Building a Collaboration to Protect Human Rights Defenders
Bringing together the public and private sector — UN Human Rights and Dataminr — resulted in a unique AI model for detecting attacks against people working to safeguard human rights.
By Amy Lynn Smith — Writer + Strategist
By definition, human rights defenders stand up for the rights of people who are persecuted and whose rights are in peril. Humanitarians, journalists reporting on human rights violations, attorneys representing victims of violence, as well as activists are all examples of human rights defenders. But the sad reality is that defending human rights can mean putting your life on the line every single day.
One of the responsibilities of the Office of the High Commissioner for Human Rights (UN Human Rights) is to watch for and protect against attacks on human rights defenders. In the Human Rights Indicators and Data Unit, there’s an increasing focus on using data to strengthen both monitoring and reporting of threats to human rights.
According to Marc Titus Cebreros, a Human Rights Officer and statistician in the unit, until recently it’s been a challenge to gather the necessary data to create accurate reports on threats and attacks against human rights defenders. Public records can be confusing. Are two news articles that refer to an incident in different ways really about the same attack? How do you identify whether an act of violence was against a human rights defender if an article doesn’t explicitly say so? The list of challenges goes on. Cebreros and his counterparts would use on-the-ground monitoring, media reports, Google searches about names and incidents, information from civil society organizations, and more in an effort to compile the data. But it was manual and painstaking work to cover not only killings but also disappearances, torture, kidnapping, and other harmful acts, he says.
“We needed to find a way to make this data collection process more efficient for everyone concerned,” Cebreros explains, “which means we needed to automate some parts of this process.”
Bringing together two experts with the same mission
Enter Dataminr, whose artificial intelligence (AI) platform detects the earliest signals of events and emerging risks using publicly available data. The UN was already using Dataminr’s First Alert platform through a partnership with UN Global Pulse (UNGP), the Secretary-General’s innovation initiative, to support the work of various members of the UN family. Still, UNGP and Dataminr sought ways to deepen the impact of this collaboration.
UNGP often acts as an enabler and convener of partnerships — in this case, brokering relationships between Dataminr and other UN agencies: “planting seeds and watching them grow,” says Gabriella Ginsberg-Fletcher, the Partnerships Lead at UNGP. She emphasises that after an initial push to get UN Human Rights on the First Alert platform, the relationship between Dataminr and UN Human Rights developed organically.
Dataminr was piloting a new program designed to double down on its commitment to social good by pairing its best-in-class AI team with experts working on interesting and meaningful projects. UN Human Rights recognized that Dataminr had exactly the expertise it needed, and their collaboration began in April 2022.
There was a firm understanding of the importance of principle-based innovation and partnerships from the beginning of the collaboration between the UN and Dataminr. Using its experience in AI models, Dataminr worked closely with UN Human Rights to understand its needs and develop a model that would capture exactly the information necessary to detect attacks on human rights defenders. The goal was to give UN Human Rights the ability to mobilise rapid responses or make proactive policy decisions accordingly.
Shihao Ran, Ph.D., and Di Lu, Ph.D, the scientists at Dataminr who led the project, set out to build an AI model that would identify and classify attacks against human rights defenders.
The most crucial steps were collecting data and then using that data to train an AI model. Ran says it was a challenge to get enough relevant data to train the model to identify important information from attacks on human rights defenders. The team eventually trained the model to extract more than 10 attributes from articles in online news outlets. These included specific references to the type of human rights defender — such as humanitarians, journalists, and trade unionists — reports of violence and their location, kidnappings or arbitrary detention, whether the perpetrator was a state or non-state actor, and more.
“AI models are trained on data labelled for the task at hand,” Ran says. “They need to learn from examples, just like a human would. But machines aren’t good at reasoning the way humans are, which makes it more challenging. That’s why getting enough data is so important.”
Showcasing the strengths of a successful public- and private-sector collaboration
The collaboration between Dataminr and UN Human Rights was close-knit, with Dataminr sharing each round of data collection for UN Human Rights’ feedback to ensure the correct data was being gathered. Dataminr incorporated that feedback, iterating on the model until it was ready for delivery on December 9, just one day before International Human Rights Day.
“Both parties were very engaged throughout the process,” Ran says. “And everyone was very open in terms of questions and issues we identified at each step.”
The final deliverable has two components. The first is a manuscript documenting all processes, steps, and instructions from data collection to model training, so UN Human Rights can facilitate future projects using a similar process. The second is a repository containing all the annotated data, data preparation and processing scripts, model training and evaluation instructions and scripts, and more.
Ultimately, just as Dataminr trained the AI model to carry out the task UN Human Rights needed to be done, the team provided the resources for UN Human Rights and other members of the UN family to implement and tailor similar models on their own.
“The project sparked ideas for similar collaborations, perhaps looking at hate speech, other aspects of civic space, and more,” Cebreros says. “Dataminr’s approach to work is, ‘We can do this for you, but we need to do it with you and we need to find a way to sustain it beyond the initial collaboration — and make sure it’s impactful and fills a real gap.’ I think that’s quite promising.”
