AI for Social Good

One of the most consequential challenges of our time is the disruptive application of AI for social good. AI for social good offers unprecedented opportunities across many domains to build inclusive, just, interconnected, and flourishing societies.

In Latin America and the Caribbean, a useful AI intervention could be using machine learning programs to identify students in the primary and secondary education systems who are at risk of not completing school, and those most exposed to adverse academic outcomes.

The aim would be to rank students by their likelihood of not completing any stage of their education, and those most exposed to adverse academic outcomes as an essential requirement for budgeting exercises. Such students may benefit from the scaffolding of a department of “institutional effectiveness” at each school.

A department of institutional effectiveness would be responsible for the systematic measurement and evaluation of the institution’s performance against its mission and goals. Such units may use data analytics and data-driven insights to guide continuous quality enhancement and strategic planning. Reports with recommendations can result in operational efficiency appraisals. Such projects can test the effectiveness of machine learning algorithms in education settings.

The aims would be to rank the students in an education district by their likelihood of not completing any stage of their education, and those most exposed to adverse academic outcomes, to predict adverse behaviour patterns as early as possible, and to inform targeted interventions.

This use of AI for social good is not to be confused with the necessary work of clinical supervision, student support services, a school inspectorate, curriculum enactment, and quality assurance. Institutional effectiveness is realized only when the academic and administrative units of the school can close the loop on student learning and student development outcomes. The effort culminates in academic improvements at the level of the classroom, the individual child, and ultimately at the institution.

In data-thick societies, AI for social good is gaining traction in policymaking circles and among AI practitioners. However, the use of AI for social good is a poorly understood global phenomenon because of AI’s novelty and ubiquitous use and adoption. Presently, there is no cogent framework for assessing the value and success of projects that aim to bring about social good.

Present metrics, like real-world demand, profitability, and commercial productivity are woefully inadequate. AI for social good proposals and projects need to be assessed against socially valuable outcomes that can parallel the processes used for “B Corporation” credentials in the for-profit environment, or for social enterprises functioning in the non-profit segment.

In this vein, AI for social good should be assessed by adopting “human well-being and “environmental welfare” metrics as opposed to financial ones. The initial work on ethical AI serves as a good starting point but it is far from a framework that is adequate to assess AI projects for social good.

The Global AI Ethics and Governance Observatory is a global resource for policymakers, regulatory bodies, scholars, private sector actors, and civil society to find solution paths to the most pressing challenges posed by the infosphere revolution.

UNESCO’s “Recommendation on the Ethics of AI” applies to all one hundred and ninety-four member states of UNESCO, and enshrines human rights and dignity as the cornerstone of the recommendation.  A distinctive feature of the UNESCO recommendation is its extensive Policy Action Areas, which permit policymakers to translate core values into action concerning data governance, the environment and ecologies, gender, education, health and social well-being, and other spheres.

A good way to identify and appraise AI for social good projects is to analyse them based on their outcomes. AI for social good is successful only insofar as it helps to reduce, mitigate, or eradicate a particular social or environmental problem. The intervention must not introduce novel harms or exacerbate existing ones. Such projects set out to design, develop, and deploy AI assemblages to resolve, mitigate, and avoid ills.

Already, the healthcare industry is overflowing with disruptive applications of AI for social good, including the discovery and development of new molecules for medications. AI offers new ways of reporting gender-based violence and monitoring corporate documents, emails, and chats for unsuitable content.

Human trafficking is a crime against humanity and a global threat to citizen security. Traffickers use social media and digital tools to place advertisements to lure potential victims. AI assemblages and computer vision algorithms can now scrape images from different websites used by traffickers, and label objects in images. AI assemblages can now combine weather data and sensor measurements to optimize, predict, and manage energy consumption.

Beyond treating disease or managing symptoms, AI offers a multidimensional shift toward holistic health that addresses our lifeworlds’ intertwined physical, mental, and social dimensions. A new multidimensional holistic health logic is rapidly emerging. However, the predictive power of AI for social good faces two risks: manipulation of input data, and excessive reliance on non-causal indicators.

Data manipulation using inflated test scores can lead to unfair outcomes that breach the principle of justice as fairness. This type of risk can impair the predictive validity of AI forecasts, and lead to the avoidance of projects that purport to use AI assemblages for social good.