AI systems turn inputs into likely outputs. They can classify, transform, compare, summarise and generate. These abilities can look like human understanding, but the operating model is different.
Five useful capability groups
| Capability | Example | What still needs definition |
|---|---|---|
| Transform | Turn notes into a report | Required structure and facts |
| Extract | Find dates and owners | Source set and missing-data rule |
| Compare | Score options against criteria | Rubric and evidence |
| Generate | Draft copy or images | Audience, constraints and review |
| Plan | Propose steps toward a goal | Authority, tools and stop conditions |
AI is strongest when the input, desired output and quality standard are visible. It becomes less dependable when the task relies on hidden context, changing facts or unwritten judgement.
The practical question
Do not ask only, "Can AI do this?" Ask:
- What evidence does the task require?
- How will a good result be recognised?
- What happens if the result is wrong?
- Who is allowed to make the final decision?