The Rise of AI in Public Services, Part II — Navigating Trust, Ethics, and Regulation

As a part II in my series on artificial intelligence (AI) and its entry into the public sector, I want to focus on how AI continues to reshape industries and reinforce the fact that its disruptive power is both inevitable and transformative.

Governments are increasingly exploring AI’s potential to automate processes, improve efficiency, enable better ways to improve sustainability, and tackle pressing global challenges like climate change. While the primary driver for the public sector is often cost, they — more than any other — must balance the promise of innovation with the need for trust, ethical considerations, and robust legal frameworks.

Trust and Legislation

The public sector operates on a foundation of trust — citizens expect fairness, transparency, and accountability. While human errors in public services are often forgiven as part of the system, AI-driven mistakes are seen as systemic flaws, creating public scepticism. This double standard underscores why building and maintaining trust is critical for AI adoption in public services — technology the public sector cannot afford not to use.

Today we trust AI to give us information, but we do not trust it to make decisions for us. While this may change in the future as the technology improves and humans adjust, the trust issue will always remain prominent and valid in one way or another.

Frameworks like the EU AI Act reflect this need, for example article 14, which prohibits fully automated decision-making without human oversight. This safeguard not only addresses the ethical implications of AI but also aligns with the public sector’s role as a steward of citizen welfare. However, it also means governments must carefully integrate AI technologies to complement and as of now, not replace, human judgement.

The Two Faces of Trust: Transactional vs. Relational

While the technology in the near future will theoretically be able to replace vast numbers of knowledge workers across industries, both private and public, trust becomes the critical barrier to adoption. It is not just about whether AI can perform the tasks — it is about whether people and organisations trust it to do so responsibly, ethically, and reliably.

Trust is, at the end of the day, all about belief, and trust comes in many different shapes and forms. You have transactional trust and relational trust, and the distinction between the two is critical when discussing trust in the context of AI and the public sector. Let me briefly explain the two.

Transactional trust is based on efficiency, predictability, and the ability to meet specific expectations in a given transaction. Let us build on the example in my first article — and let us stick with the chatbot and virtual assistant.

An AI chatbot that provides accurate information builds transactional trust by delivering what is expected in a timely manner. However, if that same AI fails or gives incorrect information, this trust is broken, and in high-risk domains, it leaves no room for recovery.

Relational trust is built over time through empathy, transparency, and the ability to respond to individual needs and failures in a human, understanding way. Its characteristics emphasise connection, communication, and mutual respect. It is resilient and can withstand mistakes if the response to failure is relational and empathetic. This kind of trust takes time to develop but fosters long-term loyalty.

If an AI-powered system used in public services works alongside human caseworkers, and the AI makes a mistake, the caseworker’s empathetic handling of the issue helps preserve or even strengthen relational trust.

AI systems excel at building transactional trust by being efficient and reliable, but they struggle with relational trust because they lack emotional intelligence and the ability to adapt to complex, human-centric situations.

Citizens expect more than just transactional trust from public institutions — they demand relational trust, which AI alone cannot easily provide.

Trust, therefore, is not just a byproduct of effective AI — it is a prerequisite. Without it, the societal resistance to replacing human knowledge workers will outweigh the theoretical benefits, especially in domains where discretionary decisions and personal interactions play a critical role.

The Strength of Deliberate Slowness

I’ve always been an avid believer in the classical adage and oxymoron; Festina Lente, meaning “make haste slowly”. A year into my tenure as a public servant, I shared a few frustrations with a colleague of mine in our ministry regarding the long and tedious bureaucratic processes surrounding the need to change regulations. The advice she gave me was simple:

“When it comes to changes in laws and legislation, you shouldn’t worry when things take a long time — it’s when things are moving really fast that you should start to worry.”

As I am writing this, I am aware of how fast the technology evolves, and that in six months we may see AI that outperforms the human mediator. I do not question the capabilities of the technology or its exponential development. However, technology’s possibilities and potential uses exist at the mercy of the speed at which regulations and legislation are developed — two systems that often operate on vastly different timelines.

Navigating the Trust Barrier

While technology evolves rapidly, driven by innovation and market demand, regulations and legislation are designed to govern according to premises rooted in caution, ethics, and societal impact.

This mismatch creates a tension where the public sector must navigate the gap between technological capabilities and the frameworks that ensure their safe and equitable deployment.

Governments are tasked with ensuring that innovation does not outpace accountability, yet the deliberative nature of democratic processes can sometimes make these safeguards seem slow or reactive. However, this slowness can also be a strength, offering time for thoughtful consideration and minimising the risks of unintended consequences.