/ 6 March 2020

South Africa must have a stake in artificial intelligence technology

Artificialintelligence
AI cannot be responsible itself, but there are ways to ensure decisions made by AI are not discriminatory. (Science Photo Library/ Sergi Laremenko)

COMMENT

Last week the daughter of the president of Russia, Vladimir Putin, Katerina Tikhonova, was appointed to head the Artificial Intelligence (AI) Institute located at Moscow State University. The university has produced 13 Nobel prizes, six Fields Medals and one Turing award, so in matters of science, putting the AI institute there is a big deal. In Russian, if a husband’s last name is, for instance, Komlev, the wife’s surname becomes Komleva. Thinking algorithmically, you add an “a” at the end of the husband’s or the father’s last name to get the wife’s or the daughter’s last name. So Katerina’s surname is Tikhonova, which means that her husband’s or one of her paternal ancestor’s last name was Tikhonov. 

Why is Tichonov such an essential name in AI? AI models are ill-posed. To understand ill-posed models requires us to understand well-posed problems. Well-posed-problems have a solution, which is unique, whereas ill-posed problems do not have a unique solution. A unique solution means there is only one solution rather than multiple but confusing solutions. To make AI models have unique solutions, one has to regularise the AI models using the Tikhonov regularisation approach. Andrey Tikhonov was the larger-than-life Russian mathematician and geophysicist who invented this regularisation approach that makes unworkable AI models workable. Even though Andrey Tikhonov is not Katerina’s husband and her last name is adapted from her great-grandfather, the choice of her to head the AI institute, although a coincidence of surname, is significant. She is a rocket scientist with an impressive doctorate in analytical mechanics.

Last week Pope Francis, IBM and Microsoft joined hands to develop ethics for AI. Pope Francis, with 1.3-billion Catholic followers, is realising that AI, with its significant effect on all aspects of society, might require “divine intervention”. Last week the Pentagon, the military wing of the United States government, unveiled the five “ethical principles” for AI in warfare. We might differ, whether any form of war, can be principled. The first principle the Pentagon proposed is the responsibility of human judgment in AI warfare. The second is ensuring equity by eliminating AI bias. The third is to ensure that humans understand how AI technology works. The fourth is to ensure that AI machines are reliable. The fifth is that humans can control AI to avoid unintended harm. This Pentagon report came just as Elon Musk, in a fire chat conversation with General John Thompson last week, remarked that the F-35 fighter jet is becoming obsolete because AI drones will be more potent. 

AI is proving to be such powerful technology that it is revolutionising all aspects of our lives. Here in Johannesburg, we have studied how it is changing the field of finance and economics. AI is fundamentally changing the principles of self-interest, nudging people to act sometimes against their best interests. AI can understand interstate conflict and can make recommendations for peacekeeping. It is changing the medical field, and recently, researchers at the Massachusetts Institute of Technology used AI to discover a new type of antibiotic. Recently in China, the company Alibaba successfully developed an AI programme that can diagnose coronavirus with an accuracy of 96%. There are no aspects of our lives that will remain untouched by AI. 

The ubiquitous nature of AI requires us as South Africa to put into place mechanisms that will allow us to have a stake in this developing technology. Following from the State of the Nation address debate, where the Presidential Commission on the Fourth Industrial Revolution’s recommendations were mentioned, I published an article on the first recommendation, which was to build human capacity for the requirements of the fourth industrial revolution. This week I discuss the second recommendation, which is to establish a national AI institute. This institute should be a collaboration between the public and private sectors, because there is more capacity in the private sector than in the public sector.

AI has three aspects, and this institute must make choices on where it will invest its efforts. The first aspect is the theory of AI. Here we mean the thorough study of AI, its architecture and the associated mathematics. This naturally includes the development of new AI methods. The second aspect is the algorithmic part, which provides for coding. Fortunately, many companies such as Google and Microsoft have developed AI codes that they provide for “free”. (Of course, nothing comes for free; the Chinese company Huawei realised that the “free” Android software from Google was no longer “free” when the interests of the United States and China clashed.) The third aspect is the application of AI. There are multiplicities of sectors and industries that we can apply AI to, such as manufacturing, agriculture, medicine and retail. The AI institute will have to choose what areas of the economy it should invest in to put the South African economy at a competitive advantage. It should simultaneously co-create solutions with the rest of Africa. 

The AI institute should pay more attention to applications as well as the creations of AI solutions and apps, rather than only the theoretical aspects of AI. Of course, to create apps and solutions, one should be able to code. The AI institute should develop competencies in the area of integrating different software with different data sources to solve socioeconomic problems. The AI institute should be governed by a board or structures that have a fair representation of AI experts, people from the public and private sectors and from society. It should have visiting experts from global centres of excellence such as Silicon Valley in the US, Zhongguancun in China and Cambridge in the United Kingdom. It should work seamlessly with other similar initiatives such as the Absa Chair of Data Science at the University of Pretoria as well as the Institute of Intelligent Systems at the University of Johannesburg.

Furthermore, this AI institute should work with the Deep Learning Indaba, which is developing AI expertise in Africa and is working in 33 African countries. Incidentally, both the Deep Learning Indaba and the Absa Chair of Data Science, are headed by my former students. Separately, it should also work with initiatives such as Google Digital Skills for Africa and the data science community called Zindi.

Africa’s 1.3-billion people, increasing to two billion by the middle of this century, presents a huge opportunity. President Cyril Ramaphosa, when he took over as the chair of the African Union, recognised the centrality of AI for Africa’s economic growth. In consultation with the Presidential Commission on the Fourth Industrial Revolution, he announced the formation of the Africa AI Forum. This forum should exploit the emergence of AI opportunities in Africa. These opportunities have led to Google establishing the Africa AI Lab in Ghana and Microsoft AI Lab in Kenya. 

The institute should facilitate the expansion of AI expertise in Africa by drawing from the local population and international expertise. It should use strategic partnerships in bodies such as the AU, the Southern Africa Development Community, the East African Community, the Economic Community of West African States, Brazil, Russia, India, China and South Africa, the US and the European Union to facilitate the movement of people, expertise, skills and technology.

This is the second in a series of eight articles unpacking the recommendations of the Presidential Commission on the Fourth Industrial Revolution. Tshilidzi Marwala is an AI professor and the vice-chancellor of the University of Johannesburg. He deputises for President Cyril Ramaphosa on the South African Presidential Commission on the Fourth Industrial Revolution