Implications of Policies and Governance for Ethical AI Algorithms on the Private Sector’s Algorithm Development Costs

There is currently an acceleration in the adoption of Machine Learning (ML), which causes concerns about the effects of algorithmic bias. Bias stems from two sources: the data upon which the algorithm operates and the human biases embedded in the algorithm’s code. It is paramount to understand its roots, its categories, and the stakeholders each of these affect. 

Booz Allen Hamilton (BAH) asked the Capstone team to explore policy options that would incentivize the adoption of ethical and non-biased performance requirements in the development of Artificial Intelligence (AI) products. This project focused specifically on ML as a subset of AI; it’s the ML that uses algorithms to identify patterns within data, and those patterns are used to create a data model that makes predictions. The project examined distinct sources of biases that affect algorithms in every step of their development.  The team focused specifically on how the algorithm is trained and the source of biased outcomes. Following, the team outlined human and technical solutions to address these previously defined sources of bias. Through their research, the Capstone team created a framework to apply to real-world scenarios and validated this framework to application to healthcare, hiring, and policing use cases. On the technical side, the Capstone team recommended framework focused on the use of proper data in the pre-, in-, and post- processing phases. On the human side, they recommended incorporating DEI efforts, auditing, training, and model card use to monitor algorithms. BAH will be able to apply these principles to future endeavors of their organization.