Page 27 - CyberSecurityEssentialsEnglish

may produce highly useful results, but they do not possess human experience, with its
meaning, values, and responsibility.

It is useful to distinguish the model, the application, and the way the system is used. A model
may be able to summarize text, while an application adds an interface for search, file access,

or action execution. The user then defines the objective, data, and review criteria. Two
applications may use the same model, yet one may be more accurate because it consults
reliable sources, or more risky because it can execute commands. We should not judge the

entire field of AI from one application experience.
Essential Terms

TABLE 1-2: TERMS THAT SHOULD NOT BE CONFUSED

Term Plain-language explanation Why it matters

A sequence of steps or rules for solving a An algorithm is not necessarily
Algorithm
problem or performing a task. a learned model.

The output depends on the
A computational or logical representation
Model model type and how it was
that converts inputs into an output.
trained or designed.

An error may lie in the data,
The model together with the interface,
System interface, or permissions, not
data, tools, rules, and users.
only in the model.

Examples or information used for training, Data quality and limitations
Data
operation, or evaluation. affect performance.

It may add search, memory, or
The layer through which a person or
Interface constraints that change the
another system interacts with the model.
output.

Using a model or rules to produce an This is what usually happens
Inference
output for a new case. each time the system is used.

Three Main Approaches to Building AI Systems
Not all AI systems learn from data. Symbolic AI relies on clearly written rules and on
knowledge represented in an explicit, readable form. Expert systems are one example: they

collect knowledge in a specific domain and apply rules to reach a result.

Machine learning trains a model on examples and data so that it can recognize patterns for
classification, ranking, or prediction. The programmer does not write a rule for every case.
Instead, the task is defined, the data is prepared, and a measure of success is established.

26