For the wealthy elites, the fear of artificial intelligence has taken root and is based on the notion that machines will take our jobs. With the rapid growth of ChatGPT, the life-like chatbot, many people in the West are beginning to worry that it’s not just assembly workers and truck drivers who risk becoming replaced by robotics, but highly-paid knowledge workers too. Data analysts, accountants, financial advisers, coders, lawyers, and even Hollywood screenwriters now worry that AI will make them unemployed.
But the effect of AI on 100 or so countries and over four billion people in the developing world will likely be quite different. Countries with lower incomes employ fewer experts, and a higher proportion of their population employs people in fields which are less tolerant of automation, specifically agriculture. In poorer countries, the essential problem is not how AI will impact millions of workers but how billions of people will benefit from AI. The most revolutionary applications in developing countries are likely not to replace human beings but are those that provide human possibilities in a new way.
Most discussions about how to promote AI and limit the risks it poses have focused on wealthy countries where some universities and companies are working with the tech. But since the impacts of AI–good and bad–will manifest differently in less developed countries, the kind of investments and regulations these countries require will likely differ. Economists, philosophers and technologists have spent their time on the future of AI in the industrialised world. It’s time to formulate an AI strategy for all of us.
It has impacted the lives of people living in poverty around the world. Consider developments in credit. Many poor people need credit scores and histories and cannot access formal loans. The year 2010 was the first time I suggested the idea of generating alternative credit scores by employing algorithms to derive inferences regarding the probability of repaying based on the data taken by mobile networks. This is one of the many methods lenders across various countries have utilized to provide small loans through mobile phones to millions of people. Researchers are also using machine learning in this set of data to determine the households within a region that are the poorest to ensure that aid can be targeted with precision during the time of a crisis. Other researchers are using satellite images to refine population estimates using trends in the human population and forecasting food shortages based on patterns in vegetation. These programs demonstrate a particular benefit of AI in developing countries where there is a lack of information. Machine learning can take signals from new sources of information.
The possibilities aren’t just there. Consider schooling. Most education systems in developing countries need help to provide top-quality education. Personalised AI tutors and chatbots with endless patience could one day meet the requirements of affluent students at remote schools. They may also aid professionals in switching between different skills, allowing repair professionals to improve their abilities and master engineering. Also, they can help with health. In many developing countries, reliable medical advice is difficult to come by. AI-powered systems could provide a more accessible and better diagnosis. Many communities have the highest rates of depression and have a shortage of therapists. Digital tools for mental health, like chatbot therapists, could meet a need for real people at a minimal cost. AI could also play the same role in aiding people to navigate bureaucracies. For instance, an Indian business person who wants to venture into an entirely new market, for example, could one day rely on an AI-powered program to complete the necessary permits.
The technology that allows these possibilities will continue to advance as countries with wealth invest massive resources into AI. The most important thing for developing countries will be to enhance this stream of investment through this technology to create goods and services to satisfy local requirements. Developing countries have many of the necessary social infrastructure required to start new ventures, including universities, tech hubs and entrepreneurial groups. Companies need more motivation to develop applications targeted at the most disadvantaged population, which is rarely profitable to support. Certain large middle-income countries like India can afford to address this issue with the help of AI technologies that are geared towards people experiencing poverty. However, many other countries need more capacity and resources to make this happen. Therefore there is a need for entrepreneurs’ networks that can collaborate across borders and international organizations, such as the World Bank, which can help coordinate investments between governments and the philanthropic sector.
There are two significant routes AI tools could follow in developing countries. First, discover the task AI is becoming proficient at in rich countries, and then adapt it to less developed countries. For instance, many entrepreneurs are working on chatbot tutors for schools with high-end infrastructure and tools that can be modified to function in locations with poor Internet connectivity and higher student-to-teacher ratios. The other is to identify entirely new AI applications or products that can be tailored to the specific requirements of people living in developing countries. For instance, using an AI-powered finance planner designed for farmers who live on subsistence could aid them in managing the risks associated with deciding the best time to plant. Some innovations began in poor nations but only reached richer countries afterwards. Kenya’s M-Pesa mobile payments system was one example. It took off before similar apps made it to the United States.
While specific AI tools from wealthy countries can be practical from the start for developing countries, some require tweaking. A significant issue is that many AI tools have been built on specific data for advanced countries, which comes from people with a high living standard and are usually written in English. Most of the world’s written information is geared towards the less fortunate or presented in languages that are not widely spoken. In addition, AI systems are typically designed to make results that satisfy wealthy customers in the West which means that they can fail when dealing with less wealthy ones in other countries, for instance, welcoming customers with their first name in a society which views such a gesture as insensitive.
Wealthy Western societies were able to begin to accumulate training data, which is why it takes some time to allow AI models to reflect the diversity of people in other countries accurately. However, this process can be speeded up. Researchers can identify applications that can be transformative, even if they only can improve the information behind them to make it more accurate. An AI-powered medical adviser, for instance, could assist people with elevated blood pressure within Silicon Valley but less useful for someone living in Lagos suffering from malaria since it doesn’t have exposure to local medical issues. A similar system could become popular among English users, but it may it isn’t accessible in Yoruba, which is Nigeria’s most famous native language.
To fill in the gap of data from the developing world To make up for this, new data must be developed for models to build on. Crowdsourcing can help in this regard. The WikiAfrica campaign, for instance, to coordinate the introduction of African material to Wikipedia. These initiatives are more significant since this knowledge can help machines make better decisions. In other domains where correctness is harder to discern–such as medicine or agriculture–crowdsourcing will not be enough. Experts must be sourced, and analogue data, such as clinic paper records, must be converted to digital. Representation is just one part of the equation, as developers must decide between groups with different beliefs. For instance, other religious communities in India might disagree on what constitutes appropriate medical advice.
Another problem when it comes to the import of AI to developing countries is one of technology. Despite considerable advances in the developing world, it is less advanced than the developed world in various technical standards. Certain AI apps will need more accessibility to mobile phones, greater Internet connection, and even digital record-keeping systems that track students’ performance at schools or medical conditions of the patients in hospitals or the outcomes of court cases. In the case of AI, like earlier waves of technological innovation, the most important thing to do is to distinguish between apps which could be beneficial quickly and those that remain within an area of sci-fi in the foreseeable future. The line between the two will shift and will differ between different fields. For instance, medicine has less tolerance for mistakes that AI systems will make, and agriculture is based on complex contextual variables that are easily understood by farmers but challenging to communicate in AI systems.
THE LIMITS OF LAWS
In both the developed and developing world, the widespread use of AI could pose a risk. In the developing world, there are many different dangers, and we have less control over the advancement of technology. The primary concern is whether AI remains centralized, which means controlled by a few tech companies. Centralized AI systems will likely be subject to regulation in big markets, such as that of the United States and the EU. Smaller markets will be able to exert small pressures and remain within the shadows of U.S. and EU regulations. While they may block access to a centralized system –for instance, blocking servers similar to what some dictatorial governments have implemented through Twitter, Facebook, and YouTube, they won’t be able to stop AI-generated content from travelling across boundaries.
It’s not sure what the future holds, but whether AI will be centralized is unclear. Open-source alternatives like Llama (a large-scale language model created by the owner of Facebook, Meta) and Stable Diffusion (an image generator created by the start-up Stability AI) are gaining popularity. Anyone with an internet connection can manipulate and run these decentralised systems. If they become helpful and efficient, it would be impossible for any country to regulate them directly. But open techniques may be easily modified to local requirements as they are generally available for use at no cost since anybody can alter their code. Due to the need for levers for regulation, developing countries might have to accept changing to the latest technology instead of controlling it. To minimize risks, they need to concentrate on regulating AI and the industries that use it—for instance, consumer protection laws that make the companies accountable when a product is dangerous, regardless of whether or not it is using AI.
AI has started the discussion about regulations in the rich world. But many of the suggestions to address the risks of AI could be inadequate in the context of poor countries. The Western regulators cannot determine how the rules function in different environments, and a system approved to be safe in Brussels could not perform as efficiently in Bangalore. Furthermore, Western regulators’ standards could be too rigorous in locations where available alternatives for using an AI application are more shaky. Forecasts for weather, for example, are optional in order to enhance the options available to farmers in countries that are developing. In higher-risk areas like medical treatment, AI may soon be superior to the existing choices that are available to the disadvantaged. A 2023 study assessed the quality of care in low-income countries to determine the percentage of cases that were handled correctly. The result was lower than 50.
In the same way people living in a developing nation is also more prone to risk to risk than a person from the industrialised world. A lot of people living in developing countries need help to contest automated decisions, like the refusal of a loan request. This new AI technology often results in lower quality than what they promise and it’s too easy for businesses to disregard issues that arise in the lower income group. That’s why it’ll be crucial for regulators to make sure that consumers have the right processes to file complaints and appeal the decisions.
A lot of people in the developing world are naive to the concept of AI and have yet to have any idea or heard of algorithmic thinking before. Therefore, care should be taken to ensure that you communicate effectively. The investigation I conducted with Joshua Blumenstock and Samsun Knight proves this is feasible. We provided poor Kenyans the app which paid their bank accounts according to how they utilized their phones by using a similar algorithm to the ones that assess creditworthiness. When the subjects were provided with clear descriptions of the algorithms’ working to determine their behavior, they changed their behaviour–a clear sign of comprehension.
Political hurdles also exist. Realistic fakes – real photos, videos and audio clips created by artificial intelligence can have a particularly negative impact on developing countries in which the systems of government tend to be fragile and trust between different groups is usually very low. When people realize that media can be created and manipulated, they are less likely to believe that content is in fact real. To avoid these issues, civil society organizations can build the trust infrastructure by spreading awareness of the possibility that content is faked, and also establishing independent media outlets which establish reputations for the vetting of the authenticity of content.
AI can also allow new ways of monitoring, for example, tracking people using smartphones and facial recognition. Most developing countries in the market for advanced surveillance tools don’t create their own, but rather import them, typically from China. This means that the use of AI-powered technology can be scattered, making it much easier for the data obtained to be released to third parties and rights to be violated in various ways. Again civil society will play an important role to play in watching out for new technologies and calling attention to violations.
BACK to the future
The new wave of AI has brought new challenges and possibilities at a record speed. We’ve witnessed similar technological changes before. While mobile phones were initially made for the wealthy, they gained popularity among those in need over the last 20 years. The developing countries benefited from the standard hardware, such as antennas and handsets, that the West manufactured. Telecom companies developed business models that helped the less fortunate, like the pay-as-you go cell phone plan. Entrepreneurs founded new organisations that let people make use of phones to transfer money, access credit, and even check prices. These inventions allowed mobile phones to rapidly reach the vast majority of the world’s poor, and bring them into the world economy.
It is precisely these links which have set the tone for the development of AI. However, despite smartphones’ popularity, this innovation has yet to fulfil its potential in the emerging world. Much of the private sector’s innovation has been concentrated on the requirements of the rich. More money has been invested in apps that connect rich consumers with drivers or vacation homes as well as cooked meals rather than apps that connect farmers who are struggling to access markets and remote children to school. Innovation in the private sector AI will likely transform numerous industries from health to education to law. However, harnessing the full potential of this technology for the developing world will require forming a broad idea of what’s possible, and paying particular focus on those who’s lives could be affected by it.