Without the Rules of the Road: Why Industrial AI Needs Governance to Leave the Pilot Phase

At the Singaporean-German Chamber of Industry and Commerce (SGC) Forum, industry leaders explored what happens when artificial intelligence steps beyond chat prompts to operate directly on physical shop floors, control loops, and automated production lines.

Opening the forum at the Mercedes-Benz Center in Singapore, Marcel Luis Mustelier Perez, President and CEO of Mercedes-Benz Singapore, drew an astute parallel between today’s AI wave and the dawn of the automotive industry. When Karl Benz patented the first automobile 140 years ago, critics dismissed the noisy, fragile contraption as a novelty compared to the proven reliability of horses.

Real disruption did not occur simply because the internal combustion engine worked. It happened when society built the roads, fuel infrastructure, traffic signals, driver licensing, and – critically – harmonized safety regulations and traffic laws that allowed vehicles to operate reliably and safely across municipal and national borders.

Without the rules of the road, the car was just a dangerous curiosity. That historical parallel defined the SGC Forum, “Riding the Wave – Best Practices to Navigate Global Shifts: Industrial AI for Competitiveness.” While enterprise discussions across MediaBUZZ frequently spotlight foundational models and conversational tools, this forum shifted the focus to the physical realm: machines, sensors, industrial safety, and legacy factory floors.

The clear consensus: industrial AI is not another software procurement exercise. It is an end-to-end capability challenge requiring deep integration across operational data, industrial protocols, security boundaries, and human workflows.

The Next AI Race is About Capability

left2857500In his keynote address, Tan Kiat How, Singapore’s Senior Minister of State for Digital Development and Information, stated that the competitive axis of AI has shifted. As computing clusters and foundation models commoditize, value accrues to whoever puts intelligence to work inside the real economy.

Manufacturing throws that challenge into sharp relief. An impressive model cannot transform a factory in isolation – it must ingest sensor telemetry, interface with legacy operational technology (OT), and augment experienced floor engineers.

The bottleneck invariably occurs after the proof of concept. Moving from an isolated pilot to plant-wide productivity requires redesigning workflows, sanitizing operational data, and retraining technicians. As SMS Tan framed it:

The winners will not necessarily be those who have the most AI. They will be those who become the most capable with AI.

Moving from Copilots to Autonomous Agents

left56515Dr. Thai-Lai Pham, CEO of Siemens ASEAN, framed this transition historically, comparing industrial AI to electrification 150 years ago – a once-in-a-century force reshaping physical production.

Moving beyond theory, Dr. Pham pointed to tangible operational transformations. In automotive chassis engineering, high-fidelity digital simulation reduces physical prototype iteration from 15 days down to half a day. Through partnerships like Siemens and NVIDIA Omniverse, industrial digital twins allow manufacturers to simulate entire factory floor mechanics, robotic kinematics, and human-machine interaction long before equipment is ordered.

More significantly, Pham highlighted the evolution from conversational copilots to autonomous agents. Instead of just offering suggestions, specialized engineering agents are now being deployed to write, validate, and execute Programmable Logic Controller (PLC) code directly. With Siemens PLCs driving roughly a third of the world’s automation lines, generating logic code autonomously circumvents critical engineering labor shortages and turns AI agents into active co-workers.

Laying the Industrial Foundation

left1905000The subsequent panel brought the focus to physical execution:

Infrastructure & Architecture: George Aprane, Managing Director at Deutsche Telekom, emphasized that industrial AI cannot be bought off the shelf. Pilots often collapse during enterprise rollout because underlying cloud infrastructure, network latency, data hygiene, and governance were treated as an afterthought.

Adaptive Machine Vision: Chin Hui Ho, Managing Director of SICK Singapore, illustrated how embedded AI transforms quality inspection. While traditional machine vision required rigid, deterministic rules and uniform lighting, modern AI vision handles organic variation – such as sorting raw rubber or detecting non-uniform defects – without requiring continuous reprogramming.

Demand-Pull Over Tech-Push: Jörg Menner, SVP Southeast Asia & Pacific and MD Singapore at Festo, argued that traditional engineering “tech push” fails with AI. Scaling requires turning that approach 180 degrees: starting strictly from customer pain points and tangible ROI before selecting the technology stack.

The Deterministic Security Paradox

As autonomous agents gain the authority to trigger physical actuators, industrial cybersecurity faces a fundamental challenge.

A central tension raised during the session centered on the mismatch between probabilistic AI systems and deterministic industrial environments. Factory floors operate with zero tolerance for unpredictable variance. The panel’s verdict: industrial AI must be architected from day one within zero-trust boundaries, segregated OT networks, and isolated cleanroom environments – ensuring that machine learning models cannot inadvertently trigger unverified physical actions.

Singapore as the Proving Ground

The forum underscored a natural synergy: combining German industrial depth with Singapore’s agile, highly connected ecosystem.

SMS Tan urged European industrial leaders to bring their most difficult, messy operational problems to Singapore – challenges where AI developers must collaborate directly with plant operators and real production telemetry. By validating solutions in Singapore’s controlled proving ground, enterprises can scale tested industrial AI frameworks across Southeast Asia’s wider manufacturing footprint.

Setting the Industrial Rules of the Road

That leaves a lingering question in today’s fractured geopolitical climate: who will establish the harmonized regulations that industrial AI requires?

In a fragmented global economy, expecting a single, universal treaty is unrealistic. Instead, the “rules of the road” are being forged from the ground up:

• Technical interoperability: Global standards bodies (such as ISO, IEC, and IEEE) are defining functional safety and AI management standards (like ISO/IEC 42001) that engineers on every continent can audit against.

• Bilateral regulatory sandboxes: Corridors like Germany and Singapore are bridging the gap – combining Europe’s rigorous risk frameworks (like the EU AI Act) with Singapore’s agile, trusted testing environments (such as AI Verify) to build frameworks that work seamlessly across ASEAN supply chains.

Much like the transition from the earliest horseless carriages to standardized highway networks, industrial AI is moving past its novelty phase. The next decade belongs to the organizations, engineers, and policymakers who lay the physical infrastructure, operating models, and regulatory guardrails required to make intelligence dependable on the factory floor.

By Philipp Bonkatz

General Manager, SP-Asia