Global Dialogue on AI Usage-Data for Labour Market Resilience

Detailed Summary

Theme: Why AI‑usage data are a strategic imperative for labour‑market policy.

1. International AI‑Safety Report – Chaired by the speaker, co‑authored by 29 countries, the EU, OECD and the UN.

   • Exposure statistics: ~60 % of jobs in advanced economies and ~40 % in emerging economies are “at risk” from general‑purpose AI.
   • Precautionary principle: Even with uncertainty about the speed of AI progress, policymakers must prepare for worst‑case labour outcomes.

2. Economic dynamics

   • Delay effects: Gap between AI capability, market adoption, and observable job displacement.
   • Power asymmetries: Wealth accrues to model owners (concentrated in two countries) while workers in many nations face job loss, amplifying intra‑ and inter‑national inequality.

3. Strategic implications

   • Access to advanced AI as a competitive advantage: Nations lacking reliable access risk a “dual disadvantage” – reduced wealth generation and rising unemployment, potentially triggering fiscal crises.
   • Need for coordinated international governance: Example of recent Germany‑Canada partnership on ethical AI.

4. Call to action

   • Emphasised the importance of granular AI‑usage data to inform transition policies and to enable equitable, globally‑coordinated responses.

---

2. Presentation – Bharat Chandar

Theme: Empirical evidence on AI exposure, adoption, and the data gaps that hinder policy.

2.1 Exposure‑Employment Findings (U.S. focus)

Metric	Result
Relative employment change	16 % decline for young workers (≤30 y) in AI‑exposed occupations, beginning late‑2024/early‑2025.
Older workers	No statistically robust decline; slight, non‑significant trend.
Robustness	Controlled for macro factors (e.g., interest‑rate movements).

2.2 Adoption Mapping (40+ countries)

• Data source: Online job‑posting feeds used to construct firm‑level AI‑adoption indices.
• Geographic leaders: Israel, Japan, India (high adoption intensity).
• Sectoral driver: Information‑sector firms show the steepest adoption curves.

2.3 Exposure vs. Adoption – A Dichotomy

• Exposure studies → consistent negative employment effects.
• Adoption studies → largely null or mixed effects.

2.4 Four Explanatory Hypotheses (Why results diverge)

1. Anticipatory hiring cuts – Firms reduce hiring before actual AI deployment.
2. Measurement error in adoption – Half of AI use occurs on personal accounts, invisible to enterprise‑level metrics.
3. Competitive pressure – Non‑adopters lose market share, leading to shrinkage rather than job creation.
4. Confounding macro‑economic trends – Observed patterns may be driven by factors unrelated to AI.

2.5 Data Needs to Resolve Ambiguities

Need	Rationale
Executive belief surveys	Capture anticipatory hiring decisions.
Fine‑grained firm‑level usage data (task‑level, not just job titles)	Better measure true adoption.
Productivity‑impact metrics	Distinguish efficiency gains from displacement.
Longitudinal tracking	Observe whether trends persist or reverse over time.

Conclusion: Better, high‑frequency, task‑level AI‑usage data are essential to differentiate genuine displacement from other economic forces and to design targeted labour policies.

---

3. Panel Discussion (moderated by Robert Traeger)

3.1 Opening Statements (Brief)

Panelist	Main Points
Dr. Shemika Ravi (India)	India’s AI adoption is high; sectors such as ag‑tech, fintech, health‑tech, ed‑tech are expanding. AI can address “last‑mile” challenges (e.g., health & education delivery).
Ambassador Philip Thigo (Kenya)	Only ~15 % of Kenya’s economy is formal; automation risk (≈52 % of jobs) must be understood relative to the informal sector. AI can boost MSMEs, but regulation and data gaps must be addressed.
Robert Traeger	Noted lack of comparable data outside the U.S.; emphasized need for systematic data collection.
Hector de Rivoire (Microsoft)	Described Microsoft’s “Economic Graph” and privacy‑preserving telemetry. Highlighted a 2025 usage report (≈37 M AI‑assistant conversations) that differentiates task‑type, intent, and sector.
Pamela Mishkin (OpenAI)	Stressed the global‑north / global‑south usage divide (24 % vs 14 % of working adults). Advocated for concrete, locally‑relevant workflow studies (public service, BPO, healthcare) to guide policy.

3.2 Audience Q&A – Key Exchanges

Question (summarised)	Respondent(s)	Core Insight / Recommendation
India’s DPI (digital public infrastructure) to Global South	Ravi	India is already sharing frameworks; 60 + countries expressed interest after G20 2023. Emphasised “humility” – governments admit limited technical understanding and thus must adopt an experimental, data‑driven approach.
Social safety nets & reskilling relevance	Ravi & Thigo	Safety nets must be redesigned for AI‑driven shocks; reskilling must be linked to what skills are needed—data‑driven labour market forecasts are essential.
Balancing AI’s cultural impacts (multilingual models)	de Rivoire	40 % of training data is English; need to increase non‑English data to improve performance & safety. Local language models and community‑driven safety benchmarks (e.g., Hindi, Tamil) are crucial.
Red lines for labour resilience	Panel (various)	No single “red line”; instead adopt scenario planning (worst‑case, median, best‑case) and embed trigger‑based policy tools (e.g., automatic safety‑net expansion).
How to obtain actionable AI‑usage data	de Rivoire	Advocate for privacy‑preserving, aggregated telemetry (the “Economic Graph” style) that governments can consume without compromising individual privacy.
Concrete data‑collection pilots for the Global South	Mishkin	Identify 2‑3 high‑impact workflows (e.g., healthcare triage, public‑service chatbots, BPO support) and systematically measure AI’s effect on productivity, substitution, and job quality.
Kenya AI Skilling Alliance – focus‑setting	Thigo	Whole‑of‑country approach: (1) industry informs government on AI trajectories; (2) build national labour‑market data infrastructure; (3) invest in lifelong learning & safety nets, especially for the ageing farmer cohort.
India’s education‑to‑employment committee – data needs	Ravi	Leverage private‑sector placement platforms (e.g., Naukri.com) for real‑time labour‑market signals; combine with government skill‑gap analyses; prioritize AI‑ready curricula for teachers and health workers.

3.3 Panel Consensus

• Data is the linchpin – granular, task‑level, and privacy‑preserving usage metrics are required to design effective transition policies.
• Policy must be iterative – experiment, monitor, and adjust as new data arrive; avoid large‑scale, static reskilling programmes detached from labour‑market realities.
• International coordination – shared standards for AI‑usage reporting (similar to the Economic Graph) can reduce north‑south gaps.
• Cultural & linguistic relevance – AI tools must be adapted to local languages and societal norms to avoid harmful side‑effects (e.g., privacy breaches, cultural insensitivity).

Key Takeaways

• Exposure vs. Adoption: Empirical work shows a 16 % employment decline for young workers in AI‑exposed U.S. jobs, while firm‑level adoption metrics produce mixed employment effects; the discrepancy stems from anticipatory hiring cuts, measurement limitations, competitive pressures, and macro‑economic confounders.

• Global Inequality Risk: Advanced‑AI access is concentrated in two countries, creating a dual disadvantage for nations lacking reliable AI‑usage data—both reduced wealth generation and heightened unemployment risk.

• Data‑Driven Policy: Accurate, task‑level, privacy‑preserving AI‑usage data (e.g., Microsoft’s Economic Graph telemetry, OpenAI’s workflow pilots) are essential for (a) anticipating labour displacement, (b) targeting reskilling, and (c) designing safety‑net triggers.

• North–South Usage Gap: Only 14 % of working‑age adults in the Global South regularly use generative AI compared with 24 % in the Global North, a gap that is widening and must be addressed through inclusive data collection and capacity‑building.

• Sectoral Opportunities in Emerging Economies: AI can accelerate growth in ag‑tech, fintech, health‑tech, ed‑tech (India) and MSME productivity, food‑system efficiency (Kenya); policies should capture these upside potentials while mitigating displacement.

• Multilingual & Cultural Safeguards: A large share of model training data is English; expanding non‑English corpora and co‑creating safety benchmarks with local communities are necessary to prevent cultural mis‑steps (e.g., misuse of medical imaging).

• Whole‑Country, Whole‑System Approach (Kenya’s AI Skilling Alliance): Combine industry insight, national labour‑market data infrastructure, and lifelong‑learning investments to create a resilient AI‑enabled workforce.

• Scenario‑Planning over “Red Lines”: Governments should develop trigger‑based policy playbooks (worst‑case, median, best‑case) rather than seeking a single immutable rule, ensuring readiness for rapid AI‑driven shocks.

• International Collaboration: Bilateral agreements (e.g., Germany‑Canada ethical‑AI pact) and multilateral forums (AI Impact Summit, G20) provide models for data‑sharing frameworks and coordinated governance.

• Immediate Action Items:

  1. Standardise privacy‑preserving AI‑usage telemetry across major tech firms.
  2. Launch pilot studies on 2‑3 high‑impact workflows in the Global South to generate baseline impact metrics.
  3. Integrate AI‑usage signals into existing labour‑market information systems (e.g., India’s Naukri.com data, Kenya’s informal‑sector surveys).
  4. Invest in multilingual model development and local safety‑benchmarking collaborations.

These takeaways underscore that robust, inclusive AI‑usage data are the cornerstone for building labour‑market resilience as AI reshapes economies worldwide.

See Also:

• ai-and-the-new-frontier-for-economic-progress-linking-innovation-to-inclusive-growth
• thriving-with-ai-human-potential-skills-and-opportunity
• policies-for-social-and-economic-resilience-in-the-ai-age-global-south-perspectives
• ai-for-industries-resilience-innovation-and-efficiency
• reskilling-for-tomorrow-ai-sustainability-and-indias-jobs-transition
• keynote-2
• ai-governance-in-the-age-of-powerful-ai-iinternational-perspectives-and-the-code-of-practice
• practical-aspects-of-using-ai-in-daily-life
• ai-impact-forum-democratising-ai-resources
• ai-beyond-moonshots-a-playbook-for-many