Priyanka Bhattacharya (Illinois Institute of Technology, Chicago-Kent College of Law) has posted A Perspective on Fairness in Artificial Intelligence on SSRN. Here is the abstract:
Data is the weapon of the future. Whoever controls data, controls the world…If we don’t put up a fight, our data will belong to the wrong people--- The Billion Dollar Code Machine Learning Algorithms: An Overview
“Predictive Analytics . . . is the application of mathematics to analyze patterns in historic data. It’s a type of forecasting that seeks to find relationships (correlations) between past and future events. You will hear terms like ‘Data mining’, ‘Machine learning’ ‘Artificial intelligence’ and ‘Deep learning’ being thrown about when people talk about predictive analytics, but it’s all just different types of math at the end of the day.” Artificial intelligence was born in the 1950s, mathematicians, scientists and philosophers began pondering upon the concept ‘artificial intelligence.’ One such person was Alan Turing who raised a question that similar to how humans make use of the available information to reason and solve problems, can machines also solve problems by generating patterns from the available data? —a question whose ramifications we’re still exploring today.
Although there is no concise definition, yet Artificial Intelligence or as we will call it for the purpose of this dissertation ‘machine learning’ is the procedure for creating algorithm that is able to learn from data. I am considering one of the formal definitions of predictive system as “the use of mathematical procedures (algorithms) to analyze data. The aim is to discover useful patterns . . . between different items of data. Once the relationships have been identified, these can be used to make inferences about the behavior of new cases.”
A machine-learning system is trained to recognize patterns in the input data and are explicitly programmed to create a function from the input to the output. r. It’s presented with many examples relevant to a task, and it finds statistical structure in these examples that eventually allows the system to come up with rules for automating the task. For instance, if you wished to automate the task of tagging your vacation pictures, you could present a machine-learning system with many examples of pictures already tagged by humans, and the system would learn statistical rules for associating specific pictures to specific tags. A machine-learning model transforms its input data into meaningful outputs, a process that is “learned” from exposure to known examples of inputs and outputs.
The field of machine learning systems is highly empirical and except some algorithms such as decision trees (explained infra), often the most experienced researchers sometimes lack an explanation for why a particular algorithm behaves in a certain manner. Machine learning systems are no longer a tool for ad placements and spam filters. The past few years have witnessed its wide range of activities in the field of healthcare, parole decisions, filtering loan applicants and many more. Gradually, such usage gave rise to a major concern that such data driven method might lead to discriminatory results or otherwise be unfair or biased. Tom Mitchell in 1980 in his paper titled “The need for biases in learning generalizations” introduced the term ‘bias’ in machine learning systems for the first time.