Your donation will support the student journalists of School for Advanced Studies-Wolfson. Your contribution will allow us to purchase equipment and cover our annual website hosting costs.
The Bias in A.I.
December 18, 2022
Artificial Intelligence (A.I.) has been featured across headlines in recent years as one of the more-promising new technologies of the 21st century. The use of A.I. could accelerate advances in fields such as medicine, finance, cybersecurity, etc. However, it is important to mention that the uses of A.I. could sometimes result in detrimental consequences.
To explain how A.I. can do more harm than good, we must first define what it is: “[artificial intelligence] works by learning the patterns in a dataset. It combines and accumulates a large set of data for a specific task through an intelligent and iterative collection process. It then finds patterns in the data to predict the outcome for specific input”. Through this process, A.I. can learn and recognize similar characteristics throughout the data which enables it to have an idea of what it is analyzing. Nevertheless, when crucial information is missing and the necessary amount of data is not presented, errors can occur. In fields such as medicine, when A.I. is presented with data for lighter and darker skin tones, the system will have a much easier time identifying diseases on those with lighter hues; this could potentially result in unidentified or misinterpreted diagnoses for darker skin tones.
The reason why these errors occur and the data may be in favor of a particular set of people is due to the lack of impartiality or awareness regarding societal differences that directly affect our way of life. For many, the thought of diversity holds little to no importance, more specifically in fields such as technology and engineering. This lack of diversity has left many with the fear that the “the technology is at risk of perpetuating historical biases and power imbalances.” The data set A.I. uses is determined by the programmers, who are responsible for what is represented in the data and what will be left out.
Since no one is perfect, the programmers may at times fail to consider things that may exclusively benefit one group of people at the expense of others. Nevertheless, it is important to realize the impact diversity plays in artificial intelligence; when everyone is accounted for, the product is improved.