liveBook · Manning

Context

As explained in Inside AI

Context is the surrounding situation, knowledge, and circumstances that influence the interpretation of an utterance or event.

A plausible answer can still miss the point. Context is the surrounding situation, knowledge, and circumstances that determine what an utterance or event means. Contextual reasoning uses those factors rather than treating the words or data in isolation. section 5.6 section 11.21

Three birds on a branch

A hunter shoots one of three birds on a branch. Subtraction suggests that two remain, but the gunshot would likely make the others fly away. The example shows why a numerically tidy answer may fail to account for the real-world situation. section 5.6

Which details matter depends on the decision:

  • In prescribing medication during pregnancy, the stage of pregnancy, medical condition, existing conditions, alternatives, and risks can change the appropriate choice. section 11.21
  • In fraud prevention, a spending surge during a holiday or vacation may be normal; ignoring the occasion can increase false positives. section 11.21
  • In language, context can distinguish a word’s intended meaning or signal that an idiom should not be translated literally. section 11.21 section 8.1

Context also changes how evidence should be read. Wartime records showed damage on planes that returned to base, not on planes that were lost. Abraham Wald recommended reinforcing the areas with less recorded damage on returning planes: damage there may have prevented a plane from returning at all. Reading the records without considering which planes were missing would lead to the opposite recommendation. section 7.2

Pattern-finding alone does not guarantee this kind of judgment. A model may produce a fluent answer while overlooking physical consequences, or draw a dangerous conclusion from healthcare data without the relevant clinical understanding. For consequential uses, check which circumstances the model has considered and involve domain experts in validating its conclusions. section 5.6 section 10.3

How Inside AI covers it

How deep it goes
Explains contextual reasoning through applied cases and AI failures, without giving an implementation procedure.
Examples
Illustrates it with birds startled by a gunshot, damage on returning planes, pregnancy-related prescribing, and unusual spending during holidays.
What you can do after
Evaluate/understand it by checking whether AI conclusions account for relevant circumstances and seeking domain-expert validation for consequential decisions.
How it connects
Connects context to common-sense reasoning, language interpretation, clinical reasoning, fraud prevention, and domain expertise.

Also explained in

Compare how 18 books explain context →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Inside AI. The section links point to the parts of the book it draws on.