Local Voices First: Why Inclusive AI for Data Systems Must Start on the Ground

Abstract

The panel examined how AI can be harnessed as public‑infrastructure for official statistics when it is rooted in local data ecosystems, languages, institutions and communities. Drawing on examples from India’s multilingual Bhashini platform, capacity‑building programmes, and civil‑society collaborations, the discussion highlighted three levers—data access, capacity development, and trust—that determine whether AI narrows or widens existing gaps. Participants debated the role of national statistical offices, the need for co‑creation and participatory evaluation, and concrete ways to bring AI to the last mile (e.g., offline tools, voice‑enabled interfaces). A brief Q &A explored connectivity challenges, the emerging relevance of quantum computing, data‑privacy legislation, and how demand‑driven solutions can be cultivated in the Global South.

Detailed Summary

1.1 Moderator’s Opening Remarks (Mercedes Fogarassy)

• Positioned the session within the broader India AI Summit and linked it to earlier talks on AI readiness and open data.
• Stressed the central question for PARIS 21: “AI for whom and built by whom?”
• Highlighted the pressure on statistical offices in the Global South: higher expectations, tighter timelines, and shrinking resources.
• Asserted that AI can both solve problems and exacerbate inequities if not grounded in local realities.
• Defined the premise: AI becomes truly impactful only when it is built on local data ecosystems, languages, institutions and, crucially, local people.

1.2 Opening Remarks (Shuchita Rawal, Civic Data Lab)

• Welcomed guests on behalf of Civic Data Lab and PARIS 21.
• Introduced four guiding keywords: localisation, inclusivity, data, AI.
• Emphasised that AI is only as good as the data fed into it; current data‑system designs often exclude the very communities they serve because they are created in boardrooms and tech hubs.
• Presented Bhashini, India’s multilingual AI platform (voice‑to‑text, translation across many Indian languages), as a concrete illustration of “AI from the ground up.”
• Stressed the need for co‑creation—communities must be partners, not just beneficiaries.
• Outlined Civic Data Lab’s work in building data‑AI collaboratives across gender, climate, disaster risk, and health.
• Closed with a litany of recurring themes from the organization’s recent “Civic Sabha 2.0”: responsible AI, ethical AI, collaboration, localized solutions, AI skilling, accessibility, safe & trusted AI, and data‑AI collaboratives.

2. Panelist Introductions (Moderator)

• Johannes Jütting – Executive Head, PARIS 21; development economist, former UN SG expert‑group member on the data revolution.
• Dr. Saurabh Garg – Secretary, MoSPI; senior Indian civil‑service officer with experience in digital public infrastructure, social policy and finance.
• Gaurav Godhwani – Co‑founder, Civic Data Lab; a decade of data‑innovation experience in climate resilience, public finance, health.
• Saachi Bhalla – Deputy Director, Gates Foundation (India); focus on gender equality, women’s economic empowerment, evidence‑driven development.

3. Problem Statement: Why “Local” AI Matters (Johannes Jütting)

1. Global vs. Local Lens – While global AI conversations are valuable, the panel needed to illuminate what local AI can actually achieve.
2. Alignment with Sustainable Development Goals (SDGs) – Emphasised that AI should serve well‑being, especially for rural and underserved populations, in line with the 2030 Agenda (“Leave no one behind”).
3. Key Levers Identified
   • Access – Availability of data and compute; raised concern about a possible “data winter” if data becomes locked.
   • Capacity Development – Need for training and technical expertise, especially among small‑holder farmer organisations (citing his own agricultural‑economics background).
   • Trust – Without trust in AI recommendations (e.g., a farmer’s phone‑based advice), adoption stalls.
4. Role of National Statistical Offices –
   • Setting standards, quality‑controlling data.
   • Integrating citizen‑generated data and private‑sector data after rigorous validation.
5. Call for Multi‑Stakeholder Collaboration – Urged the Indian panelists to share concrete experiences.

4. Government Perspective on Inclusive AI (Dr. Saurabh Garg)

4.1 Democratizing AI Resources

• Co‑chaired the AI‑Summit working group on democratizing AI resources (collaboration with Egypt & Nigeria).
• Noted the uneven distribution of compute, models, data, and talent.

4.2 AI as Public Infrastructure

• Described data as the raw material for AI; stressed the necessity of standardised, machine‑readable, interoperable metadata.
• Highlighted ongoing work to uniform metadata, classification, and standards across ministries and states.

4.3 Enhancing Access & Inclusivity

• Edge computing and small‑domain models are being explored to decentralise AI capabilities.
• Multilingual and voice‑based interfaces are key to reach illiterate or low‑literacy users; voice is language‑neutral and does not depend on reading ability.

4.4 Built‑in Safeguards

• Emphasised AI‑by‑design: ethics, safety, and fairness must be embedded from the outset rather than retro‑fitted.

4.5 Institutional Coordination

• Stressed cross‑ministerial collaboration to ensure trusted data, standard‑driven AI, and alignment with SDG targets.

5. Civil‑Society View: Co‑Creation, Evaluation, Skilling (Gaurav Godhwani)

5.1 Poll of Audience Familiarity

• Quick show‑of‑hands indicated very low uptake of AI applications in local languages, especially among elders.

5.2 Three Pillars for “Local AI”

Pillar	Core Idea	Action Points
Co‑creation	Communities should be involved throughout the AI lifecycle, not just at testing.	Incentivise participation from conceptualisation, dataset curation, early testing, to scaling.
Participatory AI Evaluation	Bridge global AI models with local contexts via local experts who can flag cultural or linguistic bias.	Create a framework that formally recognises and rewards domain‑ and culture‑experts for bias‑checking.
AI Skilling	Build a local talent pipeline so communities can develop, maintain, and use AI tools.	Offer training that leads to secure jobs, ensuring long‑term sustainability.

5.3 Emphasis on Incentives

• Suggested that incentives (financial, recognition, career pathways) keep local stakeholders engaged across the lifecycle.

6. Gates Foundation Angle: Institutional Capability & Demand‑Oriented Design (Saachi Bhalla)

6.1 AI as an Institutional Capability

• Recommended treating AI not as a standalone tool but as a capability embedded within trusted institutions and aligned with statistical standards.

6.2 Inclusion at the Outset

• Insisted that inclusion be designed in from the beginning, not as an afterthought.

6.3 Use‑Case Differentiation

• Highlighted that AI usability varies across levels:
  • National – dashboards, policy‑level analytics.
  • Community – tools for frontline workers (e.g., women’s self‑help groups).
  • Individual – personal devices, offline functionality.

6.4 Example: National Rural Livelihoods Mission (NRLM)

• AI‑enabled tools for women frontline workers to provide data‑driven advice to women farmers (crop, livestock, market).
• Emphasised that trusted frontline agents bridge technology and community, allowing AI to be truly local.

7. Audience Q &A

Questioner	Core Issue	Key Responses
Anubhuti (Nalanda University)	Connectivity in remote areas (Bharat Vistar rollout)	Garg: Mobile network coverage ≈ 98‑99 %; Bharat Vistar works on feature‑phone platforms, not smartphone‑only; connectivity is improving but not the sole barrier.
Vineet Prakash (Quantum Nebula startup)	Potential of quantum computing for AI in health & other verticals	Garg: Quantum may enhance cyber‑security and accelerate compute, but practical applications are still future‑oriented.
Himali (Data Scientist)	Public awareness of data collection & DPDP Act; bridging privacy gap	Garg: Government data sets follow strict protocols; private‑sector data (e.g., social media) remains a work‑in‑progress. Emphasised need for individual awareness and enforcement mechanisms under the DPDP Act.
Follow‑up (Sachi Bhalla)	Offline capability & last‑mile access	Confirmed that many tools (health, agriculture, education) are designed with offline‑first functionality, syncing when connectivity returns.
Additional comment (Saurabh Garg)	Role of civil‑society & multilingual voice interfaces	Stressed civil‑society as an essential conduit; highlighted the need for voice‑based explanations of data‑privacy laws for dialects lacking a script.
Additional comment (Johannes Jütting)	Supply‑driven vs. demand‑driven AI solutions; lessons from ICT rollout	Noted that latent demand (e.g., healthcare, education, sanitation) should drive AI investment; cited ICT experience where mobile phones dramatically increased financial inclusion in Kenya.

Themes Emerging from Q &A

1. Connectivity is improving, but design must accommodate intermittent or no‑internet environments.
2. Quantum computing is an emerging frontier, primarily for security rather than immediate AI service delivery.
3. Data‑privacy awareness is low; civil‑society can translate legislation into understandable, local‑language formats (including voice).
4. Demand‑creation is crucial: AI solutions must address real, everyday public‑service needs before they can be scaled.

8. Closing Remarks (Moderator & Panel)

• Mercedes Fogarassy summarised the cross‑cutting emphasis on inclusion, noting that each speaker contributed a distinct perspective (policy, civil‑society, technology, funding).
• Re‑iterated the supply‑vs‑demand tension and the need to avoid “spaghetti‑testing”—instead, invest in solutions that respond to verified community needs.
• Thanked the audience, highlighted the brief pause for an alarm clock (a humorous cue to wrap up), and closed the session.

Key Takeaways

• AI must be ground‑up: Effective AI for official statistics requires local data, local languages, local institutions, and local people throughout the lifecycle.
• Three foundational pillars: Co‑creation, participatory evaluation, and AI skilling are essential to move from pilot projects to sustainable impact.
• Access, capacity, trust: These three levers—open data & compute, technical capacity building, and building trust in AI outputs—determine whether AI narrows or widens equity gaps.
• National statistical offices as gatekeepers: By standardising metadata, ensuring machine‑readability, and integrating citizen‑generated data, they can anchor AI on trustworthy, high‑quality foundations.
• Multilingual & voice‑first interfaces are critical for India’s linguistic diversity and for reaching low‑literacy users.
• Offline‑first design mitigates connectivity constraints; many government‑sponsored tools already incorporate sync‑when‑online capabilities.
• AI as an institutional capability: Embedding AI within trusted public entities (e.g., NRLM, MoSPI) and aligning it with statistical standards maximises relevance and sustainability.
• Demand‑driven innovation: Successful AI interventions start from real public‑service problems (health, education, sanitation) rather than from technology push.
• Civil‑society’s pivotal role: Local NGOs, academic groups, and community organisations are essential for localisation, awareness‑raising, and feedback loops (including voice‑based privacy education).
• Future outlook: While quantum computing promises security gains, immediate priorities lie in standardisation, capacity building, and participatory governance to ensure AI truly benefits the most vulnerable.

---

See Also:

• exploring-a-regulatory-framework-for-open-data
• flipping-the-script-how-the-global-majority-can-recode-the-ai-economy
• ai-innovators-exchange-accelerating-innovation-through-startup-and-industry-synergy