Industrial intelligence moves from tacit know-how to physical action

MIT Technology Review described industrial intelligence as a manufacturer’s accumulated know-how, operating practices, production data and understanding of its own business. Some of that intelligence is documented; some sits with experienced workers. An operator restarting a line after a changeover may follow a procedure that sets out the sequence, but successful execution also depends on knowing how a material behaves, what happened on the previous shift and when a minor variation needs closer inspection. The opportunity, the article said, is to make more of that intelligence available wherever decisions are made and to connect it to the physical world of manufacturing. It argued that the strategic destination is not more agents but turning existing business intelligence into an engine for profitable growth.

Expertise has limited reach when it stays with one person, inside one system or within one plant, according to the MIT Technology Review article. Leaders must extend that reach without stripping away the context that makes knowledge useful. An engineer evaluating a process change needs not only a production record but also the equipment configuration, quality requirements and trade-offs an operator made. Connecting those elements can help people apply experience beyond where it originated, spreading innovation to maintenance, planning, engineering and production teams.

Physical AI, as described by MIT Technology Review, refers to systems that can perceive conditions, reason about them and take action in the physical world. By connecting machines and processes, manufacturers get a clearer view of the factory floor; agents can then identify patterns, predict problems and determine a course of action. In a robotic-handling example, vision assesses the position of a part, a model informed by material and task determines an appropriate movement, and connected controls execute an adjustment within validated limits. The value is not merely seeing variation but responding appropriately.

In quality operations, Toyota Industries connected previously fragmented paint shop data, giving engineering and operations teams context to understand defect drivers. Toyota Industries reports that analysis cycles fell from five days to under four hours, while its pilot reduced relevant paint defects by about 25%, according to MIT Technology Review. Data may show what is happening, the article said, but industrial intelligence provides context for what should happen, which constraints cannot be breached and when uncertainty requires human judgment. Simulation can let manufacturers examine proposed actions before applying them to equipment, though physical validation remains necessary. Responsible deployment begins with operational discipline: clear authority over what systems may change, reliable equipment connections, defined safety boundaries and a record of decisions and actions. People must be able to recognize uncertainty, intervene and bring equipment to a safe state.

TechRadar focused on the same problem at the frontline, citing polymath Michael Polanyi’s observation that people can know more than they can tell. Frontline workers hold tacit knowledge about workarounds, exceptions, localized efficiencies and situational problem-solving that never makes it into an official manual. TechRadar said real-life experience and multiple academic studies indicate that up to 85% of company knowledge is tacit. Traditional top-down standard operating procedures fail to capture it, and such knowledge can be unevenly distributed, gatekept or lost when experienced employees retire or leave.

AI knowledge agents trained on standard operating procedures and other onboarding and training materials are bringing long-overlooked knowledge gaps and outdated guidelines to the surface, according to the TechRadar article written by Zebra Technologies’ VP for Automation, EMEA. Instead of relying only on centralized teams to write procedures, the article described a bottom-up approach in which workers flag inconsistencies, gaps and workarounds and AI assistants route those signals into automated updates for training programs and procedures. That can turn undocumented frontline expertise into structured micro-learning and support continuous “everboarding” rather than one-off onboarding.

TechRadar said modern AI agents are multimodal and can receive and output audio, text, image and video. A worker could verbally highlight a gap or photograph a shelf, with the agent comparing the image to a visual merchandising planogram and providing corrective instructions. Smaller models make on-device inference possible, keeping data on the device or company network, reducing latency and adding security, according to the article. It also cited an MIT report finding that businesses partnering with AI providers have a 66% AI success rate, compared with 33% for those relying on internal development alone.