Artificial Intelligence for the Developing World

by: Ben Goertzel

The commercial value of artificial intelligence technology is now increasingly obvious across the board, with large companies in multiple sectors investing billions upon billions. But the importance of AI goes well beyond its direct financial value; there is a fundamental transformative potential here, which cuts at the core of human society, human life and human values.

Major governments and corporations around the world, alongside academic scientists and philosophers, are beginning to ask questions like: Will AIs eventually be more generally intelligent than humans? Will AIs eventually be conscious in the same sense as humans?   Will AIs and robots eventually take over all, or nearly all, human jobs? And will this “eventually” perhaps be, not centuries but only decades ahead?

And if we do have powerful, intelligent, self-aware AIs taking on most or all of the tasks now carried out by humans– what values will these AIs use to guide their actions? Will these values be human values, or something different? And if human values– what  variety of human values?

High-profile conferences have been convened to address these issues, e.g. the Asilomar AI conference held in California in early 2017 (organized by leaders of the US and UK tech and academic communities), and the UN conference on Beneficial AI, to be held in Europe in mid-2017. All this attention has not provided any definitive answers to these complex and thorny matters, but it has sharpened the issues involved and brought a wider variety of voices into the discussion.

One question that is not asked often enough, however is: What are the implications of AI, robotics and other advanced technologies for the developing world? Will they serve more strongly to exacerbate global wealth and income inequality, or to remedy it? How might these technologies help the world’s neediest? And what unique contributions might the skills, insights and values of people in the developing world make to the growth of these technologies and their impact on humanity as a whole?

Perhaps the best way to approach these big questions is to dive a bit into the nitty-gritty, and look at the different areas in which AI can provide value to the developing world right now. This will naturally lead into a discussion of how the unique social and cultural characteristics of the developing world may provide tremendous value to advanced artificial intelligence as it emerges and absorbs the spectrum of human values.


One of the great strengths of the developing world today is its youth.  With such a large percentage of their population consisting of young people, developing nations are especially sensitive to the quality of education. If provided with excellent and inspiring education today, the children of developing nations will do amazing things tomorrow.

But there are serious issues here. In many developing nations, well-educated teachers are in short supply. Internet connectivity and mobile computing devices like smartphones and tablets are decreasing in cost and increasing in prevalence; but without appropriate early education and acculturation, youth in rural areas of developing nations are unlikely to spend their time going through the online curricula made available by Coursera, Udacity, MIT and the like, or reading the math, physics and computer science papers available freely online at, or contributing to the open-source software projects available on GitHub.

Computer-based (including tablet and phone based) curricula and educational software can make a huge difference here, and have capability to surmount these issues.   But to be truly effective for most children, education needs to be progressive and personalized.  And this is one place AI technology can play a major role in education: assessing an individual child’s strengths, weaknesses and needs, and customizing education for that child alone.

This sort of personalized education can take many forms.  It can exist behind the scenes of educational software, affecting the information provided to the child in an adaptive way. Or it can be right out front and explicit, as in the Yanetu AI Teaching Tablet project being prototyped by iCog Labs, in which an AI avatar interacts with students and customizes the student’s interaction with educational material, in a mixture of standard educational and gamified content.

Linguistic Tools for World Languages

For educational applications alongside many others, it is important for the advancement of the developing world that their native languages become fully infused into the computer age. This means the creation of a complete battery of computational speech and language tools for developing-world languages: speech to text, text to speech, syntax parsing, semantic mapping, language generation, etc.

This set of tasks presents interesting and important research challenges for computational linguistics, with opportunities for deep technological leapfrogging.  Much of the computational linguistics infrastructure for developed-world languages is based on a species of AI technology called “supervised learning from annotated corpora”, which requires human linguists to provide detailed linguistic information about a certain “training corpus” of documents, and then uses AI to generalize this information to other documents. However, many researchers feel this technology is reaching its limits, and attention in the research community is shifting to “unsupervised learning” approaches, that infer linguistic information via pattern recognition from natural bodies of speech and text without the need for annotation of a training corpus by expert linguists. An opportunity exists here for developing nations to spearhead work in unsupervised computational linguistics, thus creating scientific advances simultaneously with extending the reach of automated language processing to the bulk of the world’s languages.

In practical terms, the result of grasping this opportunity would be, for instance, the ability for software developers to create mobile phone applications that the ordinary person could talk to, and that would understand what was said to them and reply appropriately. Imagine the implications for education, for medicine, for government and the economy. This is a chance to use advanced research  to make the world’s technology available to the totality of the world’s population, including those with limited or no literacy. And it is research that does not require any expensive equipment, only commodity computers.

Medical Support for Patients and Professionals

Medicine is an area where AI has already proven its value, for instance via IBM Watson which provides medical diagnosis at a level above that of the average human doctor, and has been purchased by numerous hospitals around the world.   One of the next steps that needs to be taken, however, is to bring this sort of AI medical wizardry to all the people of the world, not just the clientele of forward-thinking and well-funded hospitals. IBM is moving in this direction, for instance with three Watsons in sub-Saharan Africa (in Kenya, Nigeria and South Africa); but much more progress is needed.

AI medical advice and guidance should be available to every medical professional and every private citizen via smartphone apps and via voice-based AI dialogue systems. The technology exists to do this today – with the exception of computational language processing tools for the bulk of the developing world’s languages, which is a quite tractable R&D problem as outlined above. What is needed is merely the will to make it happen.

Precision Medicine

Alongside distributing the benefits of modern medical knowledge more widely, there is also an acute need for advancement of the state of this knowledge. The pharmaceutical industry is increasingly realizing that its classical “one size fits all” approach is inadequate for coping with the diversity of human bodies and situations. The new buzzword is “precision medicine” – the customization of medical treatments to the individual, based on clinical and genomic and life-situation knowledge about the individual’s situation.

Artificial intelligence is critical for precision medicine, because the patterns that make a certain therapy work for one person and not another can be quite complex, combining multiple different factors.  Finding combinational patterns in complex data is one of the classical strengths of AI technology.

The nature of precision medicine is that it depends sensitively on the individual patient’s situation; and for this reason, conclusions drawn from analyzing data regarding patients in the developed world cannot be expected to apply immediately and wholly to patients in the developing world. Rather, comprehensive data regarding healthy and unhealthy individuals across the developing world, experiencing various medical therapies and life situations, will need to be gathered, and then analyzed using AI tools. This data will yield all manner of tremendous insights, in precision medicine and beyond.

Using AI to Help Leverage Native Medical Knowledge

Modern Western medicine has demonstrated unparalleled strength at diagnosing, palliating and curing a variety of acute diseases; for chronic illnesses and promotion of general health, however, the world’s various traditional medicines often have more to offer.  The study of traditional Chinese medicine has borne impressive fruit, including the discovery of new pharmaceuticals; and is the subject of active study regarding its integrative, systemic and personalized aspects. Novel nutraceuticals have been discovered, via applying AI technology to learn new ways of combining traditional Chinese herbs so as to address combinations of genes identified by AI tools as corresponding with particular medical conditions.

What has been done with traditional Chinese medicines, can also be done with traditional African medicines. There is already work isolating active ingredients of traditional African herbs, and exploring their value as pharmaceuticals. However, no one has yet explored the application of AI to discover novel combinations of these herbs and their ingredients, and the use of these combinations in a personalized way in the vein of “precision medicine.” This is a case where artificial intelligence technology may synergize with age-old African medical wisdom to yield new and medically beneficial discoveries.

Detecting Agricultural Disease

An immediate and relatively straightforward application of current AI technology to agriculture is the identification of agricultural disease, and the prediction of the advent of agricultural disease before it reaches a damaging stage. This can be carried out using computer vision technologies similar to those used by Facebook, Baidu and Google and other companies to identify human faces. Via gathering photographs of plants at various stages of infection by various diseases, and supplying these photographs to existing AI systems capable of “learning by example” from images, one can create a system that any farmer can use.

From the farmer’s view, the process will be very simple: Simply photograph a plant or a leaf with a smartphone, upload the photo to the AI via an app, and the AI will tell you if the plant is sick and if so in what way, or the probability that the plant will get sick. Further, the dataset composed of the images uploaded by numerous farmers, will constitute a valuable and ever-growing repository of agricultural knowledge, and will enable the AI to become ongoingly smarter and more knowledgeable.

AI-Guided Animal and Plant Breeding

Pesticides and genetically-modified organisms currently are immature technologies posing significant health problems. However, the problems that they were created to solve remain acute in many places. Artificial intelligence combined with genomics and traditional plant and animal breeding provide the potential of a creative alternative solution.

If one has a diverse population of organisms of a certain species, one can then study the members of the population genomically, and use AI to determine the combinations of genomic variations underlying desired phenotypic characters. This information can then be used to guide breeding; in other words, genomic information about the population can be used to accelerate breeding beyond the rate that would be possible using only observation of phenotypic information.

For instance, if one wants to create avocados that taste sweeter but are less appealing to insects, one can gather the genomes of a variety of avocadoes, and use AI to learn the patterns of genomic variation that characterize sweet avocadoes, and that characterize avocadoes insects do not like so much. One can then determine the pattern of genomic variation needed to create an avocado that is sweet but not appealing to insects.  Knowing this, one can choose parent avocadoes based on their DNA profiles, with a high probability of creating children matching the desired genomic profile.

This “AI-guided breeding” approach will be faster than traditional breeding but slower than genetic engineering, but should product results that are more robust than those emerging from current GMO techniques, and more likely to be conducive to health for human consumers and the ecosystem.

Automated Agriculture

Today a substantial portion of citizens in the developing world work as subsistence farmers– an occupation that, for all its disadvantages, does provide some insulation against the ups and downs of economic changes. As robotics advances, however, eventually it will simply become too inefficient for society to have a large percentage of arable land farmed by humans and domestic animals. At some point, agricultural robots will become sufficiently inexpensive that developing-world governments will consider it desirable to use them to create food from the land. If there are not sufficient non-agricultural jobs for the farmers just displaced, the wealth created by the robotic farming equipment and other automated technology will need to be used to provide sustenance, education and healthcare to these individuals.

Alternative Power, Smart Power

Cheap, widely available electrical power is the first great enabler of all the modern technological advances – including of the second great enabler, which is cheap, widely available Internet access. To improve their power infrastructure and ensure its appropriate growth, developing nations must jump enthusiastically on board two trends: decentralized alternative energy generation, and AI-guided “smart power grid” regulation.

Fossil fuel based energy generation still has its place and is being made more efficient by various innovations; but cleaner, more advanced techniques such as solar and wind energy generation are now finally coming into their own. The steady improvement of solar technology in particular has been striking, during the last two decades. It is now viable for farmers in rural regions of developing nations to generate their own power using relatively inexpensive solar equipment, and sell excess power they don’t need to the overall power grid.

To maximize efficiency of power allocation around a city or a nation, and to minimize issues such as power outages or brownouts, the best known approach is to instrument the machinery underlying a power grid with Internet-connected sensors, and turn the power grid into effectively a giant, sprawling, AI-powered brain.  In this way one can proactively route power where it will be needed most, and one can predict hardware issues such as power transformer failures in advance, so that weak components can be repaired before they break and cause issues. These technologies have been deployed already, in parts of Korea and Europe for example, but are not widely used in the developing world, where arguably they are needed most. Here as in many other domains, no radical technology innovations are needed, and the cost is not high compared to various initiatives commonly undertaken in the developing world (e.g. highways, mines and dams); what is needed is merely to find the will and organization to make it happen. There seems little doubt that this will indeed occur at some point during the next 1-3 decades; but sooner will be better than later.

The Future of Additive Manufacturing

The global manufacturing sector is in the early stages of a transition away from traditional manufacturing processes toward additive manufacturing – “3D printing” and associated technologies.   The cost of 3D printers that can print metal object is rapidly decreasing. Currently these additive manufacturing devices are still too slow to replace traditional factories, but during the next 5-20 years this will surely change, in large part due to advances in AI-driven 3D-printer control systems.

As AI-controlled additive manufacturing becomes a viable and cost-effective alternative to traditional manufacturing, and eventually supplants traditional manufacturing, this will present fascinating and unprecedented opportunities for developing nations that have lagged in the creation of traditional manufacturing infrastructure. Just as mobile Internet enabled many developing nations to leapfrog past the stage of laying down “land line” data cables, similarly high-speed, low-cost additive manufacturing will enable these nations to leapfrog past the stage of building old-fashioned factories.

Currently, in many cases, natural resources are exported from developing nations to developed nations in relatively raw form. These raw materials are then refined and turned into products in the developed world and sold globally, only a small minority of the profits returning to the developing nation from which the resources were extracted   However, as additive manufacturing develops further we will see the emergence of alternative approaches. Ores may emerge from the mines of developing nations and get refined locally into powders, which are then fed into additive-manufacturing plants, producing products locally for local and global distribution.

AGI Ethics and the Spirit of Mutual Aid

Nearly everyone agrees that, as AI becomes more and more capable and verges into Artificial General Intelligence (AGI), it will be desirable for AIs to manifest “human values.” For all its problematic complexity, and all its self-contradictions, the overall morass of human values does demarcate a relatively small region of the space of all possible value systems; an AI with an arbitrary value system would be unlikely to act in accordance with human taste, ethics and aesthetics. It might be that as AIs become more and more general intelligence, they also become wiser according to some universal standard of wisdom; but this sort of phenomenon is quite poorly understood at present, and the wisest course seems to be to ensure as best we can that the powerful AIs we create display and embody the better aspects of humanity’s values.

Guiding the value systems of AGIs that may become, ultimately, as intelligent or more intelligent than humans, is clearly going to be a significant challenge. Some have taken the magnitude of this challenge as a reason to avoid development of advanced AGIs, or to slow down the development while we seek to more deeply understand the ethical issues involved (Nick Bostrom’s book Superintelligence made an impassioned and moderately influential argument in this direction a few years ago). However, the large economic value of advanced AI has already become evident to major corporations and nations, and thus a halt or slowdown to the development of AI appears unlikely. Increased generality of artificial intelligence – the “G” in AGI – also clearly has tremendous economic value; AGIs with the ability to deal with unforeseen situations and think on their feet will deliver more business and human value in a variety of contexts.

Given that advanced AI– including AGI– is clearly coming, the most rational course is to seek to derive from it the widest and deepest benefits that we can, and to infuse it with as many of the better aspects of human values as we can. And the diversity of human values is worth keeping in mind here. The values of large, predatory multinational corporate entities reflect human values; as do the values of brutal dictators and violent criminals. The values of religious mystics, dedicated scientists and engineers, wild-eyed artists and writers, young children and devoted teachers and doctors, also are part of the picture. Each aspect of human society contributes something to the totality of human values; any AGI that interacts broadly with humanity is going to absorb all these aspects, but that doesn’t mean the aspects will all be equally weighted in the AGI’s mind.

Looking across all the cultures of the world, one finds a splendid diversity of values alongside the diversity of languages, artwork, architecture and social custom. But among this diversity one may still limn some general patterns; and one of these, which has been often observed, is the deep tendency toward mutual aid and a feeling of brotherhood/sisterhood that exists in the developing world. In the developing world, more often than in wealthier and further-developed nations, the solution to a human being’s problem involves recruiting help from other human beings, based on informal person-to-person interactions and not just rigid, precisely-defined monetary exchanges. In some ways, it seems the advance toward a modern economy sometimes brings with it a diminution of everyday humanity.

The human warmth and tendency toward informal mutual aid that one finds more commonly and intensely in developing nations, may in the end be a tremendous asset to the human race as it creates powerful Artificial General Intelligences. For the spirit of mutual aid and human warmth is surely chief among the human virtues with which we would like to see our AGI “mind children” infused

The relation between AI and the developing world is thus a subtle one, with multiple intertwined dimensions that will change as technology and society unfold. There are manifold ways in which AI can benefit the developing world, and some of these have been reviewed above. There are also numerous ways in which the developing world can and will contribute powerfully to the advancement of AI and associated technologies – including obvious ones such as the brainpower of young scientists emerging from developing-world universities and startup incubators, and less obvious ones such as the contribution of the developing-world spirit of mutual aid to the psyches of powerful AGIs as they emerge.

From the editors of

This Article was originally written for the bi annual magazine (iCog Makers) by Dr. Ben Goertzel and was published on hard copy in collaboration with the Federal Democratic Republic of Ethiopia’s Ministry of Science and Technology on 5, August, 2017. It is republished here with the permission of the iCog Makers magazine editor and the verbal consent of the Ministry. 

Social tagging: > > >

Leave a Reply

%d bloggers like this: