The industrial world is currently obsessed with the transformative power of Generative AI. From predictive maintenance to supply chain optimization, the promise of "faster and cheaper" is intoxicating. But as we move into the high-stakes arena of Machinery Safety (Functional Safety), we need to have a sober conversation about where the algorithm ends and human accountability begins.
Table of content
A recent discourse in the technical community (notably highlighted by Automatyka B2B) raised a critical red flag: AI is an exceptional assistant but a reckless decision-maker. In the context of global standards like ISO 13849-1 or the new EU Machinery Regulation (2023/1230), the limitations of AI aren't just technical - they are existential.
The "hallucination" liability
Large Language Models (LLMs) are probabilistic, not deterministic. They don’t "know" the law; they predict the next most likely word in a sentence. In a field where a decimal point error in a safety distance calculation $(S = K \times T + C)$ can mean the difference between a "near miss" and a fatality, "probabilistic" isn't good enough.
We’ve seen AI confidently cite harmonized standards that don't exist or conflate Type-C standards with general requirements. For a global manufacturer, acting on a "hallucinated" safety recommendation isn't just a technical error - it’s a fast track to a massive product recall and a shattered reputation.
The black box vs. The transparent audit
One of the cornerstones of engineering is traceability. If a safety system fails, we must be able to trace the logic back to its origin. AI, particularly deep learning models used in vision systems, often suffers from the "Black Box" problem.
As Dr. Kate Crawford, a leading researcher on the social implications of AI, famously argued: "AI is neither artificial nor intelligent. It is made of data, and it reflects the biases and gaps within that data". If an AI-driven risk assessment misses a "crushing hazard" because it wasn't explicitly labeled in its training set, who is at fault? The engineer who clicked "Generate," or the company that provided the data? In the eyes of the law, the answer is always the human.
Context is the "ghost in the machine"
Safety is a tactile, sensory discipline. An experienced Safety Engineer doesn't just look at a CAD drawing; they walk the shop floor. They notice the "workarounds" - the sensors taped over by frustrated operators or the emergency stops blocked by pallets.
AI lacks "Tacit Knowledge" - the intuitive understanding of human behavior and environmental nuance. It doesn't know that a specific shift team tends to bypass guards during cleaning cycles. It sees the world as a series of static inputs, whereas safety is a living, breathing ecosystem of human-machine interaction.
The signature: The ultimate human bottleneck
In the UK, the EU, and North America, the legal framework for machinery safety relies on a "Responsible Person". This is a human being who signs the Declaration of Conformity or the UKCA/CE mark.
An algorithm cannot be sued. It cannot be held in contempt of court. It cannot feel the moral weight of a workplace accident. When we delegate the judgment of safety to an AI, we create a "Responsibility Gap." We are essentially asking a machine to make a moral calculation it is incapable of understanding.
Moving forward: Augmentation, not replacement
Does this mean we should banish AI from the safety department? Not at all. AI is a world-class "Knowledge Librarian". It can:
- Scan thousands of pages of log files to find patterns in E-stop activations.
- Summarize complex global regulations to help engineers stay updated.
- Draft the "boilerplate" sections of technical files.
But the final review must remain human. We must treat AI-generated safety reports with the same skepticism we would show a first-year intern: useful effort, but requiring 100% verification.
At our firm, we believe the future of Industry 4.0 is Human-Centric. Technology should empower the engineer, not replace their intuition. As we integrate smarter tools, let's remember: the most sophisticated safety component ever created is the human brain. Let’s keep it in the loop.
