Could AI Be The Solution To Discrimination In The Society?
Discrimination has been a long cause of concern in our society. This issue has been worsened even further through the adoption of AI systems to make decisions in many industry sectors. This is because automated AI systems make decisions based on the analysis of past data. Therefore, if the data being analysed by the systems is biased, it could lead to biased results. For example, if an AI system is being used to identify the most deserving member of staff for a promotion and historically, most staff being promoted are men, the system would take that information into account and is more likely to recommend a man for the promotion over a woman.
The fault does not lie within the machine learning algorithm, the system is simply carrying out the function is was designed to carry out, but since the data it is analysing is based on past actions, the AI System has the potential to make unfair decisions.
The researchers have created an AI tool to detect discrimination with respect to a particular attribute such as race or gender. It does so by using the concept of causality in which one variable (the cause) affects another variable (the effect). The software then uses its counterfactual inference algorithms to arrive at the best, unbiased conclusion. The researchers tested their method using income data from the U.S. Census Bureau to determine whether there is gender-based discrimination in salary determination. They found evidence suggesting that the odds of a woman being paid a salary of over $50,000 is one-third that for a man.
The development of such tools can help mitigate unfairness and inequality in society and ensure that everyone is presented with fair and equal opportunities. This could lead to the eradication of a long-lasting problem of discrimination in the society that has known to have existed throughout history. This could be the massive leap that Artificial Intelligence systems take to improve welfare and bring forward a sense of community among individuals