The first stage of the AI Bias Bounty program, a crowdsourcing effort to find bias in AI systems, has been launched by the Chief Digital and Artificial Intelligence Office (CDAO) of the U.S. Department of Defense (DoD).
The U.S. Department of Defense’s Chief Digital and Artificial Intelligence Office (CDAO) has taken a major step towards the initial phase of the artificial intelligence (AI) Bias Bounty program. The program is a crowdsourced initiative designed to identify bias in AI systems, starting with large language models (LLM).
The bounty program includes two exercises, with the first now open to the public and running from Jan. 29 to Feb. 27, 2024. The first exercise is open to the public, and everyone is advised to register. One does not need extensive coding knowledge to register, and monetary bounties will be awarded based on scoring and evaluation by ConductorAI-Bugcrowd. More details, however, on the second exercise have not been revealed yet.
The US Department of Defense (DoD) CDAO has launched the AI Bias Bounty program’s first phase. Focused on Large Language Models, the initiative aims to uncover potential risks in open source chatbots.
— CDO Magazine (@magazine_cdo) January 31, 2024
The Bounty Exercise
According to the DoD, the first bounty exercise’s goal is to identify unknown areas of risk in large language models (LLMs), beginning with open-source chatbots. This work can thus support the thoughtful mitigation and control of such risks.
Craig Martell, Chief Digital and Artificial Intelligence Officer, said that the outcome of these bounties could significantly impact future DoD AI policies and adoption. He also stated that the CDAO is actively monitoring this area, given the Department’s current focus on risks associated with LLMs.
The DoD may additionally consider the results of these bounties for further research, analysis, best practices, and policy recommendations. This highlights the department’s commitment to ensuring unbiased AI systems in all its operations. The two bias bounties are led by the CDAO Responsible AI (RAI) Division. They developed and executed partnerships with COnductorAI-Bugcrowd and BiasBounty.AI and are advised by the CDAO Digital Services Directorate.
AI Strategies
The DoD unveiled a strategy in November 2023 to expedite the integration of advanced AI capabilities. The focus was to secure enduring decision superiority for U.S. soldiers on the battlefield in the years ahead.
NEWS: DOD Releases AI Adoption Strategy https://t.co/2nLCO7lg1T
— Department of Defense 🇺🇸 (@DeptofDefense) November 3, 2023
However, the new strategy, crafted by the Chief Digital and AI Office, surpasses the 2018 AI Strategy and the revised Data Strategy from 2020. These foundational documents have, as a result, set the stage for the department’s strategy for deploying AI-enabled capabilities.
The exercises are additionally conducted to develop new approaches to algorithmic auditing and red-teaming artificial intelligence models. This will facilitate experimentation with addressing identified risks and ensuring the systems are not biased, given their particular deployment context.
The CDAO
The CDAO was established in June 2022 to integrate and optimize AI capabilities across the DoD. This office is responsible for accelerating the adoption of data, analytics, and AI within the department and also safeguarding against current and emerging threats.
'Digital On-Demand': A Game-Changer for Coast Guard AI Learning – MyCG https://t.co/xwWed8akXw
— Internet of Things A (@Internet__agent) November 22, 2023
Significantly, the DoD, with this first artificial intelligence bias bounty exercise, is setting an example for other organizations to prioritize unbiased AI systems. This step ensures that artificial intelligence is ethically used for the benefit of society, not just that it promotes fairness. The world at large is looking to anticipate the results of this exercise and how it will shape future DoD policies regarding artificial intelligence.
and then