Reducing Poverty with
Machine Intelligence
Machine Intelligence (MI) can be a game changer in the fight against poverty. Several projects and initiatives have already shown how MI technologies can help reduce the impact and spread of poverty.
According to the UN [1], there are around 736 million people in the world living in extreme poverty (at or below $1.90 a day), a majority of them in Southern Asia and sub-Saharan Africa. Besides having an extremely poor income and a low standard of living, these people often suffer from hunger and malnutrition and have limited access to education and healthcare. They suffer discrimination and are often excluded from the socio-political decision-making process. On the other hand, relative poverty – defined as having an income that is considerably lower than the average – is also present in developed countries, where it manifests itself through stigmatization, insecurity and social exclusion. Although considerable progress in reducing poverty has been made over the past decades, poverty is still a significant and challenging issue today. Thus, it is no wonder that reducing poverty remains the number one item on the list of UN’s Sustainable Development Goals [2].
UN infographic on poverty [1].
MI technologies can support nonprofit/non-governmental organizations (NPOs/NGOs), individual researchers, and government agencies in fighting poverty by optimizing existing initiatives and by creating completely new ways of mastering this complex challenge. Below, we describe examples of how MI is currently being used in the fight against poverty. We hope that they will inspire the reader to consider the problem from a different angle, to appreciate the potential of MI technologies to serve the Common Good and spawn ideas for alternative approaches to poverty reduction.
Identifying poverty
​
If we want to reduce poverty, the first thing we have to do is to identify the geographical areas in which it occurs and which of these areas require help most urgently. Census data and household surveys can be used for this purpose. However, such information is often not available for many developing countries. Furthermore, many governments manipulate their data for political purposes, so that available data might be not very reliable. Filling this gap by gathering data through volunteers is very slow, often difficult and can be very costly.
To address this challenge adequately, one needs to think unconventionally. And this is exactly what the folks from Stanford’s Predicting Poverty Project have done [3]: they have combined Machine Learning (ML) with high-resolution satellite images to provide reliable estimations of poverty and wealth in several African countries (Malawi, Nigeria, Rwanda, Tanzania and Uganda). In doing so they have used a simple but brilliant trick: first, images of the Earth at night are evaluated with regard to their brightness. Brighter regions tend to be more developed, because brightness indicates a higher availability of electricity, so that this criterion can be used as an estimator of economic development level of a region. This information is then used to label high-resolution daytime images, i.e., to divide these images into clusters of “poor” and “wealthy” regions. In the next step, an ML algorithm learns to map these daytime images to their corresponding development levels without any human help – the ML engine learns to interpret certain image features, e.g., the presence of roads, urban areas, water sources or farmland, as indicators of economic wealth. In the final step, the MI model created by the ML algorithm is used to estimate the poverty/wealth levels in new regions (regions which were not used during the machine’s learning process). In this way Stanford researchers are able to predict the distribution of poverty very precisely, even more precisely than a conventional approach! The MI approach is very cheap and scalable and can be used to evaluate poverty-stricken areas all around the world. Powered with this information, NPOs and governments can act more efficiently and effectively in their fight against poverty.
Researchers at the Qatar Computing Research Institute (QCRI) of Hamad Bin Khalifa University have shown that the above approach can be enhanced by including statistical data from social media [4]. Facebook, and other social media companies for that matter, operate by enabling advertisers to selectively target users whose profiles match certain criteria. Facebook’s tool “Audience Insights”, for example, allows access to statistics about groups of users, including information such as age and gender distribution, but also attributes like education, profession, the countries the users have lived in and what kinds of devices users have used to access Facebook [5]. These statistical analyses can even be done on a city district level and can serve as a wealth/poverty estimator. For example, if in some region there are a lot of users who access Facebook via an iPhone (rather than some other device), this region is very likely to be quite wealthy. Combining this data with satellite imagery can provide very granular statistics about the economic situation of a given country, city or even district. In collaboration with UNICEF, and using this approach, QCRI has already produced poverty maps for the Philippines and India.
Once regions affected by poverty have been identified, the next step is to define the most effective actions to combat it. Agricultural development and better access to education are often cited as the two most promising areas of action, and there are so many ways in which MI can foster agricultural development that we dedicate a separate article to this topic [6]. In this article we will just consider education as a means to this end.
​
​
Improving access to education
A UNESCO study from 2017 showed that if all adults in the world were to complete a secondary education, ca. 420 million would escape poverty [7]. That is more than 50% of the extremely poor population worldwide! In developed countries too, wealth is strongly correlated with education. So, better access to educational resources can help children from poor families to break out of the “poverty trap”. Thus, providing education to those in need is one of the most powerful ways to fight poverty around the world, and a variety of MI technologies can support NPOs/NGOs dedicated to this goal.
Standard teaching tools can be enhanced by MI and open new ways to improve education access. An example is the PowerPoint subtitling feature, which allows a teacher to transcribe her words while she is teaching and to display them as subtitles [8]. The text can appear in the same language or can be translated into another. This feature can be used, for example, to enable German-speaking teachers to give online tutoring to children in Ethiopia. Such an approach can be used to address the issue of the lack of teachers in developing countries.
Advanced analytical techniques can help educational institutions gain deeper insights into the needs of their students, their performance and struggles, and the effectiveness of the institutions’ programs. One example of such an institution is Eneza Education, a company which provides pupils in Kenya, Ghana, and Ivory Coast with primary and secondary school learning material accessible on any device [9]. Eneza cooperated with US-based social startup Delta Analytics to quantify and better tune its impact by analyzing the data of over 350,000 users of Eneza's mobile platform [10]. Similar analyses can be performed for any educational institution which has data about its students. It can lead to a better understanding of students, their behavior and the effectiveness of programs, so that the institution can make better decisions on how it should access and teach students.
One rather organic enhancement to the above idea is that of smart online tutoring systems such as the one offered by Carnegie Learning [11]. This system helps teachers and tutors to find a teaching style fitted to the needs of each student. Backed by big data and advanced analysis tools, the platform provides teachers with real-time feedback on students’ performance, strengths and weaknesses so that they can adapt their way of teaching to better serve individual students’ needs.
Poverty in developed countries
Extreme poverty in developing countries is a very important and urgent topic. But relative poverty in the developed world should not be underestimated either. For example, in 2019, 15.9% of the German population was at risk of poverty [12] and in the US more than 550,000 people were homeless [13].
To tackle such a problem several MI approaches have been developed. For example, Portuguese researchers developed an app which predicts risk factors for over-indebtedness [14]. This app can be used by authorities and NPOs to identify households which could slide below poverty line soon and need urgent help.
The city of London, Ontario in Canada is currently facing a big challenge with regard to rising homelessness. As one measure against it, the city has developed an Artificial Intelligence tool called Chronic Homelessness Artificial Intelligence (CHAI), which predicts the likelihood of a shelter user becoming chronically homeless [15]. Since this application relies on very sensitive data, developers have paid much attention to data privacy, anonymization, and interpretability of AI output. Also, the participation in program is not mandatory.
As a final example, the UK chapter of DataKind collaborated with a food bank in Huddersfield to create a Machine Learning system to help its staff identify clients who are at risk of becoming chronically dependent on the food bank’s services [16]. These clients get a holistic consultancy from social workers and help in addressing the underlying problems. The Machine Learning system supports workers in prioritizing cases so that scarce human resources can be applied in the most effective way in order to help those at the greatest risk first.
Typical process in Huddersfield's food bank to enable early intervention for people likely to be in need of additional support. Above, before and below, after introduction of Machine Learning solution [16].
Conclusion
Machine Intelligence technologies can be a game changer in the fight against both extreme and relative poverty. We have mentioned here only a few examples of intelligent machines helping NPOs, companies and authorities to battle poverty. Indeed, there are quite a number of such initiatives and even more potential applications waiting to be developed. Thus, think creatively and use the power of modern technologies in your organization [17].
References
[1] https://www.un.org/en/global-issues/ending-poverty.
[2] https://www.un.org/sustainabledevelopment/sustainable-development-goals/.
[3] http://sustain.stanford.edu/predicting-poverty/.
[5] https://www.facebook.com/business/insights/tools/audience-insights.
[6] https://www.mi4people.org/hunger.
[9] https://enezaeducation.com/.
[10] http://www.deltanalytics.org/past-grant-recipients.html.
[11] https://www.carnegielearning.com/.
[14] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7571461/.
[15] https://news.trust.org/item/20201015080726-yer4o.
[16] https://www.datakind.org/projects/identifying-food-bank-dependency-early.
[17] We, at MI4People, are eager to collaborate with you to help you understand the potential of MI and identify impactful research projects that we can undertake to find MI based solutions that help deliver Public Good better as well as assist you in meeting your organization's goals more effectively.