Good practices and lessons learned.

 

The case studies considered are often limited in scale or in the pilot stage, but serve to illustrate the potential of AI in developing countries and how difficulties can be overcome through careful implementation and cooperation among stakeholders. There are no one-size-fits-all solutions, but a good starting point in each country is to assess local conditions and technological capacities and adopt AI strategically. This may mean, for example, supporting startups and industry–university collaborations, as well as non-profit organizations that help deploy AI solutions to serve local needs. Governments should favour the emergence of AI ecosystems with investments supporting business development and networking. By showcasing successful experiences of AI adoption, they can raise awareness and diffusion and favour the accumulation of complementary assets and experience. It is also useful to engage with large companies or international organizations that can support promising local businesses with emerging technologies and connect them with international markets. This allows developing countries to accumulate relevant complementary assets and experience for the extensive and impactful diffusion of AI. There are four main takeaways from the case studies along the key leverage points of infrastructure, data and skills, as well as partnerships (figure II.4).  

AI adoption should be designed according to the available digital infrastructure. In Colombia, for example, the banana disease detection application Tumaini has an offline mode that retains most of the diagnostic functions in the absence of an Internet connection, thereby remaining accessible and useful to farmers in rural areas where Internet connectivity is limited. Similarly, AI adoption should take into account unstable supplies of electricity. The AI-assisted X-ray machines in South Sudan and Tajikistan, for example, operate on battery power and can therefore reach remote areas. Other case studies highlight different uses of AI applications based on mobile telephones, which offer a scalable platform for AI applications.

AI depends on high-quality, relevant and interoperable data sets. In developing countries, such data sets may be limited, difficult to access or expensive to pay for, and innovative ways of collecting and using new forms of data are therefore key in ensuring AI capabilities and effective adoption. In Brazil, for example, in modelling refugee lows at the border during the COVID-19 pandemic, UNHCR researchers relied on an unconventional nowcast data set, which included indicators scraped from local sources, then integrated, to produce accurate predictions of refugee movements. Alternative data sources become viable and help overcome data limitations if the right AI techniques are applied. As shown by the case studies, in China, for example, deep neural network techniques enabled the use of open-access data in rice yield predictions and, in Nigeria, the Ubenwa application used deep-learning algorithms to employ anomalies in infant cries as a reliable indicator of a health complication after birth.  

One of the main impediments to technology adoption in developing countries is a low level of digital literacy. Governments need to build greater digital capacity. In addition, designers need to consider current standards of digital capacity and build applications that are attractive and simple to use, particularly on mobile telephones. Simple interfaces help facilitate interactions by novice users with new technology solutions and thereby help promote widespread and inclusive diffusion. For example, in the United Republic of Tanzania, a chatbot for maize diseases allows users to access diagnostic information and make queries in a manner similar to messaging family or friends. Application-based AI tools and visual aids such as icons and illustrations allow for an intuitive understanding of available functions. Such designs smooth the experience for those who may be unfamiliar with new technology and are critical in AI adoption in developing countries. 

Developing countries aiming to accelerate the adoption of AI can benefit from strategic partnerships. A cross-country study by the World Bank showed that firms in developingcountries that adopted more sophisticatedtechnologies tended to be those with more external collaborations, through universities, foreign trade partners or large multinational corporations. Building strategic partnerships enables aspiring AI adopters to overcome barriers to adoption. In addition, Governments can overcome limitations of size through regional collaboration. For example, in many countries in East Africa, Swahili is a common language; a group of countries could collaborate to pool data in Swahili and jointly engage with technology companies to address common linguistic challenges. Strategic partnerships can also provide essential resources for AI. Global Grand Challenges, under the Bill and Melinda Gates Foundation, for example, currently supports the development of AI models in local languages. The AI model for predicting the risk of dropping out among subscribers of a service provided by Armman was developed with technical assistance from Google India. In addition, in Tunisia, ifarming has a partnership with IBM to use high-performance AI platforms and receive funding to expand its operations. Chapter V further discusses the importance of international cooperation in global AI governance and suggests policies for ensuring that AI works for all.