Ukraine, and Venezuela, there have not been meaningful outreach activities. As an example, the OTP created an online form, in Dari and Pashto, for submission of information during the later stage of the Preliminary Examination in Afghanistan. However, few individuals – save for interested lawyers and civil society organizations – knew about the form. While the form was available on the ICC website, to our knowledge, the ICC never advertised it within the country.
With the launch of OTPLink, it is paramount for the OTP to address the challenge of accessibility and awareness and to ensure that its new platform reaches everyone. Engagement with civil society will be key. Local and international civil society and human rights organizations can work to communicate realistic expectations about what the ICC can and cannot do with individual information and evidence. These organizations are often better situated to conduct outreach activities and information sessions about the new platform, informing the affected communities, victims, and witnesses about how they can locate and use the interface effectively. Furthermore, the OTP may collaborate with civil society and other stakeholders to increase their capacity, knowledge, and awareness about the ICC’s evidentiary standards and technicalities and how it engages with impacted communities to help obtain more relevant and admissible evidence through the new platform. The OTP may consider holding trainings on how the new platform should be used and developing accessible and easy-read guides on evidentiary standards and evidence collection techniques in compliance with applicable standards.
Additionally, to enhance the platform’s accessibility and to maximize victim and witness participation, the OTP may consider creating the same online interface in local languages (and beyond just English) for each Situation under Preliminary Examination or investigation. At a minimum, the OTP may link the platform to an online automatic translation tool, like Google Translate, to invite non-English-speaking stakeholders to submit their information and evidence in local languages and integrate text-to-speech software which could address challenges from illiteracy.
Fairness and Adversarial Examples in Submissions
Among the risks inherent in the use of machine learning are fairness and adversarial examples, both of which stem from the fact that machine learning tools are only as good as the data inputs they receive. The risks can be addressed through continual and rigorous training of machine learning models as well as expert authentication and verification.
Issues of fairness, algorithmic bias, or discrimination are omnipresent in machine learning tools, wherever they are employed. In this context, issues of algorithmic bias might mean that people of a certain race are identified more accurately than people of another race through use of facial recognition software. Microsoft, in particular, has worked for years to create “fairer” facial recognition algorithms that would not produce disparate impact for particular segments of the population, including those with darker skin. Such improvements are said to have drastically reduced rates of misclassification, and other changes have been made to safeguard against bias. Nonetheless, algorithmic systems remain only as good as the data on which they are trained, so the data input into the Project’s models are critical. Thus, a diverse set of training data, bias mitigation techniques, and regular evaluation of the machine learning model are key.
Given the context in which the database is operating, another potential concern is the possibility of adversarial examples – inputs to machine learning models that, when fed to the model, lead it to misclassify examples. For instance, a machine learning model may be intentionally misled to incorrectly classify or identify an object or person – and often with a high degree of certainty that the classification is indeed correct. In an era of deep fakes and synthetic media technology, and a context in which State actors are under investigation, the use of small, intentional data perturbations cannot be ruled out. Nor is it unrealistic to imagine a sophisticated, State-sponsored disinformation campaign that provides incorrect or misleading data. For example, it would not be difficult to conceive of one uniform being digitally exchanged for another, or a face being altered to look like someone else or no one at all. Without a doubt, Microsoft, Accenture, and Avanade are working closely to protect against adversarial examples through building and training their own models. Still, project developers should consider digital and manual detection of adversarial examples going forward. And care should be taken to ensure both that fictional information is not evaluated as true and that skepticism of such fictional information does not lead to the “liar’s dividend.” When the information arena is saturated with misinformation, the mere threat or suspicion of information having been modified may detract from the truth as much as an actual alteration would. This means that the threat of misleading data should not be overblown.
With the anticipated influx of Article 15 communications, machine learning tools will clearly help in processing massive amounts of new information and data. Nonetheless, the human aspect of information and evidence processing and review remains of supreme importance.
Conclusion
The OTP’s introduction of OTPLink and Project Harmony are undoubtedly progressive steps for international criminal justice. OTPLink will significantly contribute to evidence collection and information sharing with the goal of preventing impunity for the worst international crimes. OTPLink will also increase the direct engagement of external stakeholders with the OTP, enhance accessibility for potential victims and witnesses, and provide a tool for gathering evidence and information that might not otherwise be possible by OTP investigators. Such a tool will give survivors of, and witnesses to, some of the worst international crimes, including rape, torture, murder, and enslavement, unprecedented agency. However, OTPLink’s goal will not be fully achieved if the OTP does not invest in outreach activities and translation services to ensure that the platform and its effective use reach all stakeholders – including affected communities and potential victims and witnesses – across the Situations under Preliminary Examination and investigation.
Similarly, with the aim of harnessing the latest technology, Project Harmony will enhance the efficiency of overall evidence management at the ICC, enabling the OTP to strategically use its scarce resources and analyze large volumes of data and evidence at a lower cost and in a shorter time. However, Project Harmony is also susceptible to several risks, including issues of fairness and adversarial examples, that need to be dealt with via continual work on the model and improving authentication capabilities.