top-down reasoning

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

Top-down reasoning begins with goals, context and other high-level constructs and works down towards low-level sensing and data. For example means–ends analysis starts off with a goal and then works out sub-goals that if successful would enable the goal. In reality most systems need a combination of top-down and bottom-up reasoning. For example, in natural language processing expectations of what are likely to be topics of conversation can be fed top down into lower levels, say as probabiliities or vocabulary items, which can then be used by lower-level processes as they disambiguate words in speech-to-text ready for grammatical parsing (bottom up).

Used in Chap. 3: page 26

Also known as top-down