The Department of Defense (DoD) Chief Digital and Artificial Intelligence Office (CDAO) is conducting AI Bias Bounty exercises to detect and address bias in AI systems. The exercises, sponsored by the CDAO Responsible AI (RAI) Division, involve crowdsourcing efforts and partnerships with ConductorAI-Bugcrowd and BiasBounty.AI. The goal is to identify and mitigate risks in Large Language Models (LLMs), starting with open source chatbots. Public involvement is encouraged, and participants can earn bounties based on evaluation by ConductorAI-Bugcrowd. The outcome of the exercises may influence future DoD AI policies.
| Signal | Change | 10y horizon | Driving force |
|---|---|---|---|
| Department of Defense launches AI Bias Bounty exercises | Crowdsourced efforts to detect bias in AI systems | Improved detection and mitigation of bias in AI systems | Ensuring safe, secure, reliable, and bias-free AI systems in the Department of Defense |
| Public involvement encouraged in identifying bias in AI systems | Increased public participation in auditing AI models for bias | Greater awareness and transparency in AI system bias | Addressing and mitigating risks associated with bias in AI systems |
| First exercise focuses on Large Language Models (LLMs) | Focus on identifying unknown risks in LLMs, starting with open source chatbots | Enhanced understanding and control of bias in LLMs | Managing bias risks specifically in Large Language Models |
| Outcome of bounty exercises may shape future DoD AI policies | Potential impact on future AI policies and adoption in the Department of Defense | Improved guidelines and best practices for AI system deployment | Driving the development of responsible and unbiased AI policies |
| CDAO’s role in integrating and optimizing AI capabilities across DoD | Integration and optimization of AI capabilities within the Department of Defense | Increased AI adoption and utilization in defense operations | Enhancing the DoD’s digital infrastructure and decision-making processes |