One of last year’s top buzz-phrases – artificial intelligence (AI) – shows no sign of going out of fashion in 2018. Concepts once confined to the realms of science fiction, such as driverless cars, are now being championed as a near-future option for the everyday person. It is difficult to argue against the giant technological strides companies such as DeepMind have pioneered, despite some critics’ concerns. The notion of machines imitating intelligent human behaviour has captured the wider public’s imagination with UK adults being ‘broadly optimistic’ about the technology.
But AI has the potential to be a powerful tool in less obvious realms – like that of international development. The development industry has increasingly embraced these advances in tech: the UN Office for the Coordination of Humanitarian Affairs now provides an online training course on AI’s role in poverty reduction, while at the World Bank and International Monetary Fund (IMF) have ever-growing ICT sector teams.
When it comes to international development, the most utilised subfield of AI is machine learning. Broadly speaking, machine learning is an application of AI that allows systems to perform tasks without explicit prior programming. For instance, an autonomous hedge fund now exists, where historical and current data – including that of previous judgments – are compiled and processed as part of an automated system designed to make new decision based on precedents. Considering this fund significantly outperformed many generalised funds, the development industry may be justified in their excitement.
The role AI can play in international development still needs to be defined, although many have focused on the use of large datasets – or ‘big data’. Like most industries, the cornerstone of development lies in the ability to design policies on evidence-based solutions. But this remains a big obstacle for the development sector, where there is an overwhelming emphasis on impact evaluation. Assuming, as the Overseas Development Institute does, that an evidence-based approach does deliver higher quality outcomes, the potential roles of machine learning and big data in driving poverty reduction become evident.
Organisations such as AI-D have claimed that machine learning can improve the overall quality of development data, especially through predictive analysis. Through an automated system of data capture and handling, policy responses to development issues can be guided more effectively.
Further claims by AI-D include being able to develop workable models in environments that are difficult to observe directly. For instance, pandemic outbreaks can be combated using AI by modelling the spread of diseases using geographic and travel data.
There is also hope that utilising big data can encourage better practices in policy making when it comes to implementing the Sustainable Development Goals (SDGs) by freeing decision-making processes from bias and creating a system that is more progressive and equitable. And in times when aid budgets are being slashed in the United States – and across Europe – accurate and evidence based policy recommendations are all the more vital for the development sector.
Given all the above examples, AI and machine learning can be easily taken as a quick fix for a wide range of issues in the developing world. But it shouldn’t be assumed that ‘quick fixes’ exist at all. The history of international development is full of bubbles of excitement over potential silver bullets to combat poverty. It has been shown to be linked to the phenomenon of ‘isomorphic mimicry’ – where institutions have simply been replicated in the hope that any success is carried along with it. An example of this would be the persistent challenge of corruption identified in the Mauritius Police Force, one that is modelled after forces in England and Wales. Cases like this are suitable microcosms depicting a ‘one-size-fits-all’ approach that have gripped the development sector at the highest level.
Policymakers must also consider the potential risks that have been identified with incorporating machine learning in decision-making processes. While the industry is far from a mass-deployment of AI systems, the IMF has warned of the need for ‘scrutiny and inspection’ when it comes to computer driven decision-making. Implicit human bias is not embedded in these systems. And if responses are made fully autonomously as some have predicted, there lies the danger of unaccountability for when things go wrong. Continuing with this thought experiment – if a poor decision is made and this leads to disastrous outcomes, these can only be amplified if you consider the potential speed at which automated policy-making machine perform.
There is very little evidence to suggest that the development industry is likely to fully automate poverty reduction, which may alleviate the worst of fears some might hold. Conversely, the sheer amount of data in existence that could help guide development strategy can be seen as an untapped resource and a wasted opportunity. It is certainly the case that initiatives involving machine learning and AI might offer innovative and exciting prospects for future policymakers. This being said, it is important to remember that this sector is just one of many patiently observing the evolution of these technologies.