Notice the boundary: recommendation list

By the end of this page, you can describe a system using four observable parts: input, operation, output, and consequence.

Harbourview Public Library adds a “You may also enjoy” row to its catalogue. Mira, the service manager, hears two reactions:

• “The catalogue understands my taste.”
• “It is just a database.”

Neither statement is useful enough to evaluate the feature. Mira asks concrete questions instead:

1. What enters the system?
2. What operation changes or compares that input?
3. What comes out?
4. What happens because of the output?

For the recommendation row, a first map is:

• Input: borrowing or browsing signals associated with an account.
• Operation: a ranking method compares those signals with patterns in catalogue activity.
• Output: an ordered list of titles.
• Consequence: a patron may choose one title and overlook others.

This map does not prove which algorithm is used; that requires product documentation or testing. It does replace “understands” with questions that evidence can answer.

A scanner also has an input, operation, output, and consequence: it reads a barcode, decodes an identifier, displays a record, and updates a loan when staff confirm it. That does not make every scanner an AI system. Many scanners follow fixed decoding rules. The recommendation feature is more likely to use patterns estimated from examples, but Mira must verify that claim with the supplier.

Misconception: “If software makes a decision, it is AI.” Correction: conventional rules, formulas, and lookups also produce outputs. Ask whether performance was learned from examples and whether it can vary beyond explicitly written rules.

Complete the missing line for the library chatbot:

• Input: a patron’s typed question
• Operation: ___
• Output: a written response
• Consequence: the patron may act on the response

A careful answer is “the system generates or retrieves a response using the question and available context.” “It thinks” is not observable. “It searches the website” is too specific unless you know retrieval is connected.

Choose a feature you used today: spam filtering, route planning, a spreadsheet formula, or autocorrect. Write its four-part map. Then label each sentence as:

• fact — directly observed or documented;
• inference — a reasonable explanation not yet confirmed;
• question — evidence still needed.

If you cannot name the operation, write “unknown” rather than guessing.