Project Jetson is a platform aimed to provide UNHCR operations predictions about population movement (arrivals/departures) for specific regions or countries. Jetson – a machine-learning based application – measures multiple variables to see how changes over time that affect movement of UNHCR’s persons of concern, particularly refugees and internally displaced people. This experimental project was launched by UNHCR’s Innovation Service in 2017 to better understand how data can be used to predict movements of people in Sub-Saharan Africa, particularly in the Horn of Africa.
We asked Sofia Kyriazi, Artificial Intelligence (AI) Engineer, and Babusi Nyoni, User Experience (UX) Designer, from the project team to discuss the challenges of predicting displacement and what success may (or may not) look like for the project as it moves forward.
Why is user experience so important to this project?
BN: Project Jetson, a project about the future of displacement, is one that must be articulated carefully. This is because user experience is a process to enhance user satisfaction with a certain product or service – in this case, a website – and how it is perceived by users. It involves taking into consideration human-computer interaction. On the one hand, as a team broaching a future-facing project, a visual representation befitting of the magnitude of the work would seem appropriate. A willful disregard for user experience convention would be permitted to a degree considering the rationale that an idea of the future deserves a matching facade. On the other hand, the future, because it is unknown, is intimidating, and sometimes scary. A misstep in proper articulation could nullify the purpose of the website when users of the site failed to comprehend the content and how to interact with it.
What were some of the challenges you faced during the process of developing Jetson?
BN: Mapping the user experience had its fair share of challenges. First of all being the responsibility of presenting an interface that both the general public and UNHCR staff will find intuitive. This presented interesting challenges in creating a unified visual experience while still catering for very specific use cases such as UNHCR staff on the ground in Somalia and their more office-oriented colleagues in Geneva considering that the bandwidth disparity between the two is extensive. Another challenge was that the map visual morphed over time from something static to a very dynamic representation of conflict and displacement data over time. The use of multiple references contributed to the near-convolution of this process, but this was expected as the main consideration was that the solution would have to be bespoke and tailored to the scenario. As we continue to define the story, the visual is expected to change to something more befitting.
SK: The first and most important challenge was the question: what are we trying to show? Which number is important to us. Implementing all different based mathematical functions, to approach the actual arrivals, a lot of times the results were completely off, even though we were doing the training with all the available data. This could either mean that the data is not correlated and that we needed to expand with more information or that we were working in the wrong way.
BN: Additionally, presenting information in the most succinct manner was challenging in that while the website was meant to house the predictive engine, the map visual, and long-form content, considerations had to be made as to how much information and how to display it. A user-friendly summarisation of the engine was conceived that gave casual users of the website a brief view of the engine metrics and results with the option to view the parameters at depth.
SK: Another challenge for me was the uncertainty over correlations between datasets, of various formats, that we were collecting and how they could assist in predicting arrivals in each region. The datasets had to be cleaned, transformed and grouped per month, with the use of scripts in python or/and R, an action required to minimise the input to the modelling engine. The scope of the project had to be limited in modelling arrivals in the region of Bay. These were challenges regarding the data volume, thus to apply our effort and focus researching one use case, the pilot case, and documenting the process, in order to systematise it for the rest of the cases. Eureqa, an A.I. powered modelling engine, as a tool lacks in examples of time series predictions, future predictions. Only through forums, we were able to find a way to modify the research function to be able to predict arrivals for a month in advance. The produced functions were implemented, with the use of R, commonly used for statistical analysis, and the predictions were collected to be compared with the official real numbers of arrivals. The “winner” function was used in the final application, developed under the Shiny library, hosted in the shinyapp.io platform.
What has worked so far?
BN: An essential part of the process was the weekly standup/check-in meetings that helped track progress and kept this mapped to the project goals and deliverables.The mid-process workshop held in Geneva with all members of the team physically present fast-tracked progress on the resolution of a number of pain points. It also assisted in the rapid iteration of recommendations to the current state of the respective deliverables.The ability to tap into UNHCR Innovation’s domain expertise in big data and on-the-field information came in handy when framing the solutions and validating outputs and having the collaborators on-site meant our efforts could very easily be contextualised for UNHCR’s needs, which is something the team appreciated.
SK: It was a matter of time for the team to gain the same speed on dealing with requests, and we overcame the barrier of depending on completion of each other’s tasks, to proceed with our own. We managed to automate the process of collecting and transforming data to assist future predictions, this part is now done in a short time and with ease. We have created a systematized process we follow, to expand on other regions of Somalia, in terms of collecting results from the tools we use, implementing, testing and iterating to come up the best estimation of movements.
What will success look like for the team and for the product?
BN: Success from a user experience perspective is an intuitive interface. It is one that tells a story that can be understood without supervision and that users can articulate to non-users of the website accurately. This includes powerful imagery where necessary; concise representations of data; and interactions that convey trust all aide in creating an experience that is the best representation of a platform’s intent. For the team, success lies in presenting deliverables that articulate, with cohesion, the team’s mandate with regards to the task. This begins with defining and adhering to a team dynamic that works and also, at the same time, allowing for a level of fluidity from team members in executing their responsibilities respectively. And finally, the product should be trustable enough to use without any degree of failure, either by the product (in doing what it is expected to) or by the user (in achieving their goals).
SK: It would be a huge achievement to have the image of arrivals and departures for each region, meaning if someone would want to see, what is going to happen in Somalia over the course of the next month, they could be able, ideally in a more visual way (not just numbers) to see where big movements will take place. It would be even better if we got an “out of the ordinary” prediction, such as an alert of an unconventional movement. This would indicate that the engine has been trained enough to predict abnormalities. Regarding team success, over the last couple of months, we took time to make mistakes and sometimes we used time, expecting results from each other. Our over the weekend workshop had some amazing results and it gave the team a new pace, faster and more confident. To try to define it more, I would want to see everyone expressing their creativity and passion while being on the same track.
We’re always looking for great stories, ideas, and opinions on innovations that are led by or create impact for refugees. If you have one to share with us send us an email at [email protected]
If you’d like to repost this article on your website, please see our reposting policy.
Icon: Created by Moons for Noun Project