Interview with AI for Good Simulator team, the winners of the HackFromHome COVID-19 hackathon

A couple of weeks ago a team of enthusiasts from the AI for Good community participated in HackFromHome, the 48-hour global hackathon and won a funding grant from Just One Giant Lab for building a COVID-19 simulator for refugee camps. 

Refugee camps usually have a very high population density which makes them vulnerable to the spread of the pandemic. Within these camps, vulnerable groups extend beyond the old population and there are people with pre-existing conditions like TB, HIV, and malnutrition which exposes them at a higher risk to the virus.

Due to the lack of quarantine facilities in these camps, the virus left unchecked can spread quickly since young people and old people have regular interactions. To make matters worse, there are a relatively small number of medical staff working in the camps, and the lack of medical supplies means that medical staff are at a greater risk of getting infected.

Fortunately, the team at AIforGood Simulator believes that the latest epidemiology modeling approaches can help for the following reasons.

Since the camps are currently in shutdown, there is no influx or out-flux of people so the geographical boundaries are contained. People in the camp follow certain routines that can be predicted, for example, each section of the camp has a food distribution point where people usually spend hours together waiting in line.

These constants mean that an agent-based simulation in such an environment is going to be representative of the real situation and modeling can inform the effectiveness of the different actions the camp authorities can take. An example is swapping houses to gather vulnerable people together in one quarter and setting up a single entrance to that quarter where a hand-washing facility is used. 

The project aims to inform NGOs and local communities on the effectiveness of measures feasible within a camp setting. Through this process, AI for Good has had the pleasure of expert advice from Dr Shikta Das who is an epidemiologist and from NGO workers providing support in Moria Camp. Here’s a link to their solution: https://www.aiforgoodsimulator.com/ 

Now, let’s hear from the winners themselves at AI for Good about how much thought, effort, and passion was put into this project and what we can learn from it!

What was your contribution to the project? 

Billy Zhao, Team Lead: “Interfacing with the HackfromHome organisers to make sure we are on the right track meanwhile created a 3blue1brown style video to highlight how high population density can cause rapid spread of coronavirus.”

Alice Piterova, AI for Good Community Lead and Problem Owner: “Bringing together people from the AI for Good Community from epidemiologists, NGO workers, developers, Data Scientists, and Researchers. I’ve also collected the data about a couple of refugee camps for the modeling exercise and created the website.”

Dr. Shikta Das, Epidemiologist: “I advised the AI for Good team as an epidemiologist. I helped to conceptualise the issues faced in Refugee camps that are relevant for modeling infectious diseases. I gathered the scientific evidence supporting these issues and the problems associated with refugee camps as well as the proposed interventions suggested. I provided the methodological insights for COVID-19 modeling and the parameters required for these models.”

Harry Stamatoukos, Compartment Modeling: “The compartment SEIR model in Python and building an interactive User Interface to communicate the model to the public.”

Shyaam Ramkumar, Agent-Based Modeling: “The agent-based Netlogo model of the Moria refugee camp. I worked with the team to look at the information they gathered to try and create a close approximation of the camp in a virtual simulated environment. I developed a scenario of how a virus-like COVID-19 would spread throughout the camp, and added various interventions to simulate how that would stem the spread of the virus.”

Aarushi Kapur, Presentation and Research: “Researching basic epidemiological models and the interventions most relevant for COVID-19 in refugee camps and preparing the presentation of our work.”

Kieron Scully, Research: “Reviewing research to summarise some of the prior art in terms of best practice and what some previous work had covered around modeling disease in refugee camps. I made a cursory search for some potential sources of data for validation of our models in the future.”

Mitchell Hensman, Data Visualization: “Creating visualisations to convey key insights from our models as well as interactive dashboards enabling custom exploration of the sensitivity of interventions of various magnitudes at different points in time to transmission outcomes.”

Abdul Fattah, Front End: “Helped with technical frontend issues when the AI for Good team needed to work directly with JS/HTML in the visual website builder. Luckily, the modeling software had a .html output which I ended up hosting statically using Netlify when embedding in iFrame proved ugly.”

How did you make it happen within 48-hours – your superpowers? 

Billy: “Kept my eyes on the goal and I felt like we were doing something really important (also pg tips).”

Alice: “The superpowers I used for this project were years of experience in engaging with a wide range of stakeholders, aligning their vision, and encouraging them to work together to address the actual needs of the most vulnerable.”

Shikta: “I stayed in contact with the team during the hack to support them with any data they might need in terms of modeling and fact-checking if required. I tested the virtual simulated environment that was created and approved the scenario for COVID-19 spread and added various interventions to reduce this spread.”

Harry: “No superpowers really, like superman’s flying ability, it’s more like batman’s utility belt. The question in my head was, what kind of tools will allow me to do this as fast as possible so that I will spend more time focusing on understanding the problem and less time on implementing the actual code. So I went for a combination of Python and Streamlit to quickly set up a UI and have a visual representation of the SEIR model while all the code was written in python.”

Shyaam: “I had built an earlier virus Netlogo model which I used as a basis for this one. That last model was my first time so it took me a week to make. And since much of the heavy lifting was already done in that model, it allowed me to reuse some of the same code for the hackathon.”

Aarushi: “Leveraged the research skills learned at university as I haven’t needed to read so many academic papers in such a short time in a while!”

Kieron: “Everyone was very supportive and driven. When I came on I was given a paper to summarise and was able to explore other avenues when we were reaching a minor impasse. I found most of the complicated stuff had already been done really and had some good leads for where to explore next.”

Mitchell: “Initially, it was important to develop an understanding of some of the concepts and how the models work and what the data should look like followed by an exploration of the impact of changing parameters. To demonstrate the concept of using the model outputs to visualise in Tableau I followed a laborious method of manual extraction of sample data from a web app into a spreadsheet, using Dataiku DSS for data preparation and Tableau for reproducing common ways of visualising this data. Working with other members of the AI for Good team I was able to obtain flat file extracts from our models which enabled me to begin flexibly exploring patterns in the data and looking for new ways to communicate key insights.”

Abdul: “I was lucky the modeling team had an .html output file ready and I did not need to build any visualization for the data.”

What did you like the most about this experience?

Billy: “I liked that I was working with a group of passionate people who care about the same cause and we were supporting each other throughout the way.”

Alice: “At first, I thought our idea was too complex and an average newly emerged team won’t be able to pull it through in 48 hours, they will need at least a week or two. But we managed to put together extremely passionate and hardworking individuals that worked together like an orchestra and exceeded their own expectations in achieving the impossible. When we submitted the project at 00:00 on Sunday I was so happy with what we had built that I told my partner: “I think we have just created an Open AI level simulator to save real lives.” I’m a big believer in the impact of modeling methods in addressing uncertainties in decision-making and helping those that need it the most. We need to democratise these methods and make it accessible to humanitarians and NGOs. 

Shikta:  “The hackathon has been a very new concept for me. Being an academic, I found it an extremely rewarding experience as I got an opportunity to use my skills for a good cause. I was able to work with a really diverse team where each one of us brought a different skill-set and a range of ideas from different disciplines. It was a pleasure working with young minds and their interesting questions. The AI for Good team we had was very dedicated and were asking me the right questions. I was on stand-by during my hackathon time. There were occasions when I could not join them due to family commitments. I am glad the team trusted my research and made full use of the material I sent them. I’m really impressed that they took an opportunity to compare different models for the pandemic in this vulnerable population.”

Harry: “Working alongside the amazingly brilliant individuals at AI for Good.”

Shyaam: “I really enjoyed working with a team of passionate and dedicated people at AI for Good. It was really helpful to use concrete research about the situation in the refugee camps that the other team members gathered, and it made the model more realistic. And it was great to have others working on different models to compare the pros and cons of different approaches to understanding scenarios of a pandemic in a critical area like a refugee camp.”

Aarushi: “Working with a team of incredibly talented people was definitely the best part; working towards a common social impact goal with people who are ambitious, while also feeling like you are adding value to the process, is an extremely enjoyable and rewarding experience.”

Kieron: “It was amazing seeing how everyone at the AI for Good team worked and came together! Having a weekend deadline always helps to make progress too, it was fun getting stuff done, and it was a great feeling even when helping in small ways like resurrecting a lost slide and getting to play with the models. The work also has a personal element for me, my grandparents came to Britain as refugees from the Balkans, and I think that connection helps bring home how important work like this is.”

Mitchell: “My experience was imbued with the appreciation for finally getting involved with the AI for Good community. This has been something I’ve been intending on doing for a while as the theme is something I’m very hopeful about and have been promoting through notable efforts to evolve Dataiku’s social good programme.”

Abdul: “It was a diverse team who worked together so well, I wondered how well we’ve known each other.”

What was particularly tough and how can we learn from this experience going forward?

Billy: “It was hard to communicate over the video sometimes when you are trying to explain something across but we got there eventually.”

Alice: “It was heartbreaking for me to talk to the people working closely with refugee camps that I knew well from my previous role at Techfugees and could trust to be objective. They convinced me that if COVID-19 hits one of the camps it may lead to a disaster that neither camp authorities nor NGOs would be able to stop. I’m still hoping for the best and trying to stay optimistic but it is hard when every day feels like our last chance to do something.”

Shikta: “I personally found that the virtual team works really well for me. I just wish it was during the week instead of the weekend. Being a mother in this COVID-19 environment gives me no time during the weekend to participate in these activities. But the team worked around my restrictions and were very supportive of it! I am super proud of them and was confident of their win.”

Harry: “Working collaboratively was a bit tricky as most of us were working on different things and it was difficult to keep track of what everyone is doing  – but we managed to pull everything together in the end :)”.

Shyaam: “The virtual nature of the hackathon was a bit tough. I have participated in live hackathons before, but this was my first virtual one. I tried to keep up with all the content and messages and threads, but it was a little difficult to do. And I found myself losing track of time working on the model, so I missed a few of the scheduled team check-ins. But it was a lot of fun to do and I would certainly do it again!”

Aarushi: “This was my first hackathon so navigating how to be most useful was initially challenging. Guidance from the rest of the team helped, but I learnt that taking initiative and working on items you think are most valuable to reach the team’s goal (based on your skills set) is ultimately what it’s all about.”

Kieron: “I Had some trouble working out how to join the hackathon, so I started a little late, but I think it worked out in the end!”

Mitchell: “In a typical data project, the visualisation and communication of the data (or findings) are emphasised at the tail end. A lot of time can be expended on obtaining, understanding, and cleaning the data which can compromise the time spent effectively visualising it (which is often the only part the audience will engage with). Some time pressure at the end resulted in a more mechanical approach to the preferably creative process of creating effective visualisations. I think starting from a blank canvas with more ‘mental space’ could be more enjoyable and productive. I will also aim to choose a work destination 6 hours behind in time for the next visualisation project.”

Abdul: “I’d say time-zones. Each team member could add their time zones to their display name so it’s easy to sync our times/effort with the rest of the AI for Good team.”

What’s next?

Alice: “We are trying to conduct user research with NGOs to make our tool more useful and usable to forecast the demand for supplies and services. 

“We are trying to reach global humanitarian organisations and camp authorities to give them our AI for Good tool to help evaluate the effectiveness and appropriate time for the interventions. 

“We are trying to use different mathematical models (agent-based, compartment and interaction-based) to simulate a refugee camp environment that will represent the real situation as accurately as possible. 

“We are trying to use satellite imagery to map the layout of camps to quickly scale and replicate our models across multiple regions of the world. 

“We are looking for ways to increase computational power for our models to run simulations with more agents (people) and parameters (camp settings). 

“And finally, we are always looking for ways to conduct better research into simulation approaches and parameters to make our models more robust, accurate, and effective.”

What support do AI for Good need from volunteers?

Alice: “We need specialists in agent-based modeling (NetLogo, MASON), network modeling (R, Python), and compartment modeling (Python).

“We need web app developers in React (or similar in JavaScript) or Django (or similar framework in Python). 

“We need front-end specialists with model visualization skills (Vega-Lite or similar).

“We need UX researchers to improve the design of our tool and make it intuitive to use.  

“We need a DevOps engineer to set up cloud / virtual machines to get more computational power. 

“We are looking for collaborations with medical research groups to validate our model assumptions.”

What support do AI for Good need from partners?

Alice: “We are looking for partners in NGOs and local grassroots organisations who work with the displaced people. 

“We are looking for contacts of relevant decision-makers on a country level and in global humanitarian organisations.

“We are looking for collaborations with humanitarian and medical research groups to validate our model assumptions.”

If you have any of the skills above or are a business within the NGO and local grassroots arenas, we encourage you to get in touch with the team to get involved in this amazing initiative.

HackFromHome is a virtual hackathon sponsored by Dataswift.io and organised by the Ethical Tech Alliance, HAT-LAB,  Case Western Reserve University’s xLab, Cleveland Clinic’s Hwang Lab, and WMG – University of Warwick.