Classroom Companion: Business

 
Chapter 20 · Big Data Economics

---

311
20
Box 20.1 Artificial Intelligence, Machine Learning, Expert Systems,
and Data Mining
Big data analytics often requires com-
plex computational methods and draws 
on methods developed in other fields of 
computer science.
Artificial intelligence (AI) is “defined 
as a system’s ability to correctly interpret 
external data, to learn from such data, and 
to use those learnings to achieve specific 
goals and tasks through flexible adapta-
tion” (Kaplan & Haenline, 
2019
). Big data 
analysis is not a subfield of AI but may 
apply methods and tools developed for AI, 
for example, advanced search algorithms 
to identify hidden information, image 
analysis for face recognition, learning algo-
rithms predicting customer behavior that 
can be used for targeting advertisements 
and individual pricing, trading algorithms 
for the stock market, trend predictions, 
and several other technologies.
Machine learning uses computer 
algorithms that are automatically 
updated and modified as new informa-
tion and experience is gathered about 
the family of problems to be solved. 
Machine learning is an efficient method 
in cases where the algorithms are too 
complex or infeasible for standard algo-
rithm design, for example, for spam fil-
ters for emails and navigators for trucks 
in automated warehouses.
Expert systems consist of a knowl-
edge base and a set of inference 
algorithms. The knowledge base is con-
tinuously updated by external input 
and by internal machine learning algo-
rithms where results of earlier predic-
tions are fed back to the knowledge 
base. The inference algorithms consist 
of if-then rules to estimate the conse-
quences of a decision (scenario analy-
sis and decisions under uncertainty). 
Expert system technology may be used 
to manage business operations and cus-
tomer relations.
Data mining refers to large-scale 
data analysis methods to discover pat-
terns and dependencies in complex data 
sets, in particular, unknown patterns 
and dependencies. Among the meth-
ods used are AI, machine learning, 
statistics, mathematical inference, and 
database management. One particu-
larly challenging problem in statistical 
analysis of complex data sets is causal-
ity. The purpose of causality analysis is 
threefold (Pearl, 
2009
):
5
To determine if two variables 
actually depend on each other 
and, if so, to determine which 
of them is the cause and which 
is the effect
5
To determine if two independent 
variables are correlated because 
there is a common cause that the 
two variables are correlated
5
To determine if the variables are 
independent and the correlation 
between them is accidental
To make things even worse, correlation 
implies that the two variables are lin-
early related. If this is not the case, the 
variables may by strongly related, but 
the correlation between them is zero; 
for example, if one of the variables 
increases as the square of the other 
variable, then the variable are strongly 
related but the correlation between 
them is zero!