Algorithms for Smart World Technologies

Chapter 5Risk Measures

Algorithmic risks stem from utilizing data analytics and cognitive technology-driven software algorithms within both automated and semi-automated decision-making frameworks.

• Input data can be prone to risks, including biases inherent in training datasets, use of incomplete or outdated information, irrelevant data points, lack of sufficient sample size diversity, flawed data collection methods, and inconsistencies between training data and operational data.
• Algorithm design may be susceptible to risks such as biased reasoning, erroneous assumptions, inappropriate modeling techniques, coding inaccuracies, and the identification of misleading patterns within the training data.
• Output decisions face risks such as misinterpretation of results, improper application of the outcomes, and neglect of underlying model assumptions.

5.1 Addressing Algorithmic Bias

The proliferation of advanced data analytics and cognitive technologies has led to the widespread integration of algorithms across diverse industries and business functions. These algorithmic systems now influence critical decisions that affect individuals’ lives—ranging from the content they consume and the jobs they are offered to loan approvals, medical diagnoses, and even outcomes in the judicial system. Simultaneously, the increasing adoption of Internet of Things (IoT) devices has led to a surge in machine-to-machine interactions, many of which are algorithmically driven.

As algorithms ...