Online Program

A Responsible AI Blueprint and an Executable Roadmap for the Intelligence Community (304209)

Jennifer Jin, Accenture Federal Services 
Sridatta Kompella, Accenture Federal Services 
Sophia Lam, Accenture Federal Services 
Ian McCulloh, Accenture Federal Services 
Oscar Munoz, Accenture Federal Services 
Paul Park, Accenture Federal Services 
Aaron Pujanandez, Accenture Federal Services 
Onur Savas, Accenture Federal Services 
*Theresa Walker, Accenture Federal Services 

Keywords: ethical ai, responsible ai, human centered design, data governance, artificial intelligence, ethics of ai, digital ethics, privacy, technology risks

The U.S. Federal Government and its’ Intelligence Community (IC) face strict oversight requirements to ensure the ethical and fiscally responsible use of technology. Government oversight requirements are far more restrictive and transparent than any faced by industry. We propose a Responsible AI framework to enable the government to deliver AI-powered benefits while ensuring a commitment to moral and fiscally responsible standards.

The proposed framework consists of four pillars: Fairness, Performance, Auditability, and Training. We contend that most bias in AI systems are “transferred bias” from human systems, and our framework baselines the consistency and bias, and reduces bias at the data annotation phase to ensure fairness. Responsible performance tradeoffs must account for and fix these issues. Auditability of an AI system involves measuring consistency and bias for fairness, performance, and fiscal responsibility, as well as documenting tradeoff decisions. Training focuses on developing the workforce to implement AI within this framework. We demonstrate our application of this framework for a notional intelligence product to deliver relevant insights to a decision-maker.