NIST Revamps AI Safety Institute Cooperative Agreement

The National Institute of Standards and Technology (NIST) has updated its cooperative research and development agreement related to the US Artificial Intelligence Safety Institute (AISI). This revised agreement shifts the focus away from concepts like “AI safety,” “responsible AI,” and “AI fairness,” instead emphasizing the need to mitigate ideological bias while supporting human flourishing and economic competitiveness.

Changes in Focus and Priorities

This update, communicated in early March, outlines new expectations for AISI members. Previously, the NIST encouraged researchers to engage in efforts aimed at identifying and rectifying biases in AI models, particularly those linked to gender, race, age, or economic status. Such efforts are crucial given that biased algorithms can have significant adverse impacts on users, especially among marginalized and economically disadvantaged communities.

Removal of Key Objectives

Notably, the latest agreement has eliminated mentions of developing mechanisms for “authenticating content and tracking its provenance,” along with “labeling synthetic content.” This indicates a diminished focus on addressing misinformation and deep fakes. Additionally, a new directive urges one working group to create testing tools aimed at enhancing the United States’ position in the global AI landscape.

Researcher Concerns

One anonymous researcher from an organization allied with the AISI expressed concern regarding these changes, stating, “The Trump administration has removed safety, fairness, misinformation, and responsibility as things it values for AI, which I think speaks for itself.” The researcher highlighted the potential repercussions for everyday users, suggesting that without proper oversight, algorithms could perpetuate discrimination based on income or other demographic factors.

This sentiment is echoed by another researcher who previously collaborated with the AISI, questioning the implications of prioritizing “human flourishing”: “What does it even mean for humans to flourish?”

The Role of Elon Musk

Elon Musk, currently leading initiatives to reduce government spending on behalf of the Trump administration, has voiced criticism against AI technology developed by companies like OpenAI and Google. Musk recently shared a meme labeling AI models from these companies as “racist” and “woke.” He also referenced an instance where a Google AI debated the ethics of misgendering to avert an unlikely nuclear scenario. Musk, who oversees xAI, a competitor of OpenAI and Google, is linked to research efforts aimed at modifying the political leanings of AI models.

Research on Political Bias in AI

An increasing body of research indicates that political biases present in AI systems can influence both liberal and conservative perspectives. For instance, a 2021 study analyzing Twitter’s recommendation algorithms found that users were more likely to encounter right-leaning viewpoints. These findings raise concerns about the potential implications of biases in AI for public discourse.

Impacts of the Department of Government Efficiency

Since January, Musk’s establishment of the Department of Government Efficiency (DOGE) has led to significant changes within U.S. government operations, including the termination of numerous civil servants and a pause in government spending. This shift has reportedly fostered an environment resistant to dissent against the Trump administration’s objectives. Various departments, such as the Department of Education, have also been noted for archiving or eliminating documents related to diversity, equity, and inclusion (DEI). The NIST has also seen considerable staffing changes as part of these initiatives.