Three Google.org Fellows talk about their experience over the last several months helping HURIDOCS to leverage machine learning for human rights.
The movement for human rights is made more effective through learning, sharing and cooperation. Over the last several months, we’ve put that sentiment into action by collaborating with a group of Google.org Fellows, who have used their skills in software engineering, product management and user experience design to strengthen our work with machine learning to improve access to public human rights information.
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 have helped us to enhance how Uwazi, our open-source platform for creating databases, is used for collections of international laws and rulings related to human rights. Together, we’ve figured out ways that machine learning can automate tasks that were previously burdensome for the people who maintain these databases, resulting in collections that are more up-to-date, consistent and accessible.
Machine learning has long been of interest to us at HURIDOCS, but thanks to the Google.org Fellows we’ve been able to advance this work like never before. What we’ve achieved through this collaboration will not only feed into future development with public collections in mind, but will also indirectly benefit our partners who use Uwazi to document human rights violations.
So, just who are the Googlers who lent us a hand? As their time with HURIDOCS comes to an end, let’s hear from three of them about their experience: software engineers Benjamin Dittes and Samantha Schaevitz, and product manager Grace Danciu. All three are based in Zurich.
Click here to read more about the collaboration, this time from the perspective of our own HURIDOCS team members.
Before you got involved in this fellowship, what were you working on at Google?
Benjamin: I was working on improving Google Maps, specifically the quality of search results. My team is involved in understanding what people search for, how that varies across the world and what types of searches Google is not so good at answering yet, in order to help other teams fix them.
Grace: I was working on Google Assistant, more specifically on the part of the team trying to make Google Assistant more helpful by making it more personal and proactive. I’ve also worked on Gmail, Google Calendar, Google Flights, and several Google products for advertisers, webmasters, and even book publishers.
Samantha: For the past seven years, I’ve been a Site Reliability Engineer working on various Google Apps products including Gmail, Calendar and Meet. My job is very broadly scoped — basically, anything that could help prevent an outage or user-facing bugs from making it into our production environment. A lot of my work as a Technical Lead involves ensuring that the right people get to review changes to the system before they happen, and automating any and all invariants we want to hold true in the system. (Check out the Site Reliability Engineering books my co-workers at Google have written over the past few years to learn more about this.)
Explain what the HURIDOCS mission meant to you.
Benjamin: At Google, there is a general sense that everyone is working on ‘something meaningful’, but your actual work is very often many steps removed from a real user. I was building metrics to help other teams improve search quality to provide better results to users across many countries and devices — so the immediate meaning of my work is sometimes lost.
HURIDOCS, on the other hand, is dealing much more directly with improving the lives of vulnerable people. On a visit to Geneva we got to meet both with client NGOs working directly with the tools I was building and sit in on a UN session about human rights. It was refreshing to see my work have such a direct and “obviously good” impact.
Grace: As a long-time Googler, I deeply believe in Google’s mission to organize the world’s information and make it universally accessible and useful. When I started working with HURIDOCS, I was amazed to discover that there was this whole corner of the information universe that had seemingly been left behind — namely a lot of critical human rights documentation. While it may technically be accessible on the internet, a lot of it is missing that next level of organization that makes it actually useful. It was extremely fulfilling to continue to work on the mission of organizing the world’s information, in an area that hadn’t been receiving enough attention.
Samantha: For a long time, I’ve considered what I want the impact of my career to be on the world. As an engineer at Google, I get a lot of great breadth of impact — my code is executed billions of times per day by 100s of millions of users using products I work on. But in terms of depth of impact, working with HURIDOCS was a clear next level. The software has fewer users, sure, but those users can leverage HURIDOCS’ tools for meaningfully better and more just outcomes for all kinds of at-risk people all over the world. This could mean the difference between prison and freedom for a political prisoner, or a shift in government policy leading to more girls being in school next fall. I found this deeply rewarding.
Describe what a typical day looked like. What sorts of things did you work on?
Benjamin: My main role as a software engineer is improving the software platform, called Uwazi, that HURIDOCS builds for its client NGOs. So my day looked very much like any other software engineer’s day: writing Python and TypeScript code (and tests), coordinating product and code design with the rest of the team, trying out prototypes, etc. The two most interesting and novel things I got to do were integrating a TensorFlow machine learning learning model into the Uwazi production setup and experiencing the full software development stack outside of Google (which mostly uses internal, separate tools).
Grace: I was primarily working on defining the specific feature we were developing, which was about using machine learning to automatically categorize documents. Getting the algorithm right was critical of course, but there was also a lot of work to be done in terms of understanding the underlying need, talking to our users in-depth about the benefits and limitations of machine learning, and refining designs based on user input. “Machine learning” gets a lot of hype these days, but for many people it’s still a mysterious and daunting subject that’s either going to magically solve all their problems or lead to some inexplicable bad outcomes. We spent a lot of time explaining the basics of machine learning to our users in an accessible way, to remove the aura of mystery and to help them understand their role in the feature’s success as well.
Samantha: I spent a lot of my time in the first half of the fellowship getting to know people, attempting to understand their roles in the organization, and learning new technologies to develop Uwazi effectively. In addition to more standard programming tasks — reading code, writing code, testing code, shipping code — I also spent time trying to ensure that my understanding of The Plan was in-line with what all of our other stakeholders (e.g. Product, Project Management, UX, Users) wanted. I quite enjoyed how much time I could dedicate to learning new technologies, getting to know new people and to understanding new problem spaces throughout this fellowship!
What was the most surprising or challenging aspect of working with HURIDOCS?
Benjamin: The HURIDOCS team is spread all over the world, working mostly from home and coordinating via video calls, group messages, etc. I had never worked in a mostly-remote environment, and it took me a while to get used to it. In hindsight, it was a great way to prepare for Covid-19, though…
Grace: Adjusting to tools that HURIDOCS uses, particularly the fact that it’s much more of a chat-based organization than one that depends primarily on email like Google.
Samantha: I was both surprised at how many of the day-to-day people and organizational problems were familiar to me (as someone who went straight from university to Google), and also surprised at which specific problems were handled vastly differently.
For example, I knew in the abstract that HURIDOCS would operate differently in part due to its existence as an NGO with 20 employees than Google with its 100K+. But identifying that scaling from 15 to 20 employees is equally challenging for an organization as is scaling from 25K to 50K was a surprise. So many models of interaction and work norms must adapt. The through-line I find relevant here is the claim that “what got you here won’t get you there.”
I was also pleasantly surprised at how familiar a lot of the engineering practices were to me, even though the specific software technologies were different from those with which I was familiar. I came away with more confidence that my skills are transferable, even after almost eight years at one company!
What is your biggest takeaway from this fellowship experience?
Benjamin: The good: I was surprised how well the open source code development matched my workflow with Google internal tools. I felt productive very quickly and I can imagine contributing to open source projects more in the future because of that.
The bad: What I didn’t expect was how long it would take to start the actual project implementation. It took almost a month to settle on an initial design and user interface mock, with quite frequent changes based on our increasing understanding of what kinds of changes were easier or harder to make in the existing codebase and what the machine learning models were able to achieve.
The ugly: I learned that I personally do not enjoy building client-facing web applications. I’d never done it before, and I’m proud of some of the new flows we did build, but I didn’t enjoy the process.
Grace: That working on something I find deeply meaningful isn’t nice-to-have; it’s something that I don’t think I can do without after having gotten a taste of it.
Samantha: The Fellowship really drove home for me something I already suspected — that a small group of dedicated people are not to be underestimated in what they can achieve. The hard part for me has been figuring out what to dedicate more of my time to (there are so many problems worth solving in the world), but the fellowship let me focus on one and to see that that was enough to create real, positive social impact.
Has this fellowship experience changed how you will approach your work in Google from now on? If so, how?
Benjamin: I honestly don’t know yet, it will take time to see which skills or experiences fit into the work coming up in my Google project. I think I will seek out more immediate user-facing impact, which is the part of the work at HURIDOCS I enjoyed most.
Grace: Continuing from the above point, it’s definitely made me think a lot about where I want to turn my career next, starting with exploring more socially-oriented teams within Google.
Samantha: My time focusing on developing a feature from start to finish has helped me carve out the role I want to have once I get back to my regular team. It helped solidify for me that, while larger-scoped roles can mean that there are more interesting problems to work on, without clear methods of prioritization that span the whole area, it can also become very challenging to convince oneself that one truly works on the right problems. And so, once back, I plan to continue to focus on those problems which I can prioritize working on over others, and to keep a focus in my work wherever possible.
Click here to read more about the collaboration, this time from the perspective of our own HURIDOCS team members.