Recognise real-world forms: chatbot answer

• Classify: assign a category, such as spam or not spam.
• Predict: estimate a future value, such as likely demand.
• Rank or recommend: order options, such as catalogue titles.
• Generate: create text, images, audio, or code.
• Control: choose actions in a changing environment, such as a robot vacuum changing direction.

A product can combine families. Harbourview’s chatbot might classify a question, retrieve documents, and generate an answer.

Mira maps three proposed uses:

1. Opening-hours answers: sourced from a maintained page; easy to verify; low consequence.
2. Book recommendations: subjective; omissions matter, but patrons can ignore the list.
3. Benefits advice: rules change and mistakes may affect essential support; authoritative referral is required.

The interface looks the same, but the tasks and stakes differ. Mira approves a limited pilot for opening hours, allows clearly labelled recommendations, and rejects unsupervised benefits advice.

“The chatbot was accurate in our demo, so it is accurate.” A demo is one sample. A useful evaluation includes common questions, ambiguous wording, missing information, outdated pages, and requests the system should decline.

Classify each library feature by its main family: image-based book sorting, attendance forecasting, a writing assistant, and a robot vacuum. Some have more than one defensible answer. Explain the task boundary, not just the family name.

Find one AI feature in an app you use. Record its family, input, output, first likely failure, and how a person can recover. If the product does not disclose enough, state what remains unknown.