Page 31 - CyberSecurityEssentialsEnglish

recommendation. As a system moves closer to proposing or executing an action, permissions
and human review become more important.

A single application may combine several task types. A customer-service system may convert
speech to text, classify the request, retrieve the relevant policy, draft a response, and

recommend escalation when needed. To the user it may appear to be one tool, but it is
actually a chain of steps. Understanding that chain helps locate an error: did it occur during
transcription, request classification, source selection, or response drafting?

FIGURE 1-3: TASK FAMILIES BY OUTPUT TYPE

What to notice: Classification and numerical forecasting are two forms of prediction.
Perception converts sensory input into information, generation creates content, and
optimization searches for a plan within specified constraints.

Source: Prepared by the author.

TABLE 1-3: COMPARISON OF THE MAIN TASK FAMILIES

Category Typical output Workplace example How to test success

Accuracy in important
Route a request to the
Classification Category or label categories and the rate of
correct department
critical errors

Numerical
Number or future Estimate demand or Forecast error and stability
forecasting
value completion time across periods
(regression)

Ranking and
Rank content or Relevance and diversity of
recommendatio Ranked list
candidates for review the highest-ranked results
n

30