By Susan Holliday1
1Senior Advisor, IFC
There is a lot of emphasis right now on the impact of Artificial Intelligence (AI) on different sectors, especially financial services, and on jobs. At the World Bank and IFC, the focus is on partnering with governments and the private sector to end poverty and increase shared prosperity. This topic is becoming increasingly mainstream, as demonstrated by the United Nations Programme Development blog of 21 January 2019.1 Over 160 actual or potential use cases of AI for the “non-commercial benefit of society” have been identified. Most of these applications are commercial, showing the importance of the role that the private sector has to play, even if in some cases the customer is local or central government. This piece discusses some other examples relating to key factors in prosperity: natural catastrophe, capital markets and diversity and inclusion.
The insured losses from natural catastrophes in 2018 amounted to $76 billion but the economic loss was more than double,2 meaning there is still a massive protection gap. For global catastrophes this has been estimated by Sigma at $280 billion in 2017 and 2018 alone. They also note the rising impact of secondary and secondary effect perils. There are several examples of where AI is being used and we expect these tools to be developed further over time and adopted more widely.
One example is prioritization of rescue efforts. AI can help identify which streets are impassable due to flooding and focus rescue crews in boats to those areas. More interestingly, AI can be used based on a wide variety of data sources to predict where the flooding may hit next, helping with decisions about whom to evacuate and to where. In the case of wildfires, these are often started or made worse by unpredictable human behaviour.
AI, combined with some relatively simple Internet of Things (IoT) devices (such as sensors), could pick up early signs of new fires and also help in decision-making about the best way to fight the fires. Updated with real-time information, this becomes a powerful tool in risk mitigation and saving lives. Of course, there is a cost involved here, but in this case the AI is not particularly complex and there could be a wide array of clients from local governments to utility companies and private businesses.
The World Bank recently published a report, “Data Driven Development Response to Displacement Crisis in Uganda”.3 The proposition is that AI could be used to identify stresses in the system such as higher demand for water, with a pre-agreed mechanism in place to mobilize resources to the benefit not only of the displaced people but also the local population. This example focused on people displaced by war and conflict, but the concept could also work well in the case of populations displaced by natural disasters.
Many countries lack broad and deep capital markets, and this is becoming more of an issue as governments try to encourage long-term saving and develop private pension schemes. One of the most important issues with capital markets is trust. People tend to leave their money in cash when trust in banks is low, and similarly leave their money in the bank if they are nervous about investing.
A number of high-profile fraud cases, such as the Bernie Madoff scandal, highlight the possible risks here. AI is now being used in a number of ways by regulators to combat this. A recently adopted use is to try to detect mutual funds, financial advisers or investment firms that are either promising or reporting returns that seem too good to be true. Instead of relying on consumers to report their suspicions, regulators in the US are using AI to scan advertising and promotional material and social media to pick up on suspicious-looking claims and then investigate them further.
This is a good use case for machine learning as there is a relatively high degree of context and judgement involved. Currently, humans are investigating suspicious-looking entities, but it is likely that that over time AI and machine learning can take over a lot of this activity and also help prevent fraud.
In developing markets, where people are less familiar with investing and where confidence in financial services is low, having robust systems in place to assist often overstretched regulators is even more important, otherwise there is a real risk that innovation will be stifled.
One particularly interesting and wide-reaching focus in AI is natural language processing (NLP). This has the potential to improve interactions and customer service in a lot of areas, including financial services, travel and health. The impact can be even greater when dealing with diverse populations. Currently many bots struggle to deal with different accents or regional words or dialect. Imagine how much worse it must be for displaced persons or those who speak a language which is not widely known. There are two aspects to this issue: making AI truly conversational and incorporating more languages; there is a long way to go on both aspects.
Most of the well-known voice-based solutions, such as Alexa and Siri, operate in a way similar to digital apps. However, people do not speak in the same structure as when they use apps or search engines such as Google. A number of companies are addressing this issue using alternative approaches.
Although many of the early use cases are in areas such as digital assistants and connected homes, developing good conversational systems has important developmental use cases for people in remote areas, refugees, managing natural disaster risks and providing advice about areas like health and financial literacy. The challenge is that connectivity to the cloud is needed for it to work, which may be a challenge where Wi-Fi or a strong mobile phone signal is not available.
The other major issue is language. Most solutions today are using Google Translate, which has improved over the years. However, it is still only effective in a few languages and cannot really deal effectively with dialects although it is possible to build capabilities in any language as long as enough data is available.
The challenge from a developmental standpoint is that commercial uses are most likely going to focus on the languages spoken in wealthy countries. This could be a case where NGOs and donors may need to be involved to incentivize a more diverse and inclusive approach to language capability development. This could have significant benefits for helping minority groups and refugees and reduce pressure on central and local government resources that are currently trying to support them.
These examples suggest that there are many plausible use cases for AI in the development space. In order to make these a reality, two preconditions need to be met. The first is on the infrastructure side with the need for reliable Wi-Fi and mobile phone networks to connect devices. The second is funding. While some private sector solutions are already available to address some of these challenges, it is likely that governments, NGOs and donors will have to provide funding for implementation – at least initially – because the somewhat niche nature of most of these projects means that they are unlikely to be financed by large corporates. The good news is that as AI solutions get built out, less work will be needed to tailor them to development needs, facilitating future adoption and deployment.