How to use this section
Artificial intelligence becomes easier to evaluate when it is broken into ordinary components: a task, examples or rules, a model, an interface, an output, and a person or organization responsible for using it. This section builds that mental model before introducing product-specific features.
Begin with the task
Ask whether the system predicts, classifies, ranks, generates, plans, or controls. The word intelligence is too broad to explain what a product actually does.
Follow the information
Data is selected and represented, a model applies learned parameters, and an application turns the result into something a person can see or act on. Each stage can introduce error.
Separate models from systems
A model does not manage permissions, retrieve approved records, keep reliable logs, or take responsibility. Those functions belong to the larger software and human system.
Keep future claims separate
Today’s specialized systems can be impressive without being human-like general intelligence. Forecasts about future capability should be labelled as scenarios, not current facts.