Most industrial AI programs are drowning in data but starved of meaning. That is the sharp argument Andreas Heine makes in his recent piece: after decades of IIoT work that perfected connectivity, factories now have thousands of tags flowing into their data platforms — and many of those tags are unintelligible to the very AI systems meant to use them.
A tag like DB100.DBW12 = 42 carries no context: no indication of what it measures, which machine it belongs to, or whether it is safe to change. An AI asked to reason over such data faces the same problem an engineer would when handed a spreadsheet of raw register numbers — it has to go hunting for documentation. Heine contrasts this with self-describing values like JobState = Running, from which a language model can immediately infer the existence of jobs, lifecycle states, and production context.
The bottlenecks have shifted over the years — from connectivity, to integration, to interoperability — and the next one is understanding. The factories that win with AI will not necessarily be the ones with the most data, but the ones with the most understandable data: contextualized, standardized, traceable, and semantically consistent.
He also draws a useful distinction between exposing raw values and exposing capabilities. A machine interface offering Start(), Pause(), Resume(), and JobState is far easier for an AI to work with than one exposing StartBit and RecipeRegister. In his framing, the future factory is self-describing: systems that can explain who they are, what data they provide, and what they can do.
Source: Andreas Heine via LinkedIn
Read the full article here: https://www.linkedin.com/pulse/why-tags-killing-industrial-ai-andreas-heine-iswge/?trackingId=avg%2FIr67Ta2fBCpg2K4yZQ%3D%3D
Why it matters to Intelligent ShopFloor readers: Every AI initiative we run lives or dies on semantic context. As we connect equipment and build unified namespaces, designing for meaning — not just tags — is what turns data volume into decisions. This piece makes the case clearly enough to share with any stakeholder still asking for “more data.”