Fairness, Transparency & Data Protection - Building Blocks to People-Centred Artificial Intelligence Systems

There is an increasing call for stronger privacy and data protection legislation across the globe. This has been as a result of the growing concerns around the abuse and misuse of personal data by several key actors. This growing call has been around the issues of ethics, fairness, transparency, privacy and disregard for the fundamental human rights in the development, deployment and use of technologies including artificial intelligence. How do we protect the privacy of individuals when building large-scale, AI-based systems? Though challenging, addressing these concerns are not insurmountable. Integrating fairness, transparency and data protection in machine-learned models and systems is therefore not just essential but critical to the growing innovation and sustenance of AI technologies in the future. What has become increasing clear is that these systems and models are being developed with very limited recognition of their impact on the inherent rights of individual users. This topic therefore explores the importance of developing people-centered AI and why incorporating fairness, transparency and other data protection principles are key to building such models.

National Digital ID Projects: Do Men and Women's Data Count the Same?

From research on digital ID, some gender perspectives stand out: First, official data on digital aspects of life, for example mobile device ownership and access to the internet is not dis-aggregated for gender. Although we know that there are less women using digital devices, we are not sure of the numbers, demography etc. Second, the upcoming projects on digitalisation are premised on patriarchal laws and cultural practices. While a project such as digital ID should be an opportunity to create a system that facilitates flourishing of society, we risk entrenching patriarchy if we base digital ID on same old practices. Lack of gender perspectives in public sector digitalisation affects how women experience technology. I would like to use the opportunity to explore how public sector tech e.g digital ID can be responsive to gender needs?

ICT Access in Africa in the Age of Artificial Intelligence

Basking on the ubiquitous adoption of mobile technology in Africa, experts in the technology domain prognose a similar upswing in the application of artificial intelligence (AI) especially in the communications space and expect it to help leapfrog critical challenges on the continent. Emerging technologies such as AI portend significant opportunity for Africa in the critical areas of digital job creation, efficient e-governance and public service delivery as well as improving multilingual communication. As we know, AI is driven by three key change mechanisms - complex computing power, autonomous algorithms, and increased data (structured and unstructured) capture and storage from online searches, social media and connected devices. Therefore for countries in Africa to harness the potentials of this emerging technology, there is a need to address the array of policy issues with respect to the creation of inclusive ICT access and use especially for internet-based services. This will include providing investment friendly environments for ubiquitous new generation networks on which the new innovations will run as well ensuring service affordability across-pyramid provided within competitive markets. While in recent years there have been significant improvements in both the quantity and quality of telecommunications infrastructure, Africa still has critical gaps to fill in the light of emerging technology adoption to ensure adequate connectivity. This will be important for the fledgling techpreneural ecosystem with regards to the development of local AI technology and services, without which the continent will continue to rely on foreign expertise with critical imperatives for African sovereignty in the longer term. In this talk therefore, I shed light on the inherent digital inequalities based on hierarchical digital divide in Africa with respect to ICT access and use in the age of AI. Beyond access, I highlight critical gaps in more nuanced analogue foundations such as digital skills and technology use and its imperatives for maximizing the potentials of new technologies in the digital economy. In summary, in line with the ponderings of Ballim and Breckenridge (2018), I try to answer the fundamental question: ‘’What does it mean for Africans that the richest companies, and the most powerful governments, are investing heavily in technologies, programmes and infrastructures of artificial intelligence?’’

Explainable Deep Learning for Natural Language Understanding

Deep learning models have made ground-breaking success in all areas of AI, however, these models perform poorly on tasks that require commonsense reasoning. Because these models are black-boxes, it is hard to predict why and when they fail. Explanations make AI systems more transparent and also justify their predictions. We collected human explanations for commonsense reasoning in the form of natural language sequences and we found that these human explanations contained world knowledge that humans pick-up just by living. Our work involves training a language model to generate human-like commonsense explanations in an unsupervised way. Using these auto-generated natural languages explanations, we not only make the models more transparent but also improve the state-of-the-art on commonsense question answering by 10%.