A Civil Service Transformed: The Case of Hong Kong

Hong Kong is currently conducting one of the most significant experiments in applying Artificial Intelligence (AI) within the civil service. The aim: to increase government efficiency and address a growing fiscal deficit. According to a report by CNA on February 26, 2025, the city plans to leverage AI to manage a major civil service restructuring effort.

By April 2027, Hong Kong plans to cut around 10,000 civil servant positions — reducing approximately 2% of staff annually. These reductions are part of a strategic push to trim government spending while maintaining, or even enhancing, public service quality through digital transformation. AI is expected to shoulder some of the workload left behind. For example, the Census and Statistics Department is already using AI to handle verification tasks previously done manually.

To support this shift, Hong Kong has committed over HK$11 billion (approx. US$1.4 billion) in AI innovation and digital transformation funding. This includes a HK$1 billion allocation for R&D institutions and a HK$10 billion innovation and technology fund targeting strategic future industries.

A Global Pattern: AI as Evaluator, Not Just Executor

This ambition mirrors a broader global pattern. In Indonesia and across the Global South, artificial intelligence is no longer a distant buzzword. It is quietly reshaping the public sector — not just by automating tasks, but by evaluating the very people behind them.

Civil servants in several pilot regions are now being rated by AI systems based on data traces: collaboration metrics, email patterns, task outputs. These scores are then used to “recommend” which roles are redundant, inefficient, or low impact.

This echoes trends around the world. In the United States, the Department of Energy’s Office of the Inspector General (DOE OIG) has tested AI to flag anomalies in procurement and performance. In South Korea, AI has been trialed to detect underperformance in public health roles. Across parts of Africa and Southeast Asia, donor-funded projects use algorithmic scoring to evaluate local staff performance for continuity.

The Distorted Lens of Efficiency

On the surface, this sounds fair. After all, who wouldn’t want a government that works better?

But look deeper, and the danger reveals itself.

AI is not just a tool. It is a lens. And any lens distorts reality based on how it was shaped — by whom, for what purpose, and with which blind spots. In the name of objectivity, we risk building systems that reproduce the very inequalities we failed to fix manually.

The real question is not: “Can AI detect inefficiency?”
It is: “Who defines efficiency? And who benefits from its definition?”

Jobs with emotional, preventive, or contextual value — often held by women or marginalized communities — rarely register well on digital data. Loyalty and discretion, the backbone of many silent roles in diplomacy or social cohesion, are invisible to algorithms. The AI sees output. But not intention. It scores impact. But not nuance.

A Looming Social Risk in the Global South

Beyond governance concerns, there are critical social risks, especially in developing nations. The displacement of human workers by AI can exacerbate unemployment, particularly where alternative job opportunities are scarce. The digital literacy divide means many workers may not have the skills to transition into new roles that require AI fluency. And in countries where digital infrastructure remains uneven, the push toward AI-first public service may deepen inequality rather than bridge it.

A hopeful counterexample: Rwanda’s AI policy includes mandatory community consultations and AI literacy programs as preconditions for any government automation project. While still in early stages, this localized, participatory approach reflects an awareness of both technical and social impact.

Governance That Protects Human Dignity

Worse, the introduction of AI in bureaucratic job assessments often lacks three critical governance pillars:

Explainability – Can employees understand why they are marked “low value”? Or are they just shown a score?

Human-in-the-loop decision-making – Is there room for compassion, second chances, or clarification before action is taken?

Public transparency – Who audits the system? Who sets the parameters? And is the public informed?

Without these guardrails, AI becomes not a tool for reform — but a tool of quiet elimination. You are not fired. You are “scored out.”

In Global South contexts, this is particularly risky. Power is often personalized, and resistance to automation is framed as “anti-progress.” The pressure to adopt AI for prestige, for cost-cutting, or donor appeal creates a climate where ethical reflection is deemed a luxury.

But dignity is not a luxury.

Contextual Governance, Not Imported Frameworks

The solution is not to reject AI. It is to govern it.

We need multidisciplinary teams to co-design such systems. Ethics officers must be embedded from day one. Auditability must be built in, not patched later. And most of all, we must recognize that governance is not just about outcomes — it is about the process of deciding what counts as valuable.

Crucially, this governance must be contextually rooted. Borrowing AI regulatory frameworks from the Global North without adaptation risks deep mismatch. Social structures, political systems, cultural dynamics, and levels of digital literacy vary widely across the Global South. Most developing countries are still primarily users, not developers, of AI — making them more vulnerable to biases embedded in foreign-made systems. If not critically assessed, these biases could further marginalize local communities under the guise of algorithmic neutrality.

At the same time, reskilling and upskilling efforts must be scaled to support those displaced by AI-driven efficiency measures. Governments, educational institutions, and industry must work together to ensure that affected individuals — especially those from vulnerable communities — can transition into meaningful roles in the evolving digital economy.

What Kind of System Are We Building?

When AI becomes a gatekeeper of human worth, our silence becomes complicity.

It is not enough to build systems that work.
We must build systems that understand why people matter.