Abstract:
Technology is becoming more complex and is increasingly being used in law. Tools to assist in decision making and becoming more complex. It is important to ensure accountability structures keep up with their development so we do not lose control of decision making processes. This paper identifies four types of decision making using algorithms: human decisions, decisions using non-machine learning algorithms, decisions using machine learning algorithms, and decisions where machine learning makes the decision. Issues are identified in applying accountability mechanisms for each, focusing on challenges in pinpointing an actor to hold accountable and forums equipped to ask questions. The use of machine learning is a significant hurdle in being able to choose an actor because these kinds of algorithms are opaque and require significant expertise to comprehend. Users do not necessarily know how the machine works and so cannot provide adequate account for their use. Programmers may have to shoulder some of the accountability burden, however they too may be unable to provide complete answers. Likewise, forums may lack knowledge to ask meaningful questions and lack of transparency on the part of the algorithm. Problems identifying these parties in an accountability context need to be resolved for the future as machine learning algorithms become more common.
