Will Artificial Intelligence End Global Poverty?

In short, no—at least when it comes to lower-income countries and populations, and not in time for the 2030 Sustainable Development Goals (SDGs). Even worse, it can be a major distraction from more urgent, foundational solutions critical to achieving the SDGs.

It’s 2019. So, it’s hard to get through a week without some news story about malevolent social media ‘bots, about how we are all going to live in a surveillance society, or generally about the imminent AI invasion which will take over all of humanity. On the rosier end of the spectrum, are the stories about the next new convenience AI will bring. There’s also the occasional story about how AI will end global poverty.

Over the past couple of years, the hype surrounding “AI for good”—among technology startups, philanthropic organizations, universities, and even governments and international institutions—has reached fever pitch. Most of these conversations are highly abstract, driven by a technology-first lens which fails to appreciate what types of interventions are foundational and urgent, and specifically how data analytics and AI can help.

Consider this example of an AI-driven solution highlighted at a recent UN summit on AI for development: “helping smallholder farmers in Sub-Saharan Africa improve crop yields with less water”. Of course, only good can come out of improving the efficiency of water use, if farmers are facing physical water scarcity. The problem with such a solution is two-fold. First, there are tried-and-true methods of improving water efficiency, such as drip irrigation. The challenge most farmers face is that they cannot afford these on-farm implements. Second, most smallholder farmers in Sub-Saharan Africa face economic water scarcity, i.e., there is often plenty of shallow groundwater in their general vicinity, but the farmers cannot afford the means of accessing that water through drills and pumps. Hence, they practice rainfed agriculture.

While there is no doubt that AI algorithms could improve water efficiency in modern irrigation systems, the problem that farmers in Sub-Saharan Africa face is more fundamental: limited resources to access water, let alone the equipment to use it efficiently. Similarly, using AI to develop accurate risk models for crop insurance will have little impact without the cash reserves to provide insurance in the first place, and without adequate systems for distributing insurance policies and processing claims.

Not surprisingly, the value that can be added by data analytics and AI in global development is highly context dependent. To unpack that context, our team at ITT has released a report that parses those nuances. In the report, we point out that:

• Data-driven analytics can only add secondary value, layered on top of and improving primary interventions. The most critical interventions required to achieve the SDGs—in core developmental areas like food security, health, energy access, and education—require primary solutions such as access to irrigation, affordable medical devices for clinics, affordable energy-efficient appliances for productive uses, well-trained staff for clinics and schools, and digital tools for assisting those personnel. Data analytics and AI can help optimize decision-making for those primary interventions (e.g., via accurate medical diagnosis); but in the absence of such primary interventions, improved decision-making can add limited value.
• For most problems relevant to the SDGs, there isn’t enough data for AI. “Big data” representative of the complexity of the underlying problem is essential for AI algorithms to learn well enough to predict. Most low-income countries lack the ICT infrastructure to collect enough data to utilize such algorithms. Feeding such algorithms data that is not adequately representative can lead to incorrect conclusions. One good example of a strong data infrastructure is India’s Aadhar digital ID and India Stack systems which have laid the groundwork to gather substantial data over time. Mobile money systems like Kenya’s mPesa are a good start, but the narrow nature of the data collected (i.e., financial transactions) and its private control are limiting factors.
• A lot of value can be derived from conventional analytics, rather than new-generation AI. The concept of AI is continuously evolving; what was considered AI a couple of decades ago is now considered “conventional” analytics and has been in reliable use in many different sectors and applications. Even as data infrastructures get developed over time, much can be done using conventional analytics, rather than new-generation AI which will likely be an overkill. See the exhibit above—of the 38 most important interventions required to address food security, healthcare, energy access, and education in developing countries within the SDG timeline, 19 do not require much in the way of sophisticated data analytics; 14 interventions can be significantly improved using conventional analytics; and only 5 require AI and big data.

Organizations focused on data and AI for the SDGs should, therefore, invest first in appropriate foundational systems and data infrastructures. Otherwise their impact will be as artificial as the intelligence.

Download the full ITT report.