ETH students develop tools for the UN

Voting data made transparent 

¡°With our prototype, we want to make it easier to see how UN members have voted on various issues,¡± says team lead Benjamin Hoffman, a doctoral student at ETH¡¯s Computer Engineering and Networks Laboratory. Currently, voting results from the UN¡¯s 75-year history are scattered across numerous websites, and anyone wishing to see the bigger picture must first patch together the data from myriad sources. Hoffman and Moser are hoping to change that. With the help of their online platform, diplomats, researchers, journalists and the public will be able to view and compare the voting behaviour of all UN member states in just a few clicks. ¡°For example, our interactive maps can show how two countries ¨C say, Switzerland and China ¨C have drifted closer to?gether or further apart on specific issues over the years,¡± says Moser. 

A huge amount of coding work went into creating these maps. ¡°One unexpected challenge was dealing with countries that have changed their name ¨C in some cases, multiple times ¨C since the UN was founded in 1945,¡± says Hoffman. Since last summer, the team has been in constant contact with UN library staff to resolve such issues. ¡°It was crucial for us to understand the UN¡¯s actual needs through regular feedback from within the organisation,¡± says Hoffman. Through these exchanges, the students have gained valuable insights into international politics while also acquiring new skills.  

Facing budget cuts, the UN Digital Library lacks the resources to tackle such a project alone. ¡°Cooperation is essential for the UN, especially in the area of technology. We can achieve much more by partnering with non-state actors such as ETH,¡± says Guy Ryder, Under-Secretary-General for Policy and a leading figure at the United Nations.  

The students¡¯ work exemplifies the kind of projects that fall under the banner of the ETH¨CUN partnership. In addition to Hoffman and Moser¡¯s team, the partnership envisages other groups of students working on concrete solutions for various UN bodies. Plans include a project with the Food and Agriculture Organization (FAO) to model harvest risks faced by smallholder farmers and a collaboration with the Office of the High Commissioner for Human Rights (OHCHR) to create a new online tool that will allow users to search and analyse over 230,000 observations and recommendations on human rights issues.  

Pictograms for UNICEF 

Many people struggle to understand written information, whether in a letter from a teacher, a doctor¡¯s report or a government leaflet. The reasons are complex and varied, ranging from a lack of literacy and intellectual impairment to certain forms of neurodiversity and both children and adults are affected. 

¡°Access to simple, easy-to-understand information is crucial; we regard this as a human right,¡± says Thy Nowak-Tran from the United Nations Children¡¯s Fund (UNICEF). To ensure that written language in documents, instructions, presentations, package inserts and on the Internet is as clear and accessible as possible, UNICEF prioritises the use of simplified language along with a clear layout and supporting images ¨C an approach experts refer to as Easy Read. ¡°We see Easy Read as a tool for increased accessibility and inclusion,¡± says Nowak-Tran. 

Pictograms are central to Easy Read, but custom graphics are expensive and slow to produce. This is where another ETH¨CUN project comes in: ¡°AI image generation models could be a solution here, but the standard commercial models don¡¯t generate images that are suitable for Easy Read,¡± says Sonia Laguna, a doctoral student in the research group led by Julia Vogt, an ETH professor specialising in medical data science. Current commercial models tend to generate images that are overly elaborate and detailed. Yet Easy Read pictograms and symbols should be simple, stark and as unambiguous as possible. And they need to work across cultures to enable deployment in different countries and contexts. Existing AI models often struggle to achieve this combination. 

Working with a small ETH team and collaborating closely with UNICEF, Laguna adapted a state-of-the-art open-source image-generation model. The goal was to offer a simple way for users to auto?matically generate an easy-to-understand pictogram in response to a prompt such as ¡°the school is on fire¡±. The ETH computer scientist trained the model on curated pictogram datasets and tuned its parameters to generate Easy Read style outputs. The team also made sure that the pictograms accurately reflect cultural and regional diversity, such as skin tone. The first prototype already shows promise. ¡°The resulting images are significantly closer to the Easy Read style than those from commercial models,¡± Laguna says. 

The biggest challenge for the ETH researcher and her team came when evaluating the generated images, as no objective standards exist for Easy Read symbols. The team therefore introduced their own Easy Read score, and plans are now underway for user studies with UNICEF experts, as well as pilot projects in various countries. 

¡°This project is already showing us how AI tools can expand our symbol databases, especially in multilingual contexts where resources are scarce,¡± says Nowak-Tran. In the long term, UNICEF hopes this collaboration with ETH will greatly simplify the production of Easy Read pictograms, making them a standard option rather than a luxury. Since the AI model will be open access, it will also give organisations outside the UN an opportunity to make information more accessible for people with cognitive impairments. In a further step, the ETH and UNICEF experts plan to investigate whether AI models are also suitable for producing texts and layouts in an Easy Read format. 

Flood protection for refugees 

Cooperation between the UN and ETH predates the 2023 partnership, and ETH researchers have applied their expertise to assist the United Nations on multiple occasions in the past. One example is Bruna Rohling, a doctoral student under David Kaufmann, Professor of Spatial Development and Urban Policy at ETH Zurich. From 2021 to 2024, Rohling worked with the Geneva Technical Hub, a collaboration between ETH, the Swiss Agency for Development and Cooperation (SDC) and the UN Refugee Agency (UNHCR) to enhance flood risk mitigation in refugee settlements. Due to their location and the often fragile nature of the shelters, these settlements are highly vulner?able to flooding when heavy rain suddenly inundates the surrounding area. 

Rohling and her colleagues developed a flood risk mitigation toolkit for the UNHCR, featuring a mapping tool that combines global and local geospatial data to identify flood hazard zones and vulnerable buildings in and around refugee settlements. This tool also includes a catalogue of 22?flood risk mitigation measures for refugee settlements, plus a guide for UNHCR staff to assess risks on site together with local actors and residents. 

Identifying the hazard zones 

To pilot the toolbox, the UNHCR Country Office of the Republic of the Congo invited the ETH researchers to the ¡°Site du 15 avril¡± refugee settlement in the town of B¨¦tou, in the north of the country, in 2024. Home to 6,000 refugees, the settlement suffered severe flooding in 2023, with some homes, sanitation facilities, streets and agricultural fields under water for weeks. ¡°Our goal was to identify and map the flood hazard zones in B¨¦tou and its refugee settlement together with local actors and those residents directly affected,¡± recalls Rohling. 

After analysing existing geospatial data, the team conducted numerous site visits and interviews, and held a mapping workshop with residents from B¨¦tou and the refugee site. Based on these insights, the local UNHCR team and ETH researchers drew up a flood risk mitigation strategy for B¨¦tou with a series of short- and long-term measures ranging from improved drainage systems and elevated construction to reforestation and the implementation of an early-warning system. ¡°Flood risk mitigation should be addressed holistically and across spatio-temporal scales. In B¨¦tou, this means that urban planning and flood-resilient architecture are just as important as prioritising local know?ledge and intact ecosystems,¡± says Rohling. 

The ETH team then presented their findings at the UNHCR Country Office and to national policymakers in Brazzaville, capital of the Republic of the Congo. The toolbox is also scheduled to be deployed in other UNHCR refugee camps. Despite being launched before the official inauguration of the ETH¨CUN partnership, this project also perfectly em?bodies its spirit.