The Smart Factory: The Promise of Never Making a Mistake, and What the Figures Say

"A factory that never sleeps, never runs late and never errs." The phrase works as an advertisement but not as a description. Our question here is: what is established about AI's effect in factories, and what remains a promise?

What AI actually does in a factory

Sensors are fitted to motors, valves and furnaces, streaming readings of temperature, vibration and pressure. Models learn from the history of those readings what a machine looks like before it fails, and flag maintenance before the breakdown instead of waiting for it to break. Computer-vision systems inspect parts on the line, and other algorithms forecast demand and redistribute stock.

What matters is that the system does not "understand" the motor as an engineer does. It knows that a certain pattern preceded failures in its history, not why. That is useful as long as the motor resembles what the system has seen, and misleading when the motor, the material or the conditions change.

What the figures establish

The clearest public source is the Global Lighthouse Network, a World Economic Forum initiative launched in 2018 with McKinsey. It numbered 189 sites in the January 2025 announcement, and the Forum reported for its latest cohort an average 53 per cent increase in labour productivity, a 30 per cent average reduction in material waste, and a 25 per cent reduction in energy and water consumption. It added that 77 per cent of the top use cases across the network relied on analytical AI and 9 per cent on generative AI.

But these figures are aggregated from the selected sites themselves and presented by the body that created the network, and we do not know how they were measured or whether they were independently audited. They indicate what the front-runners can achieve, not what happens in the average factory.

Error does not vanish; it changes form

Our own analysis, rather than a statistic, is that human error on a traditional line is distributed and visible: a worker slips, a faulty part emerges, and it is caught. A model that has learnt a wrong pattern, or been fed incomplete data, may repeat the same error across thousands of parts with complete confidence. So it is not enough to measure a model's accuracy on the day it is installed; it must be monitored periodically if materials or suppliers change.

As early as 1983, Lisanne Bainbridge pointed out in her paper "Ironies of Automation" a paradox that still stands: when most of a process is automated, the operator is left with what the system cannot do, and does not practise it in daily work, so is weak when called upon. Her conclusion is that automation requires more operator training, not less.

People and the Kingdom in the equation

The World Economic Forum's Future of Jobs Report 2025, a survey of more than a thousand employers in 55 economies, estimated that about 170 million jobs are expected to be created and about 92 million to be displaced by 2030, a net gain of 78 million. Some 40 per cent of employers said they expect to reduce their workforce where AI can automate tasks. These are employers' expectations, not settled facts, and they make the better question: which skill do we raise so that the human sits above the machine, not beneath it?

The Kingdom is working to modernise its industrial base. The National Industrial Development and Logistics Program allied with an international centre based in Singapore to promote the Smart Industry Readiness Index (SIRI), with the aim, according to a report published in July 2024, of applying it in 4,000 industrial establishments. That is a declared target, not a measured achievement. The essential point remains that our factories' data should be a wealth we know how to protect and turn into local knowledge, not a raw material we sell.

Practical rules for whoever runs a factory

• Begin with data quality, before choosing a model.
• Keep the final decision with an accountable person, and retrain staff for supervision, maintenance and judgement.
• Design a "safe mode" to which the plant returns when systems fail.
• Review models periodically and secure the systems (see the article "AI and Cybersecurity").

Our religion taught us to take the means and to do our work with excellence while relying on God; the excellent smart factory is the one that combines good use of the means with a human who stays alert, not asleep in the control room.