UNESCO and partners launch the Open for Good Alliance, an initiative to leverage synergies across likeminded international, regional and national organizations that are working to stimulate the use of localised data in artificial intelligence systems to address local problems.
The Alliance was launched last month by the International Development Research Centre, FAIR Forward / GIZ, Mozilla Foundation, Radiant Earth Foundation, <A+> Alliance for Inclusive Algorithms, African Institute for Mathematical Sciences, Kwame Nkrumah University of Science and Technology, Makerere University and UNESCO with three objectives:
Making training data sets openly available, helping existing open training datasets to be found and supporting their maintenance.
Facilitating the coordination and exchange of good practices and ideas through discussions with community members around the collection of datasets and applications of AI, supporting the development of standards where needed, and share successful examples.
Increasing the public awareness for the benefits of openly available, unbiased and localized training data.
During a discussion at the launch of the alliance, Ernest Mwebaze, Lecturer at the School of Computing & IT, Makerere University, underlined that sharing of data is important as it facilitates reproducibility of results of research based on the data by other researchers and helps in the optimum use of resources by avoiding duplication in collection of similar datasets. The alliance, through its objective of creating AI commons based on localized data will help in making the data findable, accessible, interoperable and reusable.
At the Open for Good Alliance launch, UNESCO, as part of its work on Universal Access to Information and Digital Innovation and Transformation, emphasized that it would strive to strengthen North-South and South-South cooperation on the development of inclusive AI commons with localized data as part of the alliance. Cédric Wachholz outlined UNESCO’s vision of “leveraging the power of open data for innovation and social good, in order to shape an inclusive and accessible digital environment”.
The development of artificial intelligence systems requires algorithms to be trained on data. However, localized data are not always available and its absence poses several challenges. For instance, in the case of low resourced African languages enough data is not available for the users of many indigenous languages to develop AI-enabled products and services. In this regard, UNESCO is supporting the development of datasets in several low resourced African languages spoken across 22 countries and reaching 300 million speakers. Further challenges exist in making data findable, interoperable and reusable.
The training data used also influences the efficacy with which an AI system performs its tasks. As a result, another challenge exists in the form of biased or discriminatory outcomes due to the use of training data that may be biased against certain social groups. When these algorithms are integrated into products and services that enable, for example, decision-making, they can produce discriminatory outcomes.
The founding members of the alliance invited other likeminded organisations to join the alliance to address these challenges in the development of AI systems.
To join the alliance please write to the Open for Good Alliance Secretariat on the alliance website at https://www.openforgood.info/#Work