Problem Definition: What will be the purpose of the AI? Who are our stakeholders? Who benefits from this? Could this cause harm - intentional or not? Data Input: Is the data generation and collection process well understood? Do we know if there are any prior interventions or modifications done? Is the data representative of the real world? Model Building: Are our algorithms and models free of bias? Have we done the necessary due-diligence? What are the limitations of the models? Output Evaluation: What is the format and type of outputs we will produce? Is the model trustworthy? Are we able to interpret the decisions the AI makes? Implementation : What is the nature of decisions we are impacting and for whom? How will the outputs be used? Are there any AI alteration risks? Who should have access to what information?
The transformative solution Cross-examining the ethical objections To ensure the solution complied with the RAI framework, rigorous questions preceded each phase of the AI project development lifecycle with an interdisciplinary team of experts, including legal, domain, product, AI, and engineering specialists. During this process, it became evident that the solution needed to pivot on three primary principles: privacy, transparency, and accountability. Every phase of the development process incorporated these foundational principles.
6
© 2022 Fractal Analytics Inc. All rights reserved
Made with FlippingBook - PDF hosting