Research

IDRC’s white paper on AI and human development emphasizes that research on AI must be:

  • Interdisciplinary: Many of the research questions address the intersection of social and technical factors. Approaching them effectively will require multidisciplinary collaborations across the social (including economic and political), humanistic, and computational sciences.
  • Locally conducted: To ensure relevance and usefulness, the research needs to be driven and conducted by researchers in the Global South.
  • Designed to support practice and policy: To have real impact, research must be rigorous and generate actionable findings that facilitate the implementation of programs and policy.

Based on the literature, IDRC has identified three key thematic areas in AI and development that require further research:

  • Policies and regulations;
  • Inclusive and ethical AI applications; and
  • Infrastructure and skills.

Below, we elaborate on each of these thematic areas by listing a series of recommendations for research necessary to make concrete progress. This list is not intended to be comprehensive, but rather highlights the most pressing interventions emerging from scoping and review[1].

Thematic Research Area 1: Policy and regulatory structures

Foster the design of policies and regulations that enable inclusive and rights-based AI

  • Conduct baseline research on the prevalence of AI applications and policies in Global South countries: Despite pockets of AI activity in the Global South, there are no systematic overviews of the level of this activity. Baseline data collection should include the sets of AI policies, regulations, applications, existing open datasets, and skill levels.
  • Learn about effective regulatory models: Identify regulatory responses to AI and determine risk levels that are appropriate for settings with low institutional capacity. While lessons from the Global North are useful, research is critical to identify unique institutional and regulatory approaches for Southern contexts.
  • Track the impact of AI on employment and work: Conduct policy research to understand the effects of AI on employment, the nature of work, and labour markets.
  • Explore approaches to addressing liability, accountability, and redress for AI decision-making: Design regulatory systems and frameworks to determine liability and accountability for AI decision-making that is erroneous, biased, or discriminatory, and establish mechanisms for redress. Research is critical here to uncover and document which systems for accountability and redress are effective and in what contexts.
  • Study the impact of AI on human rights: Tailoring impact assessments to the risks of AI would help encourage development programs to incorporate AI technology in ways that respect and promote human rights, including privacy, equality, and freedom of expression.

Thematic Research Area 2: AI Applications

Catalyze the development of inclusive and ethical AI applications

  • Support the development and deployment of innovative AI applications for social good: Invest in research, development, and use of applications for education, health, the environment, food security, etc., and ensure that these applications are ethical and inclusive. AI applications will require homegrown solutions to be effective.
  • Research the social impact of AI innovations: Research is needed to better understand which AI applications work (or don’t work), for whom, and in what contexts. We need to know who benefits from AI applications and how, as well as who is left out or harmed. New methodologies for impact assessment and evaluation may be required.
  • Test and monitor bias in AI applications: AI systems that make or inform decision-making that affects humans’ well-being should be researched, tested, and monitored for bias and errors across different contexts and communities, both before release and continuously.
  • Explore models of participatory design for AI: Conduct research into practices that support the development of inclusive AI applications. AI stakeholders in the field should release data on diversity of participation in design and development.
  • Action research to deepen understanding of how to effectively and equitably scale proven AI applications: Research on the process of scaling AI applications, both vertically to encompass additional functionality and horizontally to expand to new locations, is critical to extending the benefits of these applications.

Thematic Research Area 3: Infrastructure and skills

Build the infrastructure and skills for inclusive and ethical AI

  • Support programs to build AI expertise in government: Promote AI expertise in all branches and at all levels of government, including regulatory entities and, potentially, new research advisory bodies.
  • Foster local capacity to lead the design, development, and deployment of AI applications: For example, supporting the growth of multidisciplinary AI centres of excellence in the Global South in order to engage in local development and research and provide evidence-based input into the shaping of national policy and regulatory decisions.
  • Develop and test cost-effective approaches to build relevant AI skills, particularly among women and marginalized populations: Research should bolster this activity through an exploration of low-cost models for developing AI tech skills and producing and testing effective curriculum and pedagogies.
  • Expand access to data and computing resources: As much as possible, AI research, tools, and training datasets need to be made freely available.
  • Study the benefits and risks of open AI: Conduct research on the short- and medium-term risks and benefits of openness in AI (i.e., sharing AI resources, datasets). Special attention should be paid to the issue of balancing the sharing of datasets with the safeguarding of privacy.