AI literacy
The duty that applies to everyone and is most often overlooked: what is required, for whom, and how it can be evidenced.
What is required
Providers and deployers ensure that their staff and other people operating AI systems on their behalf have a sufficient level of AI literacy, taking into account technical knowledge, experience, education and the context of use.
What that means in practice
| Group | What matters |
|---|---|
| Everyone with access | What the tool does, what it does not, what must not go in |
| Business owners | Limits, typical errors, review steps, reporting routes |
| People exercising oversight | How to judge an output and when to reject it |
| Procurement and legal | Roles, classes, duties, contract requirements |
| Leadership | Responsibility, risk, evidence duties |
A programme that holds
- 01
Foundation for everyone
One hour: what the tools do, the two rules, where to ask, how to report a problem. Tied to the approved tools.
- 02
Depth per role
For business areas with their own tasks: the typical errors of that task, the review steps, the metrics.
- 03
Oversight separately
Anyone exercising oversight needs more: knowledge of the system's limits, automation bias, the authority to stop.
- 04
Refresh on a trigger
New tool, new version, new task. Not annually out of habit.
The evidence
- Who was trained, in relation to which system?
- Which content, in which version of the material?
- When, and when was it refreshed?
- How was it established that the competence works? For instance through a short applied exercise.
The last point is what separates evidence from an attendance list.
Content that holds
A programme that discharges the duty and is useful covers six points:
- What the system does and does not do. Tied to the concrete tools, with examples from your own organisation.
- Typical error patterns. Invented statements, agreement bias, outdated states, false premises.
- What must not go in. A concrete list, not an abstract category.
- How review happens. Which step, by whom, to what depth.
- How to report. The route when something has gone wrong, expressly without penalty.
- What the law requires. Role, class, your own duties, at the depth the role needs.
The link to the oversight duty
AI literacy and human oversight hang together: oversight without competence is not oversight. Anyone designating a person as exercising oversight must be able to evidence that they know the system's limits and can judge the output. The designation alone does not carry.
Conversely: someone expected to review an output but given no time for it cannot discharge the duty. Process design is therefore part of discharging the duty, not its surroundings. See Duties by role and Training and enablement.