Artificial intelligence. Often associated with futuristic films and robotics, this expanding field of technology has gained pace as a huge growth market in the private sector. The international development sector is also showing an increasing interest, with enthusiasts pointing to the potential of artificial intelligence (AI) in contributing to the Global Goals. As AI has the potential to both solve and exacerbate major challenges facing our world, it is imperative that even the least technical among us understand the principles and potential pitfalls of this powerful technology.
Though AI can be used by robots and chat-bots (and forms the basis for Siri, Alexa and such), in reality AI can be most simply described as intelligence demonstrated by a machine: a computer coded programme that can perceive and perform tasks associated with human intelligence, such as decision making, translation, and identifying patterns. The automation of these kinds of tasks can enable programmes to perform functions on a grander scale, at a greater pace, and therefore at a lower cost, than humans.
Machine learning (ML), though sometimes used interchangeably with AI, is a sub-set of artificial intelligence which enables an application to ‘learn’ as it works, adapting and changing to become ‘better’ at its task. For example, an AI programme that applies machine learning will not only automate the analysis of multiple data sets, but will refine its pattern-finding over time, identifying small pattern shifts that may be imperceptible to humans. This will happen up to a point where it can predict future changes, outcomes or desires based on these pattern shifts, and will continue to refine its output the more it ‘learns’, making it better and better at its task.
Taking all of this into account, it’s easy to see why AI is gaining traction as an answer to some of the world’s most pressing and complex problems. The AI for Good Global Summit, for example, seeks to identify ways to utilise AI to accelerate progress towards the Global Goals, appreciating that AI is the most powerful technology at our disposal to analyse the unprecedented levels of data collected across the global targets, and to utilise this analysis for global impact. As a result of its potential, AI is now being deployed across a raft of development challenges, in a number of innovative ways.
Global health: In the field of global health, and in the wake of the COVID-19 pandemic, AI has been recognised at the forefront of medical innovations. In 2019, the then Department of International Development (DFID) was interested in using AI to predict and model epidemics.[i] The use of AI during the COVID-19 pandemic has been significant; in the UK for example, it is being used to improve diagnoses by analysing and identifying patterns in medical images of patients with COVID-19, in order to predict disease by identifying abnormalities at great speed and with more precision.[ii]
Big data for the planet: In the field of environmental protection and climate change mitigation, with a natural accumulation of big data[iii] sets, AI and ML is being applied in a myriad of ways. A recent digital summit on AI for the Planet highlighted a variety of applications for AI across climate change, conservation and waste management, and to tackle the challenge of meeting Global Goal targets.
Collaboration for conservation: Combined efforts, such as those between citizen scientists who collect vast data sets like wildlife imagery, and AI applications which enable the identification and analysis of these, are gaining recognition as a solution to the challenge of tracking and protecting our endangered wildlife[iv]. These human-computer partnerships reveal huge potential for making a significant impact, especially where human resources required to process the data would be prohibitive, and we simply don’t have the time to lose.
Education: AI may also contribute to the core development priority of education. For example, chatbots using machine learning can help educate children, improving their learning by personalising responses and revision techniques[v], and do so without human resource or classrooms. However, this example highlights just one potential pitfall of utilising AI in development; a digital divide that already exists in many communities could lead to further marginalisation of those without internet or technology access.
Data analysis, modelling and mining: AI can be utilised for less user-focused tasks, such as data analysis, predictive modelling and activity evaluation. AI and ML applied in data mining tasks, particularly where data is unspecific or qualitative, can speed up the evaluation process significantly. For example, programmes which involve ‘women in peace building’ but might not be a tagged as such, could be identified using AI tools. Programmes can be built allowing human evaluators to ‘supervise’[vi] the algorithm used and check errors, refining the programme’s output and allowing them to concentrate on the findings and patterns identified, without having to check thousands of documents or other data sources first. Data such as financial records, which do not require nuanced human input, can be analysed with ‘unsupervised’ algorithms.
As the vast potential of deploying AI in development is just coming to light, challenges in using this technology have also become evident.
Garbage in = garbage out: As with all analysis, modelling and evaluation, findings are only as good as the data on which they are based. Predictions will naturally exclude any groups un- or under-represented in the data, and thus AI models will replicate data biases, potentially reinforcing existing marginalisation and inequality. This is especially relevant in the case of human-based data, such as on health, migration, sentiment and behaviour. Patterns which exist in the real world, whether through bias, unbalanced power relations or simply a lack of data due to difficulties in collection, can be replicated and exacerbated by programmes which analyse and predict future patterns.
Ethics: AI created without appropriate ethical considerations can exacerbate inequality, ignore particular groups, and use data without accountability, thereby weakening public trust in organisations using these tools. In order to create ‘ethical’ AI, academics and international development experts continue to establish and refine principles and guidance on which to base new applications and programmes. The first of these principles should always be applicability; ‘innovation’ for innovation’s sake is unlikely to yield good results, and so a decision to use the latest machine learning technology should be based on whether it will truly enable better evaluation or predictions. Helpfully, USAID has recently released a Guide on AI for development practitioners, with a focus on using AI responsibly, whether a technical expert on the topic or not.
Diversity: Another key challenge that must be met is ensuring representation; this should be both within the data used and in the creation of the algorithm itself. Technology with human intelligence is still created by humans, and therefore prone to human prejudice. If we wish to solve the world’s most complex problems, then the technology we use must represent the world that is facing these problems. We need to reflect the diversity and variety of human beings when creating AI programmes, to promote fairness and accountability in their application.
Equal opportunities: The loss of jobs to AI is often mentioned as a criticism of its expansion. Yet AI and ML specialists, data collectors and AI ethics experts will become more and more necessary as this field develops. The more we can ensure that these roles are understood and jobs undertaken by a diverse range of people, the better the technology will become.
The potential for AI in international development is immense, and current capabilities are just a fraction of what might be possible in the future. But this also comes with risks. The need to include individuals and communities from around the world in the design, creation and deployment of AI will be vital to ensuring that existing challenges are tackled, not exacerbated, with the most powerful tools we now have.
Further reading on AI in development:
- World Economic Forum Strategic Intelligence – AI Topic Publications
- Data Stewards Network – Selected Readings on AI for Development
- USAID Report – Reflecting the Past, Shaping the Future: Making AI Work for International Development
- USAID Report – Managing Machine Learning Projects in International Development: A Practical Guide
- IDIA Discussion Paper on Artificial Intelligence & Development
- JPIA Article – Artificial Intelligence in International Development: Avoiding Ethical Pitfalls
[iii] ‘Big data’ refers to extremely large and complex data sets, often those which are continuously growing and cannot be processed by traditional processing software.
[vi] ‘Supervised’ and ‘unsupervised’ are two ways AI can ‘learn’. In supervised learning, a set of training data is specified by humans so that the machine can replicate predictions. In unsupervised, the machine identifies patterns independently. See page 9 of this resource for more details: https://static1.squarespace.com/static/5b156e3bf2e6b10bb0788609/t/5e1f0a37e723f0468c1a77c8/1579092542334/AI+and+international+Development_FNL.pdf