AI Fundamental Research - Managing AI Bias

A crucial principle, for both humans and machines, is to avoid harmful bias that may lead to discrimination and other harmful impacts. To achieve this goal, it is critical to design and develop AI systems by taking multiple sources of bias into account, and with impact in mind. The purpose of NIST’s work in AI bias is to enhance methods for bringing context into the evaluation of AI systems - across use cases and sectors - and improving our understanding of negative impacts and harms.

Identifying and Managing AI Bias

Towards a Standard for Identifying and Managing Bias in Artificial Intelligence (NIST Special Publication 1270), released in March 2022, reflects public comments NIST received on its draft version. Part of a larger NIST effort to manage AI risks through trustworthy and responsible AI, the document offers guidance connected to the AI Risk Management Framework that NIST is developing.  

• Final document link and News story
  • ​​​​​​​Japanese translation of NIST SP1270 (summary translation, statement). Not an official U.S. Government translation.
• First draft document link, Public comments, News story

Managing AI/ML Bias in Context 

Automated decision-making is appealing because artificial intelligence (AI)/machine learning (ML) systems ​can produce more consistent, traceable, and repeatable decisions compared to humans; however, these systems ​may have negative consequences, for example discriminatory outcomes. Harmful bias can manifest in AI/ML systems used to support automated decision making and lead to unfair results, negatively impacting individuals, potentially rippling throughout society, and leading to distrust of AI-based technology and institutions that rely on it.

A project carried out by NIST’s National Cybersecurity Center of Excellence will develop recommended guidance and practices to promote fair and positive outcomes and benefit users of AI services in the credit underwriting domain. A final description of the project draft description of the project has been issued. 

A Workshop on Mitigating Bias in AI explored related issues in depth and informs the project.

Past Activities

AI RMF Workshop 2 Day 3, Panels 9-13
Held March 29-31, 2022, the first two days of the workshop addressed all aspects of the AI Risk Management Framework; day 3 allowed for a deeper dive of issues related to mitigating harmful bias in AI.

NIST Proposes Approach for Reducing Risk of Bias in Artificial Intelligence (News Release 6/22/2021)

Bias in AI Workshop
NIST hosted a virtual workshop on August 18, 2020, to develop a shared understanding of bias in AI, what it is, and how to measure it.

Bias in AI

Contact Information: ai-bias [at] list.nist.gov (ai-bias[at]list[dot]nist[dot]gov)