Hey there! Ever wondered how the federal government ensures that the artificial intelligence (AI) systems they’re developing are accountable and ethical? Let’s dive into what I learned from the AI World Government event, where two heavyweights in federal AI development shared their approaches to AI accountability. Trust me, it’s like getting a backstage pass into how the government’s AI brains are wired!
Meet the Experts: Taka Ariga and Bryce Goodman
First up is Taka Ariga, the chief data scientist at the U.S. Government Accountability Office (GAO). Ariga isn’t just crunching numbers; he’s leading the charge with an AI accountability framework aimed at making sure AI systems are developed and deployed responsibly. Then there’s Bryce Goodman, the chief strategist for AI and machine learning at the Defense Innovation Unit (DIU). Goodman’s work involves practical applications of AI that range from disaster response to counter-disinformation efforts. Their shared goal? Ensure that AI is used responsibly and ethically across the board.
An Auditor’s Approach to AI Accountability
Taka Ariga’s work is fascinating. He leverages his auditing expertise to scrutinize AI systems through a structured framework. The initiative started in September 2020, involving a diverse group of experts, including 60% women and 40% underrepresented minorities. Talk about inclusivity!
Ariga’s framework doesn’t float in the clouds. It adopts a lifecycle approach covering design, development, deployment, and continuous monitoring. It rests on four pillars: Governance, Data, Monitoring, and Performance.
– Governance: Under this pillar, Ariga’s team examines the oversight mechanisms for AI projects. Is there a chief AI officer in place? Can they make changes? And critically, how was each AI model deliberated upon?
– Data: Here, the focus is on the quality of training data. Is it representative? Is it functioning as intended?
– Monitoring: Continuous monitoring is crucial. “AI is not a technology you deploy and forget,” Ariga emphasizes. His team ensures that AI systems continuously meet their intended purpose or decide if it’s time to retire them.
– Performance: This examines the societal impact of AI systems, ensuring they don’t violate laws like the Civil Rights Act.
Tackling AI Ethics in the DoD
Over at the DIU, Bryce Goodman is also wrestling with the ethical dimensions of AI. The Department of Defense has laid out five ethical principles for AI: Responsible, Equitable, Traceable, Reliable, and Governable. These are great in theory but translating them into actionable guidelines is where the real task lies.
Before the DIU green-lights a project, it must pass through these ethical filters. If a project can’t meet the standards, it’s a no-go. Goodman emphasizes, “There needs to be an option to say the technology is not there or the problem is not compatible with AI.”
Goodman’s guidelines involve several key questions:
1. Define the Task: Is AI really the best tool for this job?
2. Benchmark: Set clear, upfront benchmarks to measure success.
3. Data Ownership: Who owns the data? Ambiguity here can lead to major issues.
4. Data Evaluation: How was the data collected, and was proper consent obtained?
5. Stakeholder Identification: Identify those who will be directly impacted by the AI system.
6. Mission Holder: There needs to be one person accountable for the project’s outcomes.
7. Rollback Process: Have a contingency plan in case things go south.
Once all these questions are answered satisfactorily, the project moves into development. It’s a meticulous process but essential for developing responsible AI.
Lessons Learned: Metrics and Transparency
Goodman shares some insightful lessons from his experience:
– **Metrics are Key:** Don’t just measure accuracy; define what success looks like.
– **Fit Technology to Task:** High-risk applications need low-risk technology.
– **Transparency with Vendors:** Proprietary algorithms that aren’t open to scrutiny? Big red flag.
He sums it up perfectly, “AI is not magic. It will not solve everything. It should only be used when necessary and only when we can prove it will provide an advantage.”
Wrapping It Up: A Unified Approach
Both Ariga and Goodman highlight the importance of a unified, transparent approach to AI accountability. They are part of broader discussions to create a whole-of-government framework, aiming to push high-level ideals down to actionable steps for AI practitioners.
Got thoughts on how AI should be managed within the government? Drop your comments below! And hey, stay tuned for more exciting insights into the world of AI. Let’s geek out together!