We at HURIDOCS are a curious bunch. No, we don’t mean we’re strange or unusual, although we can be a bit geeky — and proud of it! What we mean is we regularly seek new ideas and perspectives related to our work supporting human rights defenders with documentation tools and strategies, because we believe that doing so makes us much more effective.
One recent example of this commitment in action? Over the last several months, we’ve collaborated with a group of Google.org Fellows to strengthen access to public human rights information through machine learning.
In 2019, HURIDOCS was selected as a Google AI Impact Challenge grantee. In addition to the financial support, it opened the door for us to take part in the Google.org Fellowship, which allows Google employees to do pro bono work full-time with a grantee.
The Google.org Fellows lent us their skills in software engineering, product management and user experience design to help us automate certain tasks within Uwazi, our open-source platform for hosting databases. By doing so, we lessened the burden on the activists and advocates who use Uwazi to create collections of international laws and rulings related to human rights.
As a result, the information contained in these collections is more up-to-date, consistent and accessible, and the people who maintain them have more time to dedicate towards other important work, such as advocacy.
Our time with the Google.org Fellows has largely come to an end. Together, we developed a solid foundation that HURIDOCS will continue to build upon, not just for human rights defenders who curate public information, but also for our partners who use Uwazi to document human rights violations.
So, what did our HURIDOCS team make of the experience? We spoke with three of our colleagues to find out: Chief Technology Officer Jaume Cardona, based in Central America; and Project Manager Mila Guilhem and Machine Learning Specialist Natalie Widmann, both based in Europe.
Wondering what the experience was like for Google.org Fellows? Check out this question-and-answer post with three of them!
Before the Google AI Impact Challenge and our participation with the Google.org fellowship, what experience had you had with machine learning at HURIDOCS?
Jaume: My journey in machine learning as applied to human rights started four years ago when Natalie approached us proposing to explore it as part of her internship. Until that moment I had a rather rudimentary knowledge of machine learning and zero practical experience. As Natalie’s supervisor and the chief technology officer, I had to ramp myself up in these technologies and learn as much as I could to properly support her and guide the process of adoption within our organization.
The main lesson I learned is that machine learning is not a silver bullet that solves every problem, but a powerful technology that enables users to do more with less–but keeping the human in the loop is essential.
Mila: Prior to the Google AI Impact Challenge and the Google.org Fellowship, I had a good conceptual idea of artificial intelligence, but did yet not have practical experience of its depth in the context of human rights. We had just started exploring the possible application of AI to optimize information collection and processing for two international organisations, UPR-Info and Plan International. These two organisations are now successfully piloting the AI-powered services, developed in the past seven months, with the support of Google.org Fellows.
Natalie: Since my internship in 2016, I’ve been working as a machine learning specialist at HURIDOCS. What started as an experiment very quickly showed the potential of machine learning to support human rights organizations with their information management challenges.
We explored use cases, developed prototypes for automatically extracting information from documents, and created a vision on how to make machine learning accessible to human rights organizations. Realizing all this requires time, a team with varied skills, user research, lots of iterations and pilot tests and, especially, many conversations with partners as the success of this vision depends on their trust and guidance.
With this in mind we were super excited and happy about being part of the Google AI Impact Challenge and about welcoming Google.org Fellows at HURIDOCS. Our vision was coming one step closer to reality.
Tell us a little bit about what excites you about machine learning for human rights work.
Jaume: Our users could be doing more if they had the resources, but they are constrained by budget limitations which tend to be scarce in the field of human rights. Machine learning frees them from many tasks that can be automated, so they can channel their human intelligence into more productive work.
Mila: This technology has the potential to offer a better user experience and more comprehensive information, and free up scarce resources for human rights advocates who stand at the forefront of advocacy and defense, and ultimately advance human rights for all.
Indeed, a broader leverage of AI capabilities in areas beyond Natural Language Processing (which was the scope of the Fellowship), such as data verification, investigations, image recognition, speech-to-text, in the human rights and social sector, is very exciting. The possibility and the vision to scale this work across the human rights sector is what’s truly exciting for me.
AI is being widely leveraged and is transformative for almost all industries. Its application for social benefit, and more specifically for human rights, is making progress, albeit slowly.
Natalie: Machine learning is very powerful. By alleviating tedious manual tasks, such as browsing through thousands of documents and extracting their title, date, affected persons and corresponding themes, human rights defenders have more resources to focus on their advocacy and litigation work. Machine learning does not only increase the efficiency of the curation process, but also opens up new possibilities for human rights work and collaboration between groups. This empowerment of human rights defenders is inspiring.
What did a typical day look like during this collaboration? What sorts of things did you work on?
Jaume: Most of the efforts went into building good models for Natural Language Processing (NLP) text classification and its integration within our existing document management platform Uwazi. On a regular basis we checked on the performance of the algorithms and discussed different approaches as well as the development of the feature.
Mila:I had two roles: (i) project manager for two pilot human rights organisations that benefited from the Google.org Fellowship on one hand, and (ii) being the HURIDOCS contact person for the Fellows, on the other hand.
I was working to ensure that users’ needs were reflected in the machine learning feature development. We spent a lot of time assessing needs, proposing and testing designs, collecting feedback on the feature usability.
An important part of this project was to understand how the machine learning feature performs and communicate it to the partners in a simple, non-technical language. For example, instead of saying ‘’training data’’, we would say that a part of the data is being used for the algorithm to learn from.
Natalie: What a typical day looked like depends highly on the project state. Before the launch I did daily check-ins with Google.org Fellows Sam and Ben to coordinate and track the development process. Other calls involved the entire product team including product management, developers and the UX/UI team. We specified the features and ensured that they meet the user needs, that they are feasible from a technical perspective, and that their functionality was well communicated to users.
In the meantime I was working with the data from our pilot partners. The goal was to build a machine learning model to assign relevant themes to a short text. To improve the model, I did a comprehensive literature research, ran a lot of experiments and exchanged ideas with Google.org Fellows Sascha and Josh as well as our HURIDOCS machine learning team, which includes Jaume and Gabriel.
When satisfied with the results from the machine learning algorithm, I communicated them to partners and made sure that they understand the underlying concepts, as well as their benefits and limitations.
What was the most surprising or challenging aspect of working with a team from Google?
Jaume: Usually, getting everyone on the same page in terms of how things are done and making sure that we build the right thing both take time; organizations absorb new members one at a time, and it requires taking the time to ramp up both in terms of technology and organizational culture. In this case, we needed to skip some steps since the Fellowship program was limited to six months, and it was a big group of people, so it felt like a clash of cultures where both sides needed to adapt to each other’s reality.
Mila: I was positively surprised by how well and fast the Google team integrated into the project and into the organisation. Any two organizations have different cultures and systems, and it requires time to adjust. In certain aspects (for example, a remote team, smaller organisation, different technical infrastructure and communication systems, etc.), this transition must have been challenging for the Google.org Fellows; however they embraced it with an open mind and enthusiasm.
Natalie: With the Google.org Fellows, our quite small team — especially on the machine learning side — grew significantly. While this brought an incredible level of skill and enthusiasm, it posed some challenges for our internal workflows and processes. The speed of decision-making and technical implementation by the Fellows was incredibly fast, sometimes too fast for me. I thought by slowing down this process we can include all perspectives in a more in-depth discussion and ensure that we are guided by user needs and not previous technical implementation.
A big surprise for me was the positive feedback from Google.org Fellows about the relevance and value of our work. Despite working for one of the most influential tech companies, some pointed out that this project was one of the most meaningful ones they have worked on.
What is the biggest insight that you take away from this collaboration?
Jaume: This sort of organizational cross-pollination is a great opportunity to boost change and innovation and challenge the status quo.
Mila: The Fellowship for me exemplified the vast potential of artificial intelligence when applied in the context of human rights and social work. This is a truly exciting chapter for HURIDOCS and for the human rights community, and I cannot wait to see what the coming years will bring.
Natalie: The past months were full of learning. My biggest insights revolve around the processes of managing a team to make the collaboration as fruitful, efficient and pleasant as possible. These processes include a product requirements document to define the purpose of the feature, progress tracking with objectives and key results, frequent reflections and writing of post-mortems to learn from mistakes and document them, a milestones run-down, and daily check-ins to ensure progress and remove obstacles.
Has the collaboration with Google.org Fellows changed how you will approach your work from now on? If so, how?
Jaume: Better communication and documentation of what is being done.
Mila: On a practical level, having worked closely with the user research and design team, I rediscovered the central role of users in any product, learned concrete methods and techniques to apply in my work.
On a broader level, the Google.org Fellowship for me was an example of a successful cross-sector collaboration, where differences become assets. I am convinced, more than ever, that such partnerships are essential to creating sustainable and robust solutions to issues that we care most about.
Natalie: With the Google.org Fellows and a growing HURIDOCS team, an efficient way of communication became crucial. Documenting meetings, decisions, runned experiments, code, team responsibilities and ready products is key to an effective collaboration.
Since the fellowship, I spend more time on documenting my work and noting down ideas such that they become accessible to team members.
Another important change for me is the adoption of the feedback culture from the Google.org Fellows. While being a bit anxious about feedback before, I learned to appreciate and to utilize it. I now actively ask for feedback and practice giving constructive feedback to others.
Wondering what the experience was like for Google.org Fellows? Check out this question-and-answer post with three of them!