The purpose of this paper is to present a case study on the influence of science diplomacy on the AI/ML growth landscape driving social impact in India. First, we will begin with context-setting: cross-sectoral collaborations within AI/ML space in India and the scope of these implementations. Next, we will unpack the roles and responsibilities of stakeholders within the regulatory landscape. Finally, the author reflects on the evidence, highlighting the risks, mitigation strategies in place that influence the long-term view of AI/ML growth & development.

Artificial intelligence (AI)—the ability of machines to simulate human intelligence by learning, reasoning, and making rational, intelligent decisions—is reshaping global power structures. Countries are investing heavily in AI to gain strategic advantages across defense, economic growth, and technological leadership. As AI increasingly underpins military systems, economic productivity, and digital infrastructure, nations that lead in its development, research, and cross-sector collaborations are poised to exert greater geopolitical influence regionally and globally in an era where technological superiority defines global standing.

The author joined as an analyst on two projects that provided exposure to AI implementation, challenges and pathways of growth within the social impact space for the case of India:

Science diplomacy allows for state (e.g., government institutions, departments and ministries) and non-state actors (e.g., non-profits, private philanthropies, technical partners) to collaborate on shared global challenges—such as hunger, poverty, and unemployment—by leveraging soft power and advancing scientific and technological innovation, while avoiding direct exposure of governance or market failures. De Bengy Puvallee et al (2025) study the Gates Foundation’s (GF) philanthropic portfolio in Europe and term it as network diplomacy, where catalytic philanthropy meets political, strategic vision, advocating to drive policy change. Breen and Kumar (2023) reviewed Gates Foundations’, Wellcome Trust, and Rockfeller Foundations’ (RF) granting patterns to move the needle. They noted an interesting pattern: non-profits for GF and RF were favored for funding from 2018-2020, digital health and biomedical technologies was a sub-thematic focus area. Further, they funded global partnerships, intergovernmental organisations, government organizations and universities, to help achieve “network diplomacy” goals in the long term (see Figure 1).

For India, AI holds promise as a catalyst to accelerate progress, while providing mechanisms to leapfrog traditional hurdles such as poor infrastructure and bureaucracy (Kalayanakrishnan et al, 2018). Most notably, this move from the Ministry emphasized the need for bilateral cooperation through leveraging partnerships and sharing technological processes for a more efficient outcome. In nearly every sector—finance, healthcare, law enforcement, transportation, agriculture, environmental conservation—one finds applications in which AI can be effective (ibid). Take, for example, the SMART cities initiative launched by the Government of India to upgrade 100 cities with enhanced services, managerial, and that aims to manage the urban exodus by integrating data-driven decision-making enhanced by AI in managing crime, managing large crowds, and improving waste management techniques. As AI/ML becomes increasingly integrated into urban governance, if implemented correctly, there is the potential for more sustainable, safe, inclusive, and resilient cities (Allam & Newman, 2018; Choudhary & Sarthy, 2022; Lourenco et al., 2018). In Table 1, I have highlighted certain key innovations in the different domains brought about by implementing AI & ML technologies in India.

The rhetoric of “AI for social good” has been adopted by organizations as well as the Indian state in its AI policy positions (Radhakrishnan, 2021). Federoff (2009) evokes Gates and his worldview on philanthropy that promotes science and technology keeping business central, he called this “creative capitalism”. It is this approach that demonstrates Gates Foundation’s many efforts to invest and hasten the AI implementation across sectors in India. Suzman, in his article, emphasizes first principles for implementing AI/ML technologies: Invest in low-income countries targeting resource-poor regions and communities ensuring transparency, accountability, and ease of access.

Science diplomacy is the use of scientific collaborations among nations to address the common problems facing 21st century humanity and to build constructive international partnerships (Federoff, 2009). There are significant downsides to stakeholders trying to influence AI regulation in India, especially given the interest of international donors and tech giants. Mohanty and Sahu (2024) highlight that MeitY released an advisory (promptly rescinded with a more balanced advisory) in March 2024, “which mandated compliance with immediate effect, directed companies to obtain the government’s permission before deploying certain AI models in India, and to take steps to prevent algorithmic discrimination and the distribution of deepfakes.” Therefore, most companies are employing self-regulation to mitigate risks such as user harm, reduce misinformation, increase accountability, and balance commercial interests. However, Digital India Bill (still under consideration at the Parliament) which will provide a legal framework requires a comprehensive risk classification framework⁶ drawing from empirical evidence.

An analysis of regulatory models of other countries demonstrates that there are two main spectrums: United States, United Kingdom, Japan and Singapore puts the onus on companies to create a self regulation framework whereas countries like Brazil, Australia, China have binding regulations managed by government authorities. Experts recommend that India implement co-regulatory framework (also in line with MeiTY’s current “light touch” approach) where only “high-risk cases” are handled by nodal agency with self-regulation in place for implementing agencies complemented by public consultations, interagency coordination for AI governance and an interdisciplinary approach.

Given the complex risks that AI/ML carries, philanthropic granting organizations in the US such as Bill & Melinda Gates Foundation, Agency Fund, Google Research India are also invested in building a robust AI infrastructure in India. This includes funding a second layer of AI implementation agencies that help deliver high-quality Indian language data, reduce bias and create open-source codes (for replication) along with centers at premier technology institutes of the country for human oversight. For instance, ARTPARK at a premier science institute in India, Indian Institute of Science (IISc) have undertaken the complex task of collecting speech and text data directly from local communities to train existing large language models. Another example is AI4Bharat, a nonprofit at Indian Institute of Technology (IIT) Madras that creates open-source datasets and most notably, creates text to speech technologies for public consumption. Therefore, this is a multi-tiered ecosystem that private philanthropies along with MeitY that helps mitigate the growing risks, legal compliance requirements, data management and incorporate feedback from grassroots (see Figure 2).

The author’s experience reveals that, given the novelty of AI interventions spurred by new philanthropy capital and investments, long-term and multi-dimensional understanding from implementing organisations remains lacking. There is a need to unpack if grantees or implementing organizations can streamline and maximize their field activities and existing community projects by implementing AI/ML technologies, or reach their last-mile beneficiary. Finally, project cycles and workflows are not prepared to practically implement AI assistants or chatbots; granting proposals for AI-specific projects at implementing organisations without investing in organizational infrastructure to incorporate technical capabilities can lead to more harm than gain.

Even while small organisations grapple with poor operational understanding around AI implementations and uneven funding mechanisms for social impact, there are newer dimensions to AI policy in India that have emerged. In 2023, India adopted the European General Data Protection Regulation (GDPR) style of law called Digital Personal Data Protection Act (DPDP) which mandates that organisations safeguard beneficiary data and with a big focus on consent, data storage and cloud computation with declarations wherever it applies. Owing to the proliferation of AI in high-risk areas, pressure is mounting to design and govern AI to be accountable, fair and transparent especially to the end users. How can this be achieved and through which frameworks? (Cath, 2018) To address this, Digital India Bill has been introduced with an intention to replace the Information Technology Act, 2000. Critiques of the Information Technology Act highlight that it is outdated in light of the rapid growth of the internet and new technologies and does not incorporate safety, trust and accountability applicable to AI/ML, blockchain and social media technologies. However, the author cautions that for new legal frameworks to be successfully implemented, it also requires an overhaul in interagency coordination across departments – a practice not easily seen in a technocratic, bureaucratic setup such as that of India.

The European Union released the first ever Artificial Intelligence (AI) Act in 2023 to regulate risks posed by AI/ML implementations. It categorizes risks based on harm & societal impact, implements robust governance guardrails around data and security and finally, focusses on democratic values rather than treat policy as technical compliance. The EU AI Act represents a major legal milestone—defining a balanced global standard that attempts to empower AI innovation while safeguarding societal values and rights. Indian companies and government authorities now monitor the territorial impact of the AI Act and high-risk applications to include human oversight to pass risk assessment under the Act.

There have been many critiques on AI/ML recreating and reinforcing structural inequalities and bias through their applications in resource poor settings where users may include remote, urban poor and marginalized communities in India. In this article, I have aimed to unpack opportunities and risks that digital integrations pose when interfaced with a multilayered society and evolving institutions. There is no doubt that the introduction of AI/ML to existing digital platforms and services actively shape and are shaped by politics, policy, ethical considerations, data use, and public expectations around technology.

In contexts such as healthcare and education, where physical access remains constrained, the increasing penetration of smartphones and internet connectivity have been seen as significant opportunities for leveraging AI and machine learning (ML) solutions to enhance service delivery and support systems. Potential use cases that integrate machine learning to digital solutions for improved care and health access include improved disease diagnosis, treatment selection, and clinical laboratory testing. India, home to over 250 million school students, the largest student population globally—has launched several digital inclusion initiatives aimed at bridging educational disparities. Flagship programs such as SWAYAM (Study Webs of Active Learning for Young Aspiring Minds) and NPTEL (National Programme on Technology Enhanced Learning) offer free online courses and digital resources, particularly targeting students in rural and remote regions. The integration of AI/ML technologies into these platforms along with multistakeholder support helps enhance personalization, accessibility, and linguistic inclusivity—crucial in a country like India.

There are several risks that we have highlighted in this article. Firstly, most actors in the Indian ecosystem are consumers or deployers of pre-trained, open source models with primary tasks of prompt engineering, fine-tuning data sets and last mile deployment (Aneja et al, 2024) whereas their agency to influence upstream processes are limited. Secondly, Sahoo et al. (2024) highlight a need for reliable benchmark datasets as current pre-trained models predominantly focus on English language and the Western context, failing to encapsulate India's unique socio-cultural nuances. or questioning gender or socio-cultural biases during chatbot implementations. Finally, private philanthropies remain important stakeholders within the AI/ML landscape in India providing funding, advisory and legal support to small organizations.

In the author’s experience, there are fundamental and practical measures that small organisations must address to create an authentic, indigenous LLM; 1. ensuring better linkages with funding agencies for long-term technical development and integration (Reddy et al, 2023), 2. secure computational support for closed source customised models for specific use cases, 3. invest in primary data collection to incorporate colloquialisms, dialects in the Indian context (Joshi et al., 2020; Bhattacharya et al., 2021), 4. critically examine embedded biases around gender, caste in training datasets, assess their social and ethical implications and eliminate them methodically.

As a way forward to expand the reach of AI/ML, the author recommends that organisations implementing solutions follow these principles: 1) foster international collaborations with a clear national vision and goals on governance, contextuality and voice 2) clarify ethical and safety standards and protocols in India within the new law regimes nationally and globally. 3) intentionally breaking new ground in private and government partnerships, especially within high risk applications where industry and corporations can help scale and customize applications transforming AI/ML through innovation and last mile reach.