Singapore has announced its "National AI Strategy 2.0 (NAIS 2.0)," an update to the National Artificial Intelligence Strategy (NAIS) launched in 2019, outlining an ambitious plan to position itself as a hub for AI innovation and regulation. With new support measures, implementation plans, and a focus on generative AI, NAIS 2.0 appears to be a significant milestone for Singapore's future.

However, behind the rosy outlook lurk shadows of data privacy breaches, AI bias, and the 'hallucination' phenomenon where AI presents nonsensical information as fact. In the humid air of Singapore's bustling business district, these questions hang heavy: Can Singapore's ambition outpace the massive, still untamed beast of AI?

David Irecki, Chief Technology Officer for APJ (Asia Pacific and Japan) at global software company Boomi, offered a positive assessment, saying, "I think it's a very good starting point." He analyzed, "I don't believe that the Asia Pacific region will follow strict regulations and penalties like the European Union's AI Act. Instead, APAC countries, especially Singapore, are focusing on increasing trust and reducing risks through generative AI frameworks."

The Challenge of Securing Trust in Generative AI

Trust is one of the most critical factors in AI adoption. Irecki CTO pointed out, "According to a Salesforce customer survey, 50% of respondents do not trust how AI processes information." This skepticism is not unfounded. Companies deploying AI systems without proper governance can face risks such as exposure of sensitive data, embedded biases, and inexplicable decision-making, which are potential minefields in Singapore's highly regulated business environment.

According to recent research conducted jointly by Boomi and MIT Technology Review Insights, 45% of companies are holding back on AI adoption due to governance, security, and privacy concerns. Furthermore, a staggering 98% of companies stated they would wait and see to exercise caution in managing personal data.

In this context, Singapore's approach stands in stark contrast to Europe's stringent regulations that emphasize immediate punitive measures. Singapore's framework focuses on encouraging innovation while establishing safeguards.

Irecki CTO emphasized, "The only way to ensure accountability for AI decisions is through human oversight. Until AI can act autonomously without human intervention, human involvement will be necessary."

Singapore's new framework also aligns with the 'Model Governance Framework for Generative AI (MGF-Gen AI)' announced in mid-last year. While NAIS 2.0 presents key objectives, MGF-Gen AI concretizes these by promoting trustworthy AI development and responsible innovation.

For example, initiatives such as the AI Verify Foundation and the Infocomm Media Development Authority (IMDA)'s AI Assurance pilot program are developing testing methodologies for generative AI applications, providing a crucial stepping stone for companies struggling with AI implementation.

However, Singapore faces unique challenges. As a global business hub where even SMEs operate internationally, Singapore's framework must consider cross-border issues, which are further complicated by the absence of robust regional guidelines or frameworks like the EU's.

Irecki CTO pointed out, "A model used in one country with specific language, culture, and data issues can produce very different results in another country." This poses a tricky problem: bias in Malaysia might not be bias in Thailand, and Singapore's framework needs to be flexible enough to accommodate these regional nuances.

The Need for an Agent Registry System

For companies looking to implement AI responsibly, data quality remains a core foundation. Irecki CTO emphasized, "Ultimately, what we need to understand first within the organization is the data, and we need to improve data quality to lay the groundwork for these models."

But what happens when multiple teams deploy various AI solutions without coordination? The result is often AI sprawl – the implementation of disparate systems with varying levels of governance across different departments.

An agent registry system is a centralized oversight system that tracks AI deployments across the organization. Irecki CTO explained, "The purpose of an agent registry system is to provide a synchronized view of all agents operating within the organization, monitor their activities, and ensure compliance with all frameworks."

The agent registry system, part of Boomi's broader AI Studio platform, becomes even more critical when considering what Irecki CTO calls the 'credit card swipe' phenomenon. He noted, "You might want to scale down some projects and clean up data, but someone in HR can swipe a credit card and immediately use a large language model that's readily available."

Its importance grows when considering Agentic AI, which operates autonomously and orchestrates other AI systems. Irecki CTO provided a specific example: "A master agent communicates with a human about a destination, but in the background, it automatically orchestrates and communicates with a flight booking agent, a hotel booking agent, and a car rental booking agent."

The ramifications are immense. As these systems proliferate, humans will struggle to monitor them effectively. Irecki CTO warned, "Eventually, it will operate at a scale that current tools and humans cannot handle."

The Dilemma of Accountability

Singapore's framework emphasizes accountability, but enforcing it becomes increasingly challenging as AI systems become more autonomous. Irecki CTO reiterated, "The only way to ensure accountability for AI decisions is through human oversight. Until AI can act autonomously without human intervention, human involvement will be necessary."

This ability to intervene aligns precisely with the 'kill switch' that Boomi is building into its AI governance platform – a feature that takes agents offline if inappropriate behavior is detected.

However, Singapore's framework assumes static AI models, while the reality is far more complex. Models fluctuate over time, presenting a constantly moving target for governance.

Irecki CTO stated, "Understanding model degradation due to input changes, whether it's learning from its own output, and how to correct it, is what I think we will grapple with together going forward."

In Singapore's highly regulated industries such as banking, healthcare, and transportation, the 'black box' nature of many AI models poses a particularly difficult challenge.

Irecki CTO explained, "In highly regulated markets, how you test AI is very important. We've seen customers trying to build testing policies with specific algorithm outputs within an approved range. As long as it stays within that approved range, it gets a pass on how the model works."

However, this approach hinges on the proper design of the testing model itself, potentially leading to complex and recursive scenarios where AI tests AI.

Practical Advancements

For Singaporean and ASEAN companies struggling to implement AI governance, Irecki CTO offers practical advice: start with quick wins.

"If you're looking to introduce AI into your business, find quick wins where you can get a fast return on investment." Two use cases stand out: improving chatbots through Retrieval-Augmented Generation (RAG) and document summarization.

However, regardless of the application, standardization is essential for effective governance. Irecki CTO cautioned, "Governance will be very difficult until there is standardization between agents."

As Singapore's AI strategy unfolds, it presents a middle ground between innovation and regulation. However, for companies grappling with legacy systems and the need to adopt AI, the journey remains challenging.

Irecki CTO pointed out, "Many companies are still stuck with legacy systems and outdated technologies. They are in a dual situation where they need to adopt AI to generate new revenue streams and stay competitive, but they are struggling due to legacy systems, data silos, organizational silos, and limited resources."

It is within this tension between ambition and capability that the true test of Singapore's AI strategy lies. The key will not just be creating frameworks, but helping businesses fundamentally transform.

[Copyright (c) Global Economic Times. All Rights Reserved.]