AI & Machine Learning

 

(Sriram, Danely, Alvin, Xeon)

Artificial Intelligence and Machine Learning are more and more prevalent in every aspect of our lives, from policing to banking. As such, it is important that these algorithms are designed in ethical ways. Unfortunately, in the banking sector, that is not always the case. These algorithms have advantages including increasing speed, reducing mistakes, reducing labor costs, and enhancing customer experience. However, algorithms are not without bias. Bias can come from unrepresentative or biased data samples used to train the model, bias in the training of the model, or bias in the original design or adaptation after contact with new data. This is especially bad when credit scoring company uses shopping patterns or social media activity, which could be incorrect, in a non-transparent process, and result in denying credit without the consumer knowing why. Furthermore, an algorithm could use factors such as race or writing style, which have no direct connection to creditworthiness and deny them based on a correlation the algorithm found with its data. In short, algorithms can discriminate. AI is still worth using, as it is undeniable the great positive impact they have. Furthermore, if this blog post were going to say that AI should not be used, none of us would be majoring in Computer Science. However, as computer scientists, we have an ethical responsibility to ensure that biases are minimized in our training and our data sets. This is difficult, but possible. Some researchers at Microsoft Research figured out a method to cancel out biases by injecting data with an equal and opposite bias into a biased set. There are definitely methods to minimize bias in machine learning. It may be difficult, but millions of livelihoods depend on it, and so must be done. https://www.whitecase.com/publications/insight/algorithms-and-bias-what-lenders-need-know https://www.americanbanker.com/news/is-it-ok-for-lending-algorithms-to-favor-ivy-league-schools