Many organisations are buying AI courses as they once bought cybersecurity courses: tick a box, file a certificate and hope the problem has disappeared. AI literacy is not a day of enthusiasm. It is the distributed ability to recognise what a system can do, what it cannot do and when it should not be used.
Competence must follow the role
Board members must understand risk, accountability and investment criteria. IT leaders must govern access, data, integration and suppliers. Healthcare professionals need to understand limitations, oversight and possible distortion. Administrative staff need to know authorised tools, shareable data and what requires verification. Giving everyone the same course is efficient only at producing certificates.
From theory to behaviour
Training becomes real when it changes a choice. Exercises must therefore use the organisation’s own processes: examining a plausible but incorrect output, recognising a request that exposes sensitive data, documenting a control and deciding when to stop automation. A final test should assess decision quality, not memory of definitions.
Learning to disagree with the machine
People tend to trust confidently presented output, particularly when it comes from a sophisticated system. Automation bias is not solved by saying “always check”. People must learn which signals to seek, have alternative sources and work within a process where challenging the output is not treated as wasted time.
A culture that continues
Models, interfaces and rules change. Training must be periodic, traceable and connected to observed incidents. Small quarterly reviews, simulations and internal communities of practice are more valuable than a large annual event forgotten the next day. Culture is not a slide. It is the behaviour that remains after the trainer leaves the room.
A map of capabilities
Every organisation should know who uses which tools, how often and for what decisions. Different learning paths follow: foundations for everyone, laboratories for those producing content or analysis, advanced work for people evaluating suppliers and specific controls for sensitive roles. Maturity is not measured by licence count but by the distributed ability to select the right tool and to avoid it when it is inappropriate.
Document in order to learn
Minor incidents, challenged outputs and bypassed procedures are valuable educational material. Hidden, they cause the same errors to return; observed without blame, they become real cases for the next learning cycle. A competent culture is not one in which nobody makes mistakes. It is one in which mistakes leave usable memory.
Technology does not replace responsibility. It makes it harder to hide.
This content is educational and is not clinical guidance. Editorial reconstructions are distinguished from documentary material.