Page 96 - ConceptGP_C8_Fb
P. 96
• There are three research institutes currently at Oxford University that mainly focus on
AI Ethics.
Reboot
1. Is AI threat to human dignity? Justify.
2. How is an AI system accountable for the actions it takes?
AI Bias
AI bias is an irregularity in the output of machine algorithms. This may happen during the algorithm
development process in which prejudiced assumptions were made or the training data was biased.
Types of AI Bias
There are mainly three types of AI bias:
1. Perceptive biases: There may be operative feelings towards a particular group based upon
the group, one belongs to. Approximately 180 human biases are defined by psychologists.
These can affect the decisions we make. These may also disturb the machine learning
algorithms by the designers, unknowingly introducing them into the model or the data set
on which the machines are trained.
2. Incomplete data biases: When the data is not complete, it lacks accuracy. For example,
when research is done initially using a particular group, it may not represent the whole
population. Machine learning from such data may generate biased output.
3. People: The developers of AI can also be a reason for the bias. The designers focussing
on achieving a specific goal with the available data, may not think of the other broader
aspects which may land onto these biased results.
Popular AI Bias Examples
Let us learn about some popular AI Bias examples:
Amazon Recruitment
With the aim to introduce AI in its recruitment process, Amazon introduced an AI project in 2014.
94 Premium Edition-VIII

