OTI Urges NIST to Prioritize Transparency, Create Additional Guidelines for AI Bias Assessments

This month, New America’s Open Technology Institute submitted comments in response to the National Institute of Standards and Technology (NIST) consultation on managing bias in artificial intelligence (AI). This is one stage of a broader effort by NIST to develop a risk management framework for AI, with an ultimate goal of encouraging the design of trustworthy and responsible AI. OTI’s comments focus on including transparency as a guiding principle and encouraging developers to conduct meaningful assessments both prior to and after deployment to mitigate risk.

The NIST proposal defines three stages where bias should be addressed: pre-design, design/development, and deployment. OTI’s comments urge NIST to create guidelines or encourage legislation that requires the creators of high-risk algorithms to test their systems prior to deployment to better safeguard individuals’ fundamental rights. Additionally, OTI noted that the NIST proposal does not address the need for developers to address bias during the post-deployment stage. Developers should continue conducting relevant assessments after deployment, especially because the pre-deployment test environment may not encompass the full unpredictability of the real world environment.

In its proposal, NIST identified certain values that will guide its forthcoming work, including accuracy, explainability and interpretability, privacy, and reliability. OTI emphasized that transparency must be added to this list. Ensuring that the user knows when they are using a technology with AI capabilities, and how the system broadly operates is an important step to increasing public trust and decreasing fear. OTI has put forth numerous recommendations on how internet platforms and government agencies can promote fairness, accountability, and transparency around algorithmic systems.

The following quote can be attributed to Spandana Singh, policy analyst at New America’s Open Technology Institute:

OTI welcomes NIST’s efforts to promote the design of trustworthy and responsible AI. NIST’s proposal for reducing the risk of bias in artificial intelligence is a good first step towards achieving this goal. As NIST continues to drive this process forward, we encourage the organization to prioritize transparency and consider creating guidelines for pre-deployment and post-deployment assessments of high-risk algorithms.