Some things are hard to explain in words: for example, how a craftsperson feels that a part has been machined correctly, or how the sound of a machine can tell you something has gone wrong. This kind of knowledge usually resides in the hands and mind of an individual and isn't passed on when they retire.
This is precisely the problem that the Japanese company ARUM is solving with the help of artificial intelligence.
Precision engineering is a field where millimeters matter. Aerospace parts, medical equipment, and electronic components – all require not just adherence to a process, but a fine-tuning that comes with years of practice.
The problem is that such specialists are becoming increasingly rare. The older generation of craftspeople is retiring, and they are unable to pass on their expertise in a clear form – not because they don't want to, but because this experience is largely non-verbal. It's an intuition built up over years.
This situation is particularly acute in Japan, a country known for its culture of monozukuri – a deep respect for craftsmanship and mastery in manufacturing. But even here, time is taking its toll.
ARUM's Solution
ARUM took an unconventional approach to this task. Instead of simply automating processes, the company focused on digitizing the knowledge itself – the intuitive actions an experienced operator takes at the machine.
Simply put: they observe how a master works, record various parameters – movements, settings, decisions in non-standard situations – and convert it all into data. An AI model is then trained on this data, enabling it to replicate a similar level of precision without a human expert's involvement.
This isn't about replacing the craftsperson. Rather, it's a way to preserve their knowledge and make it available at scale – across multiple machines and multiple factories, without any loss of quality.
Why This Matters Right Now
The manufacturing industry has long faced dual pressures. On one hand, there's a shortage of skilled personnel. On the other, demands for quality are constantly rising, especially in high-tech sectors.
AI in manufacturing is nothing new. But most solutions work with easily measurable variables: conveyor speed, temperature, pressure. ARUM has set its sights on something different – on what was previously considered fundamentally unquantifiable.
This is changing the fundamental logic of how AI is applied in the industry. Where the goal was once to «automate routine», it has now become to «preserve and scale expertise.»
In implementing its approach, ARUM relies on the Microsoft Azure cloud infrastructure and AI-based tools from Microsoft. This allows the company not only to process large volumes of manufacturing data but also to deploy trained models across different plants without having to «retrain» the system from scratch each time.
In essence, the company is building a platform: knowledge captured from one craftsperson can be applied where neither the craftsperson nor their apprentice is present.
What This Means for Manufacturing as a Whole
If ARUM's approach proves to be scalable, it could change several conventional notions about how modern manufacturing is structured.
First, the barrier to entry into high-precision manufacturing will be lower. Companies will no longer need to spend years cultivating specialists – access to a trained model will suffice.
Second, the risk of losing expertise as generations change will become manageable. This is crucial for industries where the cost of an error is high.
Third, the very process of accumulating knowledge will become more deliberate. Companies will begin to consider how to document not just process charts, but also the living knowledge of their employees – while they still have it.
Open Questions
Of course, uncertainties remain. How accurately can AI replicate what a craftsperson does in a non-standard situation – when a material behaves unexpectedly or when a task falls outside the scope of its training data?
Manufacturing is about more than just repeatability; it's also about adaptation. It remains to be seen how flexible models trained on the experience of specific people under specific conditions will prove to be.
Nevertheless, the very fact that someone is seriously tackling this task – and apparently making progress – speaks volumes. Perhaps the most valuable knowledge in manufacturing isn't what's written in instructions, but what resides within people. And now, it has a chance to outlive its human carriers.
