SDG 3: Good Health and Well-being
SDG 5: Gender Equality
SDG 7: Affordable and Clean Energy
SDG 9: Industry, Innovation and Infrastructure
SDG 10: Reduced Inequality
SDG 13: Climate Action
SDG 17: Partnerships to achieve the Goal
Please describe Other
Health, Clean Energy and potentially others
2. Project Details
Company or Institution
FAIR Forward – Artificial Intelligence for All
General description of the AI solution
The German Development Cooperation initiative “FAIR Forward – Artificial Intelligence for All” strives for a more open, inclusive and sustainable approach to AI on an international level. To achieve this, we are working together with six partner countries: Ghana, Rwanda, Kenia, South Africa, Uganda and India. Together, we pursue three main goals: 1) Strengthen local technical know-how on AI, 2) Remove entry barriers to AI by helping to build open AI training data sets as digital public goods, and 3) develop policy frameworks ready for AI. In this context, the initiative develops pilot applications built on open AI training data to improve the life of millions of people and inspire local AI developers. In the first phase, we focus on conversational AI systems in local languages, as open voice technologies promise to make basis services available to millions of citizens currently underserved. Together with local partners and the Mozilla Foundation, FAIR Forward builds open training data sets in Kinyarwanda, Kiswahili, Luganda and a set of Indian languages. First applications include a Covid-19 chatbot that provides crucial information to Rwandan citizens and will serve as the central health information hub of the Rwandan government. In a second phase, FAIR Forward focuses on AI and climate change, expanding and working with geospatial datasets, ground truth data and other openly available data sources to builid i.e. an application for site-identification for green minigrids. These activities are complemented by work with the Lacuna Fund to address the shortages of openly available AI training data sets for NLP, agriculture and climate change. Additionally, the projects cooperates with UN Global Pulse and Smart Africa to develop AI policy frameworks, e.g. with the government of Ghana, and offers traineeships, webinars and e-learning courses on AI to the youth in partner countries.
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
With the first applications, we develop conversational AI systems in local languages, applying machine learning and natural language processing (NLP). With collecting and making available representative, problem-specific training data sets under a CC-0 license for languages such as Kinyarwanda and Luganda, FAIR Forward addresses a critical market failure. So far, the lack of data has prevented local innovation ecosystems to develop voice assistants and related applications specific to local contexts, leaving millions of people, especially illiterate citizens, digitally excluded and underserved with basic services. The applications we develop are one of a kind, owning to the scarcity of AI training data sets and constrained capacity for local AI development. We develop these applications in collaboration with global partners such as the Mozilla Foundation, the Bill and Melinda Gates Foundation and local partners such as Digital Umuganda and Masakhane. We have a strong focus on local AI ecosystem development to ensure sustainability of the approach and see digital public goods as one component to do so. So far, we and/or partners presented our approach on some conferences, including the UNESCO conference on language technologies and re:publica 2019. We also follow closely conferences such as ICLR but have not yet presented our work there. Our approach was mentioned as an example in the GPAI report on “Areas for future action in the responsible AI ecosystem”. Our applications have the potential to reach millions of people; the Covid-19 chatbot in Rwanda is proof of this, reaching 150,000 users and facilitating more than 2,000,000 interactions in the first 10 days of its launch.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
Conversational AI systems in local languages have the potential to reach millions of citizens with critical information and enable access to services. Kinyarwanda speakers amount to 15 million across Rwanda and Uganda, while Kiswahili is one of the biggest languages spoken by circa 80 million people in East Africa. Since the underlying data used to train our algorithm is openly available under a CC-0 license, literally everyone across the globe can make use of it, especially local innovation ecosystems consisting of start-ups, non-profits, businesses, government, and research institutions. As a project of the German development cooperation, the project works according to a log frame and implements rigorous monitoring and evaluation practices, including, but not limited to customer satisfaction surveys and downloads of the AI training data set repository via Mozilla Common Voice. It is the principal aim of the project to solve development problems with AI by giving local communities the agency, tools, and capacities to do so, bearing in mind and mitigating critical aspects such as data colonialism, biases, and surveillance. In addition, FAIR Forward supports global initiatives of AI for SDGs such as the Lacuna Fund and is a founding member of the Open4Good Alliance.
Since the project’s objective focuses on AI training data sets as public goods, including language data and geospatial data, applications developed based on the data can span a wide range of SDGs (see indication above). Our pilots especially address SDGs 3, 5, 7, 9, 10, 13 and 17. They will be openly available on GitHub to be taken up, localized, and improved by whomever it might serve.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
The first user numbers of the Covid-19 chatbot (see question 1) suggest the application delivers valuable information to citizens and has a positive impact on the knowledge of current Covid-19 regulations in Rwanda. The chatbot proofs conversational AI systems are attractive for citizens and promise high uptake of services provided, under the precondition that systems are developed bearing in mind disparities in dialects, gender, geographies, prior knowledge etc. Challenges to scaling include 1) knowledge of the chatbot especially in rural areas; 2) updating health information beyond Covid-19 on a regular basis, if not daily; 3) liabilities in case the chatbot is to provide not only information but also gives advice on health issues (future scenario); 4) integration with ticketing system of government partner. Regarding scaling to other countries and geographies, the main hurdles include 1) open AI training data for language adoption; 2) localized health information; 3) local AI development capacity. In addition, the project needs a local partner with a business model and/or internal or external funding to ensure long-term sustainability.
Since the project’s objective is the creation of open AI training data sets (CC-0) and includes capacity building for local AI ecosystems, the emergence and growth of local companies is one of its success measures. The team works actively on the local level to support the take-up and usage of data sets by holding meetups, offering fellowships etc. On the global level, FAIR Forward is a funder of the Lacuna Fund, which supports the emergence of a global AI ecosystems for sustainable development by providing grants to universities and research institutions in the Global South. The project is currently introducing an oversight board to navigate compliance with GDPR and local standards, and get valuable input from local experts to ensure appropriateness and sensibility to local contexts and communities.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
As a project of the German development cooperation, it is our goal to support the ethical and lawful development of AI solutions to address development challenges. We are very aware of the challenges arising with AI development, especially in the Global South and have different approaches in place to address these: 1) we provide trainings and handbooks for the ethical development of AI systems to local developers; 2) we require our own development partners to adhere to local and international standards as well as ethical guidelines, and offer support in doing so – one case, for instance, included a bias in voice data sets in Kinyarwanda which where biased towards young and educated people; 3) we are currently introducing a project-wide oversight board to give guidance; 4) we support government in the development of ethical AI regulation and frameworks, inspired by the European Union’s approach. The first application (Covid-19 chatbot) is trustworthy in the sense described above and will function as a pilot for similar applications in additional partner countries. However, robustness testing will need to be further undertaken.
The chatbot is an AI solution made possible by Rwandans for Rwandans, and mirrors FAIR Forward's approach to make AI development more open, inclusive, and fair. In doing so, we challenge existing power asymmetries and aim to support local AI communities' struggle to obtain a place on the global AI table.