Five AI buzzwords in China's latest government work agenda - and what public leaders should do next
China's new government work agenda puts AI at the center of the smart economy. The message is clear: productivity, public services, and security will run on data, models, and compute - with strong governance baked in.
Here are five terms you'll hear often, what they mean in practice, and how to move from slides to implementation inside government.
1) "AI+" across sectors
AI+ is short for "AI applied to X." It's a directive to infuse AI into priority industries and public services - manufacturing, healthcare, agriculture, finance, transport, and city operations.
Why it matters for government: Budget, procurement, and regulation will tilt toward projects that show measurable impact on throughput, cost, and service quality.
- Stand up an AI+ project list tied to national/municipal KPIs (e.g., clinic throughput, port dwell time, energy load balancing).
- Create shared components: APIs for identity, payments, and data access so agencies don't rebuild the same pipes.
- Mandate outcome metrics per project: baseline, target, and audit plan before funding release.
2) Foundation models ("large models")
These are general-purpose models (text, vision, speech) fine-tuned for domains like law, tax, customs, and emergency response. Expect multi-modal models to become standard interfaces for staff and citizens.
Why it matters for government: One core model can support many use cases if you control data, safety policies, and fine-tuning.
- Adopt a "shared model, many apps" approach: one vetted base model, multiple domain adapters.
- Set red-team, eval, and logging policies before deployment. Align with risk tiers by function (advice vs. decisions).
- Prioritize retrieval-augmented generation so outputs cite official sources and reduce errors.
3) Computing power and data infrastructure
Compute clusters, data centers, and high-bandwidth networks are now public utilities. So is high-quality, consented, and well-labeled data.
Why it matters for government: Capacity planning, data liquidity, and cross-region routing become as important as roads and power grids.
- Build a compute and storage registry: who has what capacity, at what cost, under what security class.
- Publish data catalogs with access policies by sensitivity level; standardize schemas for health, land, logistics, and SME credit.
- Use usage-based budgets for training/inference to stop over-provisioning and reveal real unit costs.
4) Industrial AI and smart manufacturing
Computer vision, predictive maintenance, process control, and digital twins are moving from pilots to line-wide deployment. The goal: higher yield, lower energy use, safer operations.
Why it matters for government: Grants and tax incentives will favor measurable OEE gains, energy intensity cuts, and safer factories.
- Require factories to submit before/after metrics and model audit logs with every subsidy application.
- Create shared "model playgrounds" using synthetic and de-identified data so SMEs can test before capex.
- Tie export finance and park admissions to interoperable standards for sensors, IDs, and maintenance data.
5) AI governance, safety, and standards
Trust will decide scale. Expect stricter testing, content controls, incident reporting, and lifecycle management of models.
Why it matters for government: Clear rules lower vendor risk and speed procurement.
- Adopt a risk-based framework with tiered controls. Reference global tools like the NIST AI RMF and the emerging ISO/IEC 42001 AI management system.
- Stand up an AI incident registry and require vendors to report model failures and mitigations.
- Bake evaluation and monitoring into contracts: bias tests, jailbreak tests, and post-deployment drift checks.
Policy moves you can execute this quarter
- Issue an "AI+ Priority List" with 5-7 projects tied to concrete KPIs and a public dashboard.
- Create a central model evaluation service agencies must use pre-procurement.
- Launch a secure data exchange with common schemas for three high-value domains (e.g., health referrals, port logistics, urban mobility).
- Switch AI budgets to milestone-based disbursement triggered by audited outcomes, not demos.
- Publish a short AI safety directive: risk tiers, human-in-the-loop rules, logging, and incident reporting.
Capability building for public leaders
Policy only sticks when teams have the skills. Give your staff a path from awareness to deployment.
- Upskill policy and operations teams with the AI Learning Path for Policy Makers to standardize risk, procurement, and measurement practices.
- For implementation leaders, see AI for Government for governance templates, vendor checklists, and case studies.
Final note
AI+ is no longer a slogan. Pick a few critical services, define the metric that matters, and ship. Tight feedback loops - policy, infrastructure, governance, delivery - will separate pilots from programs that move the economy.
Your membership also unlocks:
