By Kerstin Nothnagel.
With artificial intelligence transforming many aspects of society, from healthcare to education to economic development, how can it be used to reduce rather than perpetuate inequalities? In particular, given that artificial intelligence can widen gaps by exacerbating existing inequalities through biased datasets, lack of infrastructure, and limited access to resources, how can the benefits of artificial intelligence be brought into the reach of low-income nations and marginalised communities? What practical steps can be taken to ensure artificial intelligence is developed and applied in a way that is inclusive and benefits everyone?
My work has been in the health field, but the findings are likely to be more broadly applicable. I suggest three strategies that would enable artificial intelligence to reduce inequities. The first two are key contributions that researchers can make. The third is a call to policy makers and funders. An example is provided for each strategy.
1. Develop context-specific artificial intelligence-based solutions
One-size-fits-all artificial intelligence models rarely work in diverse global contexts. A key role for researchers is therefore to develop artificial intelligence systems that are adaptable to the unique needs and constraints of different regions. In particular:
- Artificial intelligence tools should be designed with local infrastructure in mind, ensuring they work effectively in areas with limited internet access and outdated technology.
- Language diversity must be considered—many artificial intelligence applications rely on dominant languages, excluding those who speak under-represented languages and dialects.
- Data collection should prioritise inclusivity, using diverse sources that reflect the realities of all populations, not just those in wealthier nations.
Example: A study assessing the generalisability of artificial intelligence models across hospitals in low- and middle-income countries and high-income countries found that models trained on data from high-income settings often underperformed in low- and middle-income countries. Retraining these models with local data significantly improved their performance, highlighting the necessity of context-specific adaptation (Yang et al., 2024).
2. Collaborate globally
A key role for researchers is to ensure broad collaboration between researchers, policymakers, and local communities. Artificial intelligence solutions should not be imposed from the outside but co-developed with those who will use them. In particular:
- Partner with researchers and organisations in low-income countries to co-design artificial intelligence models suited to their specific challenges.
- Engage with local governments and policymakers to align artificial intelligence projects with national development goals.
- Help develop and support open-source artificial intelligence initiatives, allowing underfunded regions to access and modify artificial intelligence tools for their needs.
Example: An analysis of international collaboration patterns in artificial intelligence research revealed that countries like Vietnam, Saudi Arabia, and the United Arab Emirates have high degrees of international participation. Such collaborations have facilitated the development of artificial intelligence solutions tailored to local contexts, demonstrating the value of global partnerships (Hu, Wang and Deng, 2020).
3. Focus on sustainable infrastructure
For artificial intelligence to create lasting change in low-income nations and marginalised communities, there must be long-term investment in the underlying infrastructure. Without access to the internet, computing power, and data storage, even the most well-designed artificial intelligence tools remain unusable in many parts of the world. In particular, policy makers and funders could beneficially pay attention to:
- expanding internet and digital access in underserved regions to support artificial intelligence-based applications;
- providing affordable artificial intelligence training programs to build local expertise and reduce reliance on external developers;
- supporting policies that encourage responsible data collection and ethical artificial intelligence deployment.
Example: A study examining artificial intelligence deployment in healthcare in low- and middle-income countries highlighted the challenges and opportunities related to data sharing. Investments in digital infrastructure and data governance frameworks have enabled the successful implementation of artificial intelligence-driven health interventions in these regions (Kaushik et al., 2024).
Relevance to complex problems
The need for inclusive artificial intelligence development becomes even more crucial when tackling complex societal and environmental challenges. Artificial intelligence has immense potential to address global issues such as climate change, public health crises, and economic inequality—but only if it is designed and implemented with equity in mind. In these contexts, artificial intelligence must navigate systemic interconnections, adapt to diverse local realities, and avoid reinforcing existing disparities. To fully harness the potential of artificial intelligence, interdisciplinary collaboration, ethical governance, and inclusive innovation are essential in ensuring that artificial intelligence solutions create meaningful and lasting change for all.
Concluding questions
What other strategies should researchers, policy makers and funders consider for making the adoption of artificial intelligence more inclusive? Have you encountered successful approaches to equitable artificial intelligence implementation? I’d love to hear your insights!
References:
Yang, J., Dung, N. T., Thach, P. N., Phong, N. T., Phu, V. D., Phu, K. D., Yen, L. M., Thy, D. B. X., Soltan, A. A. S., Thwaites, L. and Clifton, D. A. (2024). Generalizability assessment of AI models across hospitals in a low-middle and high income country. Nature Communications, 15, 1: 8270. (Online – open access) (DOI): https://doi.org/10.1038/s41467-024-52618-6
Hu, H., Wang, D. and Deng, S. (2020). Global collaboration in artificial intelligence: Bibliometrics and network analysis from 1985 to 2019. Journal of Data and Information Science, 5, 4: 86–115. (Online – open access) (DOI): https://doi.org/10.2478/jdis-2020-0027
Kaushik, A., Barcellona, C., Mandyam, N. K., Tan, S. Y. and Tromp, J. (2025). Challenges and opportunities for data sharing related to artificial intelligence tools in healthcare in low- and middle-income countries: A systematic review and case study from Thailand. Journal of Medical Internet Research, 27: e58338. (Online – open access) (DOI): https://doi.org/10.2196/58338
Use of Generative Artificial Intelligence (AI) Statement: Generative artificial intelligence was used to improve grammar and spelling in the development of this i2Insights contribution. (For i2Insights policy on generative artificial intelligence please see https://i2insights.org/contributing-to-i2insights/guidelines-for-authors/#artificial-intelligence.)
Biography: Kerstin Nothnagel is a PhD student in Population Health Sciences at the University of Bristol, UK, where her research focuses on the implementation of artificial intelligence technologies in healthcare. She also examines the policy-making processes surrounding artificial intelligence in the healthcare sector. Currently, she is working on an international artificial intelligence policy project, which explores the impact and regulation of artificial intelligence technologies. This project has been awarded the prestigious Alan Turing Institute Award. She is now based at Human Technopole, an Italian Government Research Centre in Milan, where she contributes to global discussions on artificial intelligence policy.
Acknowledgement: Professor Luisa Zuccolo, Research Group Leader of the Health Data Sciences Centre at Human Technopole, supervises my work at that Centre.