Reimagining Women’s Inclusion in the Future of Work

Key Topics Covered

• Women's participation in the digital economy — current barriers, opportunities, and scale of potential impact
• Ecosystem fragmentation — siloed skilling initiatives without job placement outcomes
• Human-in-the-loop work in AI — data annotation, RLHF, content moderation as sustainable job creation
• AI-driven agricultural advisory — using multimodal AI to support smallholder farmers, with focus on women
• Skill training infrastructure — government, nonprofit, and private sector approaches to digital skills
• Demand-side vs. supply-side skilling — mismatch between where training happens and where jobs exist
• Mobility and time constraints — how women's caregiving responsibilities and safety concerns limit participation
• AI bias, hallucination, and quality assurance — why human validation remains critical and creates jobs
• Local language AI — critical gap in making AI accessible to non-English speakers
• Income inequality in agriculture — wage gaps between male and female farmers and how AI can help narrow them

---

Key Points & Insights

1. Work location mismatch is critical: Most Indian skilling happens at the source of labor (rural areas, small towns), not at demand centers. Women have lower mobility than men due to caregiving responsibilities, safety concerns, and time constraints. Solution requires bringing aggregated work to women rather than expecting migration.

2. AI automation ≠ job elimination: While 60-92 million jobs may be automated, 100-170 million new jobs are projected in AI-adjacent work over the next decade. Volumes will increase because: (a) more AI applications create demand, and (b) AI's probabilistic nature means human validation is always needed to prevent bias and hallucination.

3. Siloed initiatives don't produce outcomes: India has 22+ government schemes for digital skills training but most create skilled workers without job connections. Skills training aimed at quick (3-6 month) livelihood outcomes differs fundamentally from child education investments and requires different ecosystem design.

4. Last-mile intermediaries are missing: Successful models require demand-side aggregators that understand job markets and connect them to women workers. Example: KA (mentioned in portfolio) brought 100,000 workers into formal AI work (annotation, content moderation) in just 3 years—yet investment in such platforms remains negligible relative to need.

5. Human-in-the-loop work is not temporary: Even with 50-60% automation, the remaining 40-50% must be validated for bias and hallucination by human workers. This creates sustainable, scalable roles. NextWealth's example: an all-women medical coding center achieves 99% accuracy (vs. 97% target) on complex ICD code assignment.

6. Contextual/task-based skilling outperforms generic training: Women in rural areas doing fashion accessory tagging for global retail must understand local customer context and preferences. Generic IT training doesn't prepare workers for this. Contextual skilling + fine-tuning + prompting expertise creates skill ladders with progression potential.

7. Multimodal AI significantly improves access: Digital Green's FarmerChat shows 40% of queries use voice/image rather than text—critical for low-literacy populations. Zero-rating data access removes cost barriers (demonstrated in Kenya/Ethiopia). Starter questions and auto-generated follow-ups lower the "capability overhang" (imagination bottleneck).

8. Women farmers face distinct advisory needs: A woman farmer buying cheap chicks received generic advice ("vaccinate them") that assumed income availability she didn't have. Fine-tuned, locally-annotated AI advisory must account for different labor capabilities, income levels, and preferences—not one-size-fits-all responses.

9. Government digital infrastructure is nascent but scalable: NLIT's Digital University platform reached 45,000 registrations in 4 months with 10,000 in AI courses. 160+ industries onboarded. Hands-on virtual labs rival physical environments. Yet awareness and mindset barriers remain—only 25 of 25+ startups from northeast centers were women-led despite inclusive design.

10. Policy and incentives must be intentional: Generic policies (40-hour work weeks, 7-day availability) disadvantage women. Dedicated government funding mechanisms ("fund of funds" for women in digital/AI) are absent. Recommendation: align policy with women's constraints (flexible hours, work-from-home compatibility) and create specific budget allocations for women-centered initiatives.

---

Notable Quotes or Statements

• Atul Satija (Nudge Institute): "Unless we really invest in demand side aggregator platforms, we won't be able to save this all... If you don't have lastmile aggregators of work, you won't be able to take work where women are."

• Mitila Romesh (NextWealth): "AI is probabilistic and can never be 100% perfect. So you'll always need the human in the loop to make it do the quality check, do the validation, to do the testing... It requires human in the loop to do the validation testing make it more trustworthy, reliable, more responsible and explainable."

• Mitila Romesh (NextWealth): "The next 100 million jobs are going to be generated in this space... it's like if out of 100, 60 gets automated and reduces to 40, the 100 itself is going to go up."

• Yona Reshi (Digital Green): "What we're building is agency... it feels like a safe space to ask questions. I don't feel judged. I don't feel dismissed. I can actually ask questions that I wouldn't feel comfortable going to other people around."

• Dr. Mohan (NLIT): "If you want intentional gender outcomes, we need to create a very specific fund of funds for women inclusion in digital and AI."

• Mitila Romesh (NextWealth): "It is a business imperative and not just about inclusiveness... to have women doing this human in the loop work... especially in these industries [financial services, retail, marketplaces, fashion]."

---

Speakers & Organizations Mentioned

Speaker	Organization	Role
Dr. Sharon BH	Tria University	Panel Moderator, Executive Director
Atul Satija	Nudge Institute	Founder & CEO; Ecosystem Builder
Mitila Romesh	NextWealth	Co-founder & Managing Director
Yona Reshi	Digital Green	Assistant Director, Global Strategy
Dr. Mohan Moantraati	NLIT (National Institute of Electronics and IT)	Director General
Dhiraj Dolwani	B2R (Rural BPO)	Participant; 17 years in rural operations (Uttarakhand)

Other Organizations/Programs Referenced:

• KA — Portfolio organization bringing 100,000 workers into AI annotation/content moderation work
• FarmerChat (Digital Green) — AI agricultural advisor; 1M farmers reached across 14 languages
• NLIT Digital University — Free digital skilling platform; 45,000 registrations in 4 months; 160+ industries onboarded
• SafariCom (Telco) — Partnership for zero-rated FarmerChat data access in Kenya/Ethiopia
• Infosys — Referenced workforce projections (92M jobs automated, 170M created)
• Microsoft, KPMG, Intel, Nvidia, IBM — Industry partners on NLIT platform
• Digital Green — 18-year-old organization working on smallholder farmer advisory

---

Technical Concepts & Resources

AI/ML Concepts

• Human-in-the-loop (HITL) — workers validating, annotating, and refining AI outputs; core to bias mitigation and quality assurance
• RLHF (Reinforcement Learning from Human Feedback) — using human feedback to fine-tune models
• Prompt fine-tuning — adapting AI prompts for specific contexts and user populations
• Agentic AI — autonomous AI agents that delegate tasks; creates demand for validation work
• Bias and hallucination detection — critical roles for human workers to prevent AI errors
• Multimodal AI — systems accepting voice, image, text, and other input modalities

Tools & Platforms

• FarmerChat — Agricultural advisory chatbot; multimodal interface; available in 14 languages; 60% of queries use voice/image
• NLIT Digital University (nlet.edu.in) — Free hands-on skilling platform with virtual labs (semiconductors, AI, cybersecurity, industry 4.0)
• Interview Simulator — NLIT tool for job prep
• Recommendation Engine — Personalized learning on NLIT platform
• Credit Paper System — Portfolio tracking on NLIT platform
• Starter Questions — FarmerChat UX feature to lower capability overhang
• Auto-generated follow-up questions — Prompts to deepen farmer engagement

Methods & Frameworks

• Demand-side aggregation — Bringing work to workers rather than training workers to migrate
• Contextual/task-based skilling — Training workers on specific tasks with local context (e.g., fashion tagging for global retail with local knowledge)
• Skill ladders — Progression pathways: task-based → contextual → fine-tuning/prompting expertise
• Randomized controlled trials (RCTs) — Digital Green conducting RCTs in Kenya, India, Ethiopia to measure FarmerChat's impact on farm income and productivity

Data & Statistics

• 1 million farmers reached by FarmerChat (as of last week mentioned in talk); ~60% in India, rest in Kenya, Nigeria, Ethiopia, Brazil
• ~33% of FarmerChat users are women; goal is to reach 45% next year
• 40% of NLIT platform registrations are women; 100,000 registered in AI alone
• 45,000 registrations on NLIT Digital University in 4 months; 5 new students per minute
• 100,000 workers placed by KA in AI annotation/content moderation work in 3 years
• 20,000 workers placed by NextWealth; 60% women; 5,000 across 11 centers
• 22+ government skilling schemes in India, largely siloed without job placement outcomes
• 1 in 1,000 farmer-to-extension-agent ratio in agriculture; women even more neglected
• 7 in 10 farmers adopting FarmerChat advice within 30 days of receiving it
• 99% accuracy achieved by all-women medical coding center at NextWealth (vs. 97% target)
• 92 million jobs automated; 170 million created (Infosys projection)
• 40% of 8 million FarmerChat queries are multimodal (voice/image)
• 25 startups created via NLIT northeast initiatives; none women-led to date

Policy & Infrastructure

• DPI (Digital Public Infrastructure) — Government-backed digital systems enabling the digital economy
• Platformization of work — Moving traditional jobs onto digital platforms
• Zero-rating — Free data access for specific apps (demonstrated with SafariCom/FarmerChat)
• Flexible work policies — Recommendations for 40-hour weeks, work-from-home compatibility for women
• Fund of funds — Dedicated government financing mechanism for women-centered digital/AI initiatives (proposed, not yet implemented)

---

Policy Recommendations & Calls to Action

1. Create dedicated government funding: "Fund of funds" specifically for women inclusion in digital/AI ecosystems (mentioned by Dr. Mohan)
2. Invest in demand-side aggregation: Shift from supply-side training to platforms that bring aggregated work to women
3. Align policies with women's constraints: Flexible hours, work-from-home options, caregiving time, safety measures
4. Localize AI aggressively: Prioritize local languages and contextual fine-tuning; treat as business imperative, not afterthought
5. Integrate industry into skilling design: Ensure training outcomes map to actual job demand, not theoretical skills
6. Support "AI factories" model: Hub-and-spoke approach where urban centers develop AI and small towns execute human-in-the-loop validation work