In conversation with Solidarités International: Artificial Intelligence in a humanitarian context

Since September 2020, Solidarités International has been a part of our GSMA Mobile for Humanitarian (M4H) Innovation Fund portfolio. The Fund aims to promote innovation in the use of mobile technology to address humanitarian challenges. Solidarités International, a French NGO, assists populations affected by conflicts and violence, epidemics, natural or climate-related disasters and economic collapse. They aim to meet vital needs and are mainly active in providing Water, Sanitation and Hygiene (WASH) services and providing shelter for refugees. In Lebanon, they are focussing on the Syrian refugee crisis by providing assistance to around 100,000 refugees living in the country across more than 3,000 sites. Assistance provided includes the provision of safe water and sanitation to people, keeping individuals and communities informed about services they are eligible for, conducting awareness campaigns and collecting feedback on service delivery. Providing this assistance in a widespread area has its challenges and in an attempt to address these, Solidarités International have previously made use of a standard dial-in hotline to deal with requests for services. Whilst this hotline increased effectiveness of service delivery, it has still had limitations for the users in terms of affordability and line availability.

Utilising an M4H Innovation Fund grant, awarded in 2019, Solidarités International have developed the ‘SOLIS bot’, a chatbot that works via WhatsApp, which acts as a two-way communication channel between refugees and Solidarités International.

The use of Artificial Intelligence (AI), such as chatbots, in the humanitarian space remains relatively new and so I sat down with Francois-Xavier Dosogne, Information Management Coordinator at Solidarités International, to discuss the SOLIS project in more detail and understand how the team manages ethical considerations and mitigates any potential risks associated with this technology.

Can you explain the origins of the SOLIS bot project – why have Solidarités International chosen to develop this?

Among a population where more than 80 per cent of individuals are regularly using WhatsApp to keep in touch with their family or seek employment, the introduction of WhatsApp to bridge Solidarités International and the people it serves appeared as an opportunity too good to miss. Available 24/7 and free of charge, the SOLIS bot can handle a massive volume of information in both ways. Connected to our internal case management software, SOLIS, it stores and retrieves data at a glance. It is a new method to act as a dedicated communication channel for any user, enabling access to specific personal files or receiving notifications and updates on their service request. For Solidarités International it is also a game changer, allowing our team to provide more services and attention without overloading the capacity of our team. It opens many possibilities for various populations in need of humanitarian assistance across the world, a challenge Solidarités International is seeking to address.

What has been the impact of the bot so far?

The bot has reached a total of 18,760 requests for information or assistance since its launch in October 2020. It exchanged more than 285,000 messages with users, supporting more than 9,000 households.

Through the broadcast feature, a tool to rapidly send a message to a mass number of people, we have been able to launch several communication campaigns and were able to reach our complete list of beneficiaries (13,000 households in February 2021) in a 24 hour period to alert them of fraudulent activities occurring in their area.

We also launched a ‘form’ module through the bot that we are mostly using for our feedback and complaints sector. The form allows any user in the area to submit for free and 24/7, several types of feedback or complaint forms that we have adapted to a chat-bot format. Through a standard chat conversation via WhatsApp, the bot asks questions and completes the form with the user’s answers. It is very useful for aid service requests that we can then follow up in a fast way. This new feature has been a great success with more than 3,000 forms being submitted since its introduction in mid-July 2021.

To assist our COVID-19 field response, Solidarités International used the same survey module to screen all households on post distribution satisfaction of COVID-19 safety kits. We sent users a request to participate in our survey, then when users had accepted the request, the bot asked five questions to enquire if the kits they received were useful or not. We were really impressed to reach 32 per cent engagement, more than 2,900 forms completed and submitted, in only two days.

Following every communication campaign, we also witness a high peak in the use of the information centre part of the bot, where users can find useful information such as vaccine eligibility criteria or local contact information.

The SOLIS bot is bridging the communication gap between ourselves and our intended audience; we can already see ease the benefit that it is having amongst the refugee populations we are working with.

Are there concerns that a chatbot offers a less personalised experience for users? Can chatbots lead to a disconnect between user and organisation?

Any digital project in the humanitarian sector must prioritise human involvement in aid action. Bots and smartphones will never replace the need for human interaction.

But in a limited resource setup, we need to make sure that available staff are prioritising refugees with the most pressing needs. By automating responses to general information requests, we can refocus workload were human interaction is most needed by providing prioritised assistance to the most vulnerable. Combined, our team and the SOLIS bot are then working together to improve and facilitate efficient connection between refugees and Solidarités International.

There is a concern amongst researchers that some human biases can be introduced into chatbots through the design of algorithms and the assumptions by coders of the needs of the users, which can impact the messages and information that is sent. Is there a way that the SOLIS bot is designed to acknowledge or avoid the potential biases in the bot’s algorithm?

The Natural Language Understanding (NLU) module is the process through which the bot can interpret open phrases written or spoken by humans to identify the actual intention behind the request. For example: “Can you send me the phone number of the Red Cross please?”, should lead the bot to understand that what you want is the contact information. This process uses algorithms to match any phrase with a single keyword that is linked to a specific message the bot will send as reply.

Bias may be introduced due to several reasons but they will mainly come from cultural differences and misunderstanding. This is why our ‘building with the local communities’ approach is so important. The bot is equipped with a real Machine Learning module, meaning that it can learn and improve its accuracy through its experience with human interaction, using their feedback (“Was this content useful?, Yes/No” for example), as a compass.

This helps the SOLIS bot store and understand an array of phrases and terminologies, meaning it has the ability to understand and respond to hundreds of ways which requests for the Red Cross contact details may be shared, in several dialect or languages. To support the bot and to try and guard against any potential biases, we’ve built a manual management interface where our community manager can help the bot with this learning task by manually correcting mismatch or unfound matches.

Through our Machine Learning module, with the bot’s development and ability to respond being supported and enabled by thousands of users, this is how we are looking to avoid algorithm biases.

With increasing numbers of users able to register complaints through the bot, how are Solidarités International making sure they are able to respond and follow up to these complaints in a timely manner?

Through a specific API, we are able to automatically encode the requests we receive and forward these requests through to the relevant field activity manager. This enables us to automate a major part of the request process, allowing us to provide targeted, relevant and rapid assistance, saving time on having to perform administrative tasks and manually assigning requests to the relevant agent. This gives us a greater chance of responding to all of our requests in a timely manner, maintaining the level of trust that we have built up amongst our user base. To date, we have reduced the average response time for information requests from four hours and 30 minutes (prior to the SOLIS bot) to three minutes and 20 seconds.

How do you navigate the issue of cultural and/or language sensitivities?

When we originally designed the SOLIS bot, it was designed to cater specifically for the Syrian refugee population in Lebanon. One of the key pillars of our approach is to make the bot relevant to the targeted end users. This is why our very first activity was to take the needed time to design the content with the target communities through focus group discussions. We prepared it all in Arabic, making sure the wording and concepts would make sense in the Syrian dialect.

We visited many sites, checking at first what information users needed and what content they consumed in their daily lives, making sure that the language used on the bot was understood by the end users through iterative tests and feedbacks. To improve the bot’s performance, we recruited a Community Manager whose role was to consistently improve the language and user friendliness of the bot by including several adjustments throughout the launch phase and the project. Part of that role was to introduce a number of different features on the bot, including Emojis, as well as voice notes registered in a dialect for people that cannot read or write. A user centric design has been crucial for the success of the SOLIS Bot.

How do you overcome the data safeguarding concerns that are often associated with chatbots?

Solidarités International has a high level of data protection and security safeguarding. Protection by design and by default are the basis of our digital projects.

By default, all SOLIS bot users are granted a high level of data protection and we make sure to guarantee these rights while using data management tools. This is why when designing the forms and the content, we pay particular attention to only collect the required data for the completion of the service, in the most anonymous possible way. The strategy is simple; the best protected data is the data we don’t have.

By design, any Information Management tool we use, including the bot, must have a minimalistic approach regarding the data circulation. This means that we are looking for systems or software that, due to the way they were built, allow us to install them on local servers or download all data locally to be stored and secured by strong firewalls. We aim to avoid as much as possible, data being transmitted and stored online. The SOLIS bot is not an exception and, whilst it can be deployed in the cloud, we opted to use a local server deployment to maximise data protection.

How do you see the long term future of this project and the opportunities for scaling?

We believe that the SOLIS bot is a game changer in bridging the communication gap between humanitarian organisations and refugees. It allows people in need of humanitarian assistance to have access to support 24/7. As more users actively use the tool, providing feedback, the bot will continue to expand and improve. By designing a bot to be used via WhatsApp, a messaging platform that we know is extremely popular amongst Syrian refugees, we were able to overcome the adoption barriers that we were initially expecting. Our next goal is to develop the bot to be used via several different communication channels, including via SMS, to further increase accessibility.

The SOLIS bot is already being adopted by an additional international NGO as a way of communicating with their user base, and we are in conversations with other similar organisations. This is very exciting news for us, as it validates the work that we have done and shows that the SOLIS bot has a strong use case outside of Solidarités International. We are looking forward to collecting feedback, and we are excited that our technology is being used to build a greater community of users, whilst most importantly improving delivery of humanitarian services.

We have designed the SOLIS bot with open sourced technology, meaning that it can be used in any country and context, restricting its use to humanitarian organisations committing to comply to the sector’s protection principle. The next question is, who is ready to adopt it?

The M4H-Solidarités International project is in its final stages, and we look forward to sharing more lessons from the project in the near future. To learn more about the M4H Innovation Fund and the organisations we are working alongside, you can download the M4H Innovation Fund Portfolio (2017-2022).

This initiative is currently funded by the UK Foreign, Commonwealth & Development Office (FCDO), and supported by the GSMA and its members.
Donor logos