Page 99 - ComputerGenius_V2.1_Class8
P. 99
Bias in Real-World Data
Artificial Intelligence collects data and learn from
the real-world. Based on those observations, it
draws conclusion and plan accordingly for similar
situations in the future. The problem here, is
that the real-world, with all its irregularities and
confusion, is not a very ideal environment to take
inspiration from. The data collected by the AI
systems from the real-world is full of bias.
Real-world is full of bias, therefore, the data collected by the AI systems from the real-world is also
full of bias. For example, a computer system trained on the data for the last 200 years might observe
that more females were involved in specific jobs or that more percentage of successful businesses
were established by men. It might conclude that specific genders are better equipped for handling
certain jobs (gender bias).
Understanding or detecting such biases is not an
easy task as the behaviours of AI systems is not
simple or even rational. The reason behind their
decision-making is not easy, and in some cases,
just impossible to understand. Many times, the
programmers of AI systems themselves cannot explain
the logic behind decisions taken by the AI systems.
The Problem of Inclusion
The consequence of training AI-systems on biased
data is that they create the problem of inclusion, i.e.
the problem that some people will be left out of the AI
decision-making system. For example, the AI system
used by Amazon service for recruitment got affected
by gender-bias in its data. This created a situation in
which many eligible females were left out of consideration.
The Problem of Facts and Their Interpretation
All AI systems are based on facts or data and their
interpretations. This is the root of all bias problems
related to AI.
AI systems are trained to scan data and find learning
from it, but they are not equipped to understand the
reason behind a particular conclusion or learning.
Access to AI and Ethical Issues 97

