When I completed my master’s degree in artificial intelligence (AI) 20 years ago, it had yet to emerge from the “AI winter” of limited funding and interest. Today, AI can beat world chess champions, diagnose medical conditions, predict our emotional states, and drive unmanned vehicles. Many consider AI to be the leading opportunity for economic growth, with the potential to increase global GDP by $15.7 trillion by 2030.
On the other hand, we have already seen the troubling potential for AI to cause harm. Biased AI applications have discriminated against women and minorities in corporate hiring and criminal justice matters. AI has been used in efforts to sway elections, manipulate public opinion, and inflame ethnic tensions. Some experts predict that in the years ahead, AI-driven automation could massively disrupt labour markets, resulting in large scale unemployment.
In short, AI is something of a double-edged sword, and examining its Jekyll and Hyde nature can lead to an emotional roller-coaster ride — from excitement, to concern, to hope.
In simple terms, AI is a branch of computer science dedicated to developing systems that can carry out typically “human” tasks such as decision-making or forecasting, natural language processing, pattern recognition, and solution optimization. In practice, AI is a set of foundational technologies that are often working behind the scenes to enable revolutionary new ways of working, organizing, and producing. Advances in computational power and, more recently, the ability to gather massive amounts of data through the Internet and social media, have brought us to a point where AI is beginning to play an active role in our daily lives. The future remains unwritten, but it’s clear that AI will be transformative.
While the impact will be global, the most transformative change will happen in the Global South, where the need and the potential benefits are greatest. In areas ranging from education to agriculture to medicine, AI could improve the lives of tens of millions of people by addressing illiteracy, identifying threats to crops, and improving access to high-quality healthcare in remote and under-served regions. In fact, AI-based applications are already showing promise in each of these areas by directly addressing shortages of expertise and expensive equipment.
But as with all technologies, AI can produce negative or problematic outcomes, and because its power and impact can be far-reaching, it has the potential to increase existing inequality and instability. In fact, an existing AI divide in the Global South could make such outcomes even more likely.
At one level, this divide relates to the raw materials of AI: technology infrastructure and data. The availability of technology varies widely, with about 4 billion people still without Internet access. At the same time, the large datasets required to train AI systems tend to be generated in the Global North. As a result, this data is often not appropriate for, or even available in, the Global South.
In addition, the data itself can reinforce existing inequalities or embody economic, social, and cultural biases. AI trained on these datasets can learn, reproduce and, through their application, even amplify those biases.
We are already seeing examples of this close to home. Amazon’s AI-powered recruiting technology, which was trained predominantly on men’s CVs, was found (not surprisingly) to favour male applicants. In Florida, an AI system used to inform sentencing and parole decisions by predicting an individual’s probability of reoffending wrongly labelled black defendants as reoffenders twice as often as white defendants.
Another level of the AI divide relates to the design, development, and use of AI technologies. Of the $15.7 trillion in predicted GDP gains by 2030 that is noted above, only 11% are expected to accrue in the Global South. In terms of job losses due to AI-based automation, the World Bank estimates that two-thirds of all jobs in developing countries could be affected, and these countries are often the least equipped to provide social safety nets. Then there is the limited capacity to prevent and respond to intentional misuse of AI for illicit purposes.
This analysis may be sober, but it is also essential for understanding and navigating potential pitfalls. Our report, Artificial Intelligence and Human Development: Toward a Research Agenda, seeks a way forward in which locally relevant, accountable, and ethical AI applications can flourish and contribute to advancing sustainable development goals.
IDRC is starting several lines of work to address the AI divide as part of a long-term plan for building local capacity to ensure the ethical development and deployment of AI applications. We will launch a network of excellence in sub-Saharan Africa, the first such regional network to connect AI researchers with social scientists, ethicists, development actors, policymakers and sources of funding.
We are working to improve education in the Global South so that children can develop capacities for problem solving, innovative thinking, continuous learning, and resiliency. We are also engaging with other like-minded funders to set up an “AI for development” initiative that includes a research-to-policy and capacity building mandate.
I’m still concerned about the future and the possibility that AI could intensify both inequality and instability in the Global South, but most of all I feel energized and hopeful about the role IDRC can play to help forge a positive AI future for all.