Inventing Artificial Intelligence in Africa
Alan Blackwell is Professor of Interdisciplinary Design at the Computer Laboratory in the University of Cambridge and an alumnus (m. 1995) and Fellow of Darwin College. He is also Director of the Crucible network for research in interdisciplinary design and co-Director of Research for the University of Cambridge Global Challenges Initiative. Here he talks about the opportunities and challenges of AI in the Global South.
Increasingly many aspects of our lives are influenced by, or even controlled by, Artificial Intelligence (AI) algorithms. But we wonder what benefits result from these algorithms, and who the benefits go to. Governments, health services, and even schools and universities can be enthusiastic about the savings that might be achieved by automating public services. But many of us are uncomfortably aware that these technologies are driven by the agenda of the companies that create them. These new public services (and even university research) are increasingly dependent on the tools and technical vision of companies like Google, Tesla, Amazon and Facebook.
As an AI engineer since the 1980s, who came to Darwin for a PhD studying humancomputer interaction, my own research is dedicated to making AI research more human-centred. This has always required working across disciplinary boundaries, but in recent years two particular interdisciplinary initiatives have inspired a unique approach to understanding new opportunities in AI. One of these has been the work I have done with Darwin alumnus David Good and many other collaborators, developing Cambridge Global Challenges, a university-wide strategic research initiative that is focused on addressing the Sustainable Development Goals in partnership with researchers in the low and middle-income countries of the Global South. The other interdisciplinary initiative feeding in to this work is a fascinating series of workshops convened by previous Darwin Master Geoffrey Lloyd, who has extended his work in the history and philosophy of ancient science with anthropological and cross-cultural perspectives under the rubric Science in the Forest, Science in the Past.
These two activities, each very stimulating in themselves, demonstrate everything that readers will join me in appreciating about the interdisciplinary research community of Darwin. But beyond the pleasures of interdisciplinary curiosity, this mixture of activities has led me to ask why so much development and critique of AI technologies is concentrated on the problems and concerns of people who live in wealthy countries. I started to ask what AI would look like if it were invented by people in Africa, drawing on things I had learned from other Darwin alumni including anthropologists James Leach, Lee Wilson and Amiria Salmond.
An initial plan for a new ethnographic approach to AI was hatched at one of the Science in the Forest, Science in the Past workshops, and developed into projects in specific countries through contacts I had made in Cambridge Global Challenges. The year of sabbatical leave originally dedicated to this work has been interrupted by the pandemic, but I did manage to complete two fascinating periods of field research, including two months in Ethiopia at the end of 2019, and a month in Namibia in early 2020, before it became necessary to return to lockdown in Cambridge.
Although seriously interrupted by the pandemic, it has been possible to publish some findings from this research, coauthored with colleagues in Ethiopia and Namibia, and also with my wife Helen Arnold, a high school mathematics teacher with educational research experience (and known to some Darwin alumni as the soprano soloist on Darwin College choir tours to Portugal in the late 90’s), who was able to work with collaborators in both countries. The initial results of this research have been re-evaluating the mathematical concepts underpinning contemporary AI research, working from the perspective of teachers and students in these countries.
For example, one of the real challenges for members of the public today is how to interpret and act on the outputs of machine learning and data science algorithms. The Covid pandemic has especially emphasised how data in itself does not seem sufficient for us to know how to “follow the science” as the UK government has advocated. A master’s student is currently experimenting with teaching concepts of probability in relation to the risk of contracting Covid in Nigeria. In fact the mathematical principles of information and inference that underlie all current AI, machine learning and data science, as described in the classic book by Darwin fellow David MacKay, ‘Information Theory, Inference and Learning Algorithms’, are hardly taught in Western schools, let alone students in Africa. At the start of this project, we had the good fortune to meet with experts in the Cambridge Mathematics project who were revising the fundamental conceptual principles of future teaching in probability and statistics. To the extent that future populations are comfortable with AI, whether in Africa or the UK, it will be these aspects of the school curriculum that prepare them. Instead of assuming that new curriculum is always exported from the West to the rest of the world, why not start in places like Africa, where different ways of teaching and learning can enrich us all?
Findings from our research demonstrated how new concerns for international mathematics curriculum could be derived, not only from the skills needed in universities like Cambridge, but from the perspectives of students on the shore of Lake Tana in Bahir Dar, Ethiopia, or in Tsumkwe Senior Secondary School in the Namibian Kalahari. In both locations we met talented data scientists, very different in some ways to students in Cambridge, though also local AI researchers developing their own tools and agendas, for example in Ethiopia’s Bahir Dar Institute of Technology and the International University of Management in Namibia. We found it particularly interesting to work with the gifted senior high school students at the Bahir Dar STEM Centre, and the ancient data science traditions of huntergatherer Ju|’hoansi people in the Kalahari. In both places, we saw opportunities for AI systems to be designed that built on the cultural strengths specific to a country, such as the traditional observational skills of hunter-gatherers, or the collaborative community of the Amharic tradition. More technical detail of these findings in relation to computer science literature and other disciplines is described in the publications mentioned below.
The original research programme was ambitious and wide-ranging, and very much continues at the time I write this. Current Darwin PhD student Joycelyn Longdon, who I supervise together with Jennifer Gabrys from the Cambridge department of Sociology, Adham AshtonButt from the British Trust for Ornithology, and Emmanuel Acheampong from Kwame Nkrumah University of Science and Technology in Ghana, is currently conducting fieldwork to understand how data science contributes to forest conversation. Joycelyn’s PhD is the first fieldwork-based social science project carried out in the Centre for Doctoral Training in AI for Environmental Risk, founded in 2019 by Darwin fellow Emily Shuckburgh.
Other students and researchers, including many Darwin members, are building on these results to learn from the Global South, understanding what kinds of technology will best equip all of us to address global challenges.
Acknowledgement I am very grateful to my hosts and collaborators in Africa, especially Dr Tesfa Tegegne at Bahir Dar University, Ethiopia and Prof Nic Bidwell at the International University of Management, Namibia. Thanks also to the friends in those countries seen in these photographs, and to those listed as co-authors in the papers below.
Further reading: Blackwell, A.F. (2021). Ethnographic artificial intelligence. Interdisciplinary Science Reviews 46(1-2), 198-211, DOI: 10.1080/03080188.2020.184022
Built with Shorthand